[DISCUSS] phoenix-connectors and phoenix-queryserver release process

[Thomas D'Silva <td...@salesforce.com>]

We are currently working on moving the connectors and queryserver to their
own repos (see https://issues.apache.org/jira/browse/PHOENIX-5062 and
https://issues.apache.org/jira/browse/PHOENIX-5063).
IMO we should have a single branch for each repo that depends on a released
version of Phoenix/HBase. I thinnk we should start the pom versions at
1.0.0. We can then have connectors or queryserver releases independent of
Phoenix. For the connectors, I think we should release all of them when we
have a connector release. Same with queryserver (release both the server
and client).
What do folks think?

Re: [DISCUSS] phoenix-connectors and phoenix-queryserver release process

[Vincent Poon <vi...@apache.org>]

Right now the "presto connector" I'm working on also builds an uber jar
that is like phoenix-client but with some additional excludes and
relocations.  So I would need to build one for each Phoenix release.  So I
was thinking of something like:
phoenix-client-4.14.1-HBase-1.4-presto-1.0.0.jar

I would think it would be the same for e.g. PQS, if we ultimately want to
build a tarball that includes everything you need to run PQS, which is the
easiest thing for users.  You need a specific Phoenix client version to
include in the bundle.

Yet at the same time, we want to avoid having many branches like Phoenix.

So maybe we can just have a single master branch, and have a 'target'
Phoenix version as a variable, with a default of the latest Phoenix release.
Then at release time we can deploy with different Phoenix version targets
as needed.

On Tue, Jan 8, 2019 at 12:50 PM Josh Elser <el...@apache.org> wrote:

> +1 for queryserver starting out at 1.0.0, for sure. We've really made
> only a few changes for PQS over the years (the majority of code is in
> Avatica). Doesn't make sense to follow the 'normal' Phoenix versioning.
>
> Assuming the connectors is similarly slow moving, I'm in favor of the
> same for that too.
>
> I asked on the PR for the queryserver repo: how are people going to
> interact with these? Something as simple as:
>
> * How will PQS know where the phoenix-client.jar is?
> * Where do users invoke sqlline-thin.py from?
>
> On 1/8/19 2:34 PM, Thomas D'Silva wrote:
> > We are currently working on moving the connectors and queryserver to
> their
> > own repos (see https://issues.apache.org/jira/browse/PHOENIX-5062 and
> > https://issues.apache.org/jira/browse/PHOENIX-5063).
> > IMO we should have a single branch for each repo that depends on a
> released
> > version of Phoenix/HBase. I thinnk we should start the pom versions at
> > 1.0.0. We can then have connectors or queryserver releases independent of
> > Phoenix. For the connectors, I think we should release all of them when
> we
> > have a connector release. Same with queryserver (release both the server
> > and client).
> > What do folks think?
> >
>

Re: [DISCUSS] phoenix-connectors and phoenix-queryserver release process

[Josh Elser <el...@apache.org>]

+1 for queryserver starting out at 1.0.0, for sure. We've really made 
only a few changes for PQS over the years (the majority of code is in 
Avatica). Doesn't make sense to follow the 'normal' Phoenix versioning.

Assuming the connectors is similarly slow moving, I'm in favor of the 
same for that too.

I asked on the PR for the queryserver repo: how are people going to 
interact with these? Something as simple as:

* How will PQS know where the phoenix-client.jar is?
* Where do users invoke sqlline-thin.py from?

On 1/8/19 2:34 PM, Thomas D'Silva wrote:
> We are currently working on moving the connectors and queryserver to their
> own repos (see https://issues.apache.org/jira/browse/PHOENIX-5062 and
> https://issues.apache.org/jira/browse/PHOENIX-5063).
> IMO we should have a single branch for each repo that depends on a released
> version of Phoenix/HBase. I thinnk we should start the pom versions at
> 1.0.0. We can then have connectors or queryserver releases independent of
> Phoenix. For the connectors, I think we should release all of them when we
> have a connector release. Same with queryserver (release both the server
> and client).
> What do folks think?
>