You are viewing a plain text version of this content. The canonical link for it is here.
Posted to dev@dubbo.apache.org by Liujie Qin <li...@apache.org> on 2018/07/09 07:32:01 UTC

Proposal for Dubbo 3.0

  Hi All, Dubbo has become a excellent RPC framework with the efforts of
various partners in the community. However, in the process of development,
we found some shortcomings of Dubbo 2.x, which is listed as follows:

   - The maintainability is relatively poor. The code uses a lot of
   dynamically generated classes. It is difficult to understand and difficult
   to debug. The code hierarchy is not clear enough, and the calling logic is
   confusing. As an experienced Dubbo developer, sometimes it is difficult for
   me to sort out the calling process of Dubbo 2.x; the more developed, the
   more difficult it is to maintain, so it needs to be refactored in order to
   develop better.
   - Some extensions cannot meet current needs, such as service addresses
   push monitoring, performing Filter early interception requests before
   service addresses selection, printing the process of route selection, etc.
   - The underlying Dubbo 2.X is based on synchronous calls. The
   asynchronous call is wrapped synchronously. This introduces inaccuracies
   such as RT during asynchronous calls.
   - At present, the service metadata information and the address of the
   Dubbo service are coupled together. Most of the metadata is the same, and
   the metadata is redundant. At the same time, as the size of the machine
   increases, the Registry is under great pressure, especially when machine
   restart.
   - The service registry and the configuration center are coupled
   together, and after splitting, the management and storage of service
   governance rule data such as routing rules can be better handled.
   - The core is written based on JDK 1.6 and cannot use the newer JDK
   features. It is recommended to upgrade JDK to 1.8 and above.

For the above reasons, in order to better develop Dubbo and attract more
people to participate in the development of the Dubbo community, I propose
to refactor the Dubbo 2.x code. The newly developed version number is Dubbo
3.x. Refactoring Dubbo 3.x will have the following advantages:

   - Refactoring core modules to make the code more concise and
   understandable
   - Refactoring the routing process, Making the route process to Pipeline
   style,Making the route process can be observed
   - Add new extension points to make it easier for developers to extend
   the framework
   - Add the Service Domain Model to enable developers to handle its
   lifecycle
   - Supports asynchronous interface, unified call process returns
   CompleteFuture, synchronous call process uses CompleteFuture wrapper; Solve
   the problem of inaccurate statistics such as RT during asynchronous call
   - Splitting out the Configuration Center and Registry for better control
   and support for larger clusters
   - Prepare for new features such as Http2 stream、Reactive stream
   - Performance improvement
   - And so on

Of course, refactoring Dubbo 3.x also faces the following challenges:

   - A wide range of refactorings that can cause extension points to be
   incompatible
   - The Dubbo team may need to maintain two branches at the same time for
   a longer period of time

For the following challenges, we can do this:

   -

   For major extension points, such as Fitler, etc., be sure to ensure
   compatibility and enable developers to smoothly upgrade
   -

   For some extension points, the Dubbo team will listen to the community
   and selectively fix compatibility issues with some extension points.

Comments welcome!

thank you ~

Re: Proposal for Dubbo 3.0

Posted by Andrea Del Bene <an...@gmail.com>.
Hi Minxuan,

I followed the steps you suggested. Once I stop BasicConsumer I actually
receive a notification about it but in the 'Applications' page I still see
both 'demo-consumer' and 'demo-provider'. The only difference I noticed is
that 'demo-consumer' has both 'Providers' and 'Consumers' columns to 0,
while 'Consumers' was 1 before I stop it. This make perfectly sense, but I
would expect that once you have no more running instances of a given
application, this one would be removed from 'Applications' page list.
I also had a look at the code of class RegistryContainer from project
'incubator-dubbo-ops/dubbo-monitor-simple/' which is the class that
receives notifications from registry.  I noticed that applications are
stored in a Set variabile where 'remove' method is never called (here is
the variable 'https://github.com/apache/incubator-dubbo-ops/blob/
master/dubbo-monitor-simple/src/main/java/com/alibaba/dubbo/monitor/simple/
container/RegistryContainer.java#L46') so I guess it was meant to work like
this (i.e. applications are never removed), right?

<https://github.com/apache/incubator-dubbo-ops>


On Thu, Jul 12, 2018 at 10:57 PM, Andrea Del Bene <an...@gmail.com>
wrote:

> Hi,
>
> thank you for your reply. So far I used a stand-alone broadcast registry.
> I will try to follow the steps you suggested me.
>
>
>
> On 12/07/2018 16:46, Minxuan Zhuang wrote:
>
>> Hi
>>      Thank you for being interested in Dubbo monitor, If you unregister a
>> a
>> consumer, the monitor will get the information Immediately,
>> you can try this by the steps below:
>> 1. start a zookeeper
>> 2 start Dubbo monitor by running MonitorStarter.java
>> 3. download dubbo samples here <https://github.com/dubbo/dubbo-samples>
>> and
>> run dubbo-samples-monitor module
>> If you start both BasicProvider and BasicConsumer, you can see provider
>> and
>> consumer info on dubbo monitor, once you stop BasicConsumer, this consumer
>> info will disappear. the monitor can detect the change by NotifyListener
>> and this mechanism is very efficient for me, do you have any
>> suggestion for implementing
>> this?
>>
>> On Wed, Jul 11, 2018 at 5:29 PM Andrea Del Bene <an...@gmail.com>
>> wrote:
>>
>> Hi,
>>>
>>> I'm very new to Dubbo but I'd like to share some ideas. I am playing
>>> around
>>> trying to implement a simple monitor for Dubbo (this one
>>> https://github.com/bitstorm/wicket-coreui). To start grasping what I
>>> have
>>> to do I've looked at what dubbo-monitor-simple (
>>> https://github.com/apache/incubator-dubbo-ops) already does. While the
>>> overall mechanism is quite clear I find difficult working with the
>>> current
>>> URL-based notification mechanism (I'm referring to interface
>>> NotifyListener). Extracting informations from url is not trivial and I
>>> think we might use a more complex and clear object for this task. In
>>> particular is difficult to be notified if a consumer application has been
>>> unregistered.
>>>
>>> Hope this could be helpful for future improvements.
>>>
>>> Thank you.
>>>
>>> On Mon, Jul 9, 2018 at 9:32 AM, Liujie Qin <li...@apache.org> wrote:
>>>
>>>    Hi All, Dubbo has become a excellent RPC framework with the efforts of
>>>> various partners in the community. However, in the process of
>>>>
>>> development,
>>>
>>>> we found some shortcomings of Dubbo 2.x, which is listed as follows:
>>>>
>>>>     - The maintainability is relatively poor. The code uses a lot of
>>>>     dynamically generated classes. It is difficult to understand and
>>>> difficult
>>>>     to debug. The code hierarchy is not clear enough, and the calling
>>>>
>>> logic
>>>
>>>> is
>>>>     confusing. As an experienced Dubbo developer, sometimes it is
>>>>
>>> difficult
>>>
>>>> for
>>>>     me to sort out the calling process of Dubbo 2.x; the more developed,
>>>>
>>> the
>>>
>>>>     more difficult it is to maintain, so it needs to be refactored in
>>>>
>>> order
>>>
>>>> to
>>>>     develop better.
>>>>     - Some extensions cannot meet current needs, such as service
>>>> addresses
>>>>     push monitoring, performing Filter early interception requests
>>>> before
>>>>     service addresses selection, printing the process of route
>>>> selection,
>>>> etc.
>>>>     - The underlying Dubbo 2.X is based on synchronous calls. The
>>>>     asynchronous call is wrapped synchronously. This introduces
>>>>
>>> inaccuracies
>>>
>>>>     such as RT during asynchronous calls.
>>>>     - At present, the service metadata information and the address of
>>>> the
>>>>     Dubbo service are coupled together. Most of the metadata is the
>>>> same,
>>>> and
>>>>     the metadata is redundant. At the same time, as the size of the
>>>>
>>> machine
>>>
>>>>     increases, the Registry is under great pressure, especially when
>>>>
>>> machine
>>>
>>>>     restart.
>>>>     - The service registry and the configuration center are coupled
>>>>     together, and after splitting, the management and storage of service
>>>>     governance rule data such as routing rules can be better handled.
>>>>     - The core is written based on JDK 1.6 and cannot use the newer JDK
>>>>     features. It is recommended to upgrade JDK to 1.8 and above.
>>>>
>>>> For the above reasons, in order to better develop Dubbo and attract more
>>>> people to participate in the development of the Dubbo community, I
>>>>
>>> propose
>>>
>>>> to refactor the Dubbo 2.x code. The newly developed version number is
>>>>
>>> Dubbo
>>>
>>>> 3.x. Refactoring Dubbo 3.x will have the following advantages:
>>>>
>>>>     - Refactoring core modules to make the code more concise and
>>>>     understandable
>>>>     - Refactoring the routing process, Making the route process to
>>>>
>>> Pipeline
>>>
>>>>     style,Making the route process can be observed
>>>>     - Add new extension points to make it easier for developers to
>>>> extend
>>>>     the framework
>>>>     - Add the Service Domain Model to enable developers to handle its
>>>>     lifecycle
>>>>     - Supports asynchronous interface, unified call process returns
>>>>     CompleteFuture, synchronous call process uses CompleteFuture
>>>> wrapper;
>>>> Solve
>>>>     the problem of inaccurate statistics such as RT during asynchronous
>>>>
>>> call
>>>
>>>>     - Splitting out the Configuration Center and Registry for better
>>>>
>>> control
>>>
>>>>     and support for larger clusters
>>>>     - Prepare for new features such as Http2 stream、Reactive stream
>>>>     - Performance improvement
>>>>     - And so on
>>>>
>>>> Of course, refactoring Dubbo 3.x also faces the following challenges:
>>>>
>>>>     - A wide range of refactorings that can cause extension points to be
>>>>     incompatible
>>>>     - The Dubbo team may need to maintain two branches at the same time
>>>>
>>> for
>>>
>>>>     a longer period of time
>>>>
>>>> For the following challenges, we can do this:
>>>>
>>>>     -
>>>>
>>>>     For major extension points, such as Fitler, etc., be sure to ensure
>>>>     compatibility and enable developers to smoothly upgrade
>>>>     -
>>>>
>>>>     For some extension points, the Dubbo team will listen to the
>>>> community
>>>>     and selectively fix compatibility issues with some extension points.
>>>>
>>>> Comments welcome!
>>>>
>>>> thank you ~
>>>>
>>>>
>>>
>>> --
>>> Andrea Del Bene.
>>> Apache Wicket committer.
>>>
>>>
>


-- 
Andrea Del Bene.
Apache Wicket committer.

Re: Proposal for Dubbo 3.0

Posted by Andrea Del Bene <an...@gmail.com>.
Hi,

thank you for your reply. So far I used a stand-alone broadcast 
registry. I will try to follow the steps you suggested me.


On 12/07/2018 16:46, Minxuan Zhuang wrote:
> Hi
>      Thank you for being interested in Dubbo monitor, If you unregister a a
> consumer, the monitor will get the information Immediately,
> you can try this by the steps below:
> 1. start a zookeeper
> 2 start Dubbo monitor by running MonitorStarter.java
> 3. download dubbo samples here <https://github.com/dubbo/dubbo-samples> and
> run dubbo-samples-monitor module
> If you start both BasicProvider and BasicConsumer, you can see provider and
> consumer info on dubbo monitor, once you stop BasicConsumer, this consumer
> info will disappear. the monitor can detect the change by NotifyListener
> and this mechanism is very efficient for me, do you have any
> suggestion for implementing
> this?
>
> On Wed, Jul 11, 2018 at 5:29 PM Andrea Del Bene <an...@gmail.com>
> wrote:
>
>> Hi,
>>
>> I'm very new to Dubbo but I'd like to share some ideas. I am playing around
>> trying to implement a simple monitor for Dubbo (this one
>> https://github.com/bitstorm/wicket-coreui). To start grasping what I have
>> to do I've looked at what dubbo-monitor-simple (
>> https://github.com/apache/incubator-dubbo-ops) already does. While the
>> overall mechanism is quite clear I find difficult working with the current
>> URL-based notification mechanism (I'm referring to interface
>> NotifyListener). Extracting informations from url is not trivial and I
>> think we might use a more complex and clear object for this task. In
>> particular is difficult to be notified if a consumer application has been
>> unregistered.
>>
>> Hope this could be helpful for future improvements.
>>
>> Thank you.
>>
>> On Mon, Jul 9, 2018 at 9:32 AM, Liujie Qin <li...@apache.org> wrote:
>>
>>>    Hi All, Dubbo has become a excellent RPC framework with the efforts of
>>> various partners in the community. However, in the process of
>> development,
>>> we found some shortcomings of Dubbo 2.x, which is listed as follows:
>>>
>>>     - The maintainability is relatively poor. The code uses a lot of
>>>     dynamically generated classes. It is difficult to understand and
>>> difficult
>>>     to debug. The code hierarchy is not clear enough, and the calling
>> logic
>>> is
>>>     confusing. As an experienced Dubbo developer, sometimes it is
>> difficult
>>> for
>>>     me to sort out the calling process of Dubbo 2.x; the more developed,
>> the
>>>     more difficult it is to maintain, so it needs to be refactored in
>> order
>>> to
>>>     develop better.
>>>     - Some extensions cannot meet current needs, such as service addresses
>>>     push monitoring, performing Filter early interception requests before
>>>     service addresses selection, printing the process of route selection,
>>> etc.
>>>     - The underlying Dubbo 2.X is based on synchronous calls. The
>>>     asynchronous call is wrapped synchronously. This introduces
>> inaccuracies
>>>     such as RT during asynchronous calls.
>>>     - At present, the service metadata information and the address of the
>>>     Dubbo service are coupled together. Most of the metadata is the same,
>>> and
>>>     the metadata is redundant. At the same time, as the size of the
>> machine
>>>     increases, the Registry is under great pressure, especially when
>> machine
>>>     restart.
>>>     - The service registry and the configuration center are coupled
>>>     together, and after splitting, the management and storage of service
>>>     governance rule data such as routing rules can be better handled.
>>>     - The core is written based on JDK 1.6 and cannot use the newer JDK
>>>     features. It is recommended to upgrade JDK to 1.8 and above.
>>>
>>> For the above reasons, in order to better develop Dubbo and attract more
>>> people to participate in the development of the Dubbo community, I
>> propose
>>> to refactor the Dubbo 2.x code. The newly developed version number is
>> Dubbo
>>> 3.x. Refactoring Dubbo 3.x will have the following advantages:
>>>
>>>     - Refactoring core modules to make the code more concise and
>>>     understandable
>>>     - Refactoring the routing process, Making the route process to
>> Pipeline
>>>     style,Making the route process can be observed
>>>     - Add new extension points to make it easier for developers to extend
>>>     the framework
>>>     - Add the Service Domain Model to enable developers to handle its
>>>     lifecycle
>>>     - Supports asynchronous interface, unified call process returns
>>>     CompleteFuture, synchronous call process uses CompleteFuture wrapper;
>>> Solve
>>>     the problem of inaccurate statistics such as RT during asynchronous
>> call
>>>     - Splitting out the Configuration Center and Registry for better
>> control
>>>     and support for larger clusters
>>>     - Prepare for new features such as Http2 stream、Reactive stream
>>>     - Performance improvement
>>>     - And so on
>>>
>>> Of course, refactoring Dubbo 3.x also faces the following challenges:
>>>
>>>     - A wide range of refactorings that can cause extension points to be
>>>     incompatible
>>>     - The Dubbo team may need to maintain two branches at the same time
>> for
>>>     a longer period of time
>>>
>>> For the following challenges, we can do this:
>>>
>>>     -
>>>
>>>     For major extension points, such as Fitler, etc., be sure to ensure
>>>     compatibility and enable developers to smoothly upgrade
>>>     -
>>>
>>>     For some extension points, the Dubbo team will listen to the community
>>>     and selectively fix compatibility issues with some extension points.
>>>
>>> Comments welcome!
>>>
>>> thank you ~
>>>
>>
>>
>> --
>> Andrea Del Bene.
>> Apache Wicket committer.
>>


Re: Proposal for Dubbo 3.0

Posted by Minxuan Zhuang <z8...@gmail.com>.
Hi
    Thank you for being interested in Dubbo monitor, If you unregister a a
consumer, the monitor will get the information Immediately,
you can try this by the steps below:
1. start a zookeeper
2 start Dubbo monitor by running MonitorStarter.java
3. download dubbo samples here <https://github.com/dubbo/dubbo-samples> and
run dubbo-samples-monitor module
If you start both BasicProvider and BasicConsumer, you can see provider and
consumer info on dubbo monitor, once you stop BasicConsumer, this consumer
info will disappear. the monitor can detect the change by NotifyListener
and this mechanism is very efficient for me, do you have any
suggestion for implementing
this?

On Wed, Jul 11, 2018 at 5:29 PM Andrea Del Bene <an...@gmail.com>
wrote:

> Hi,
>
> I'm very new to Dubbo but I'd like to share some ideas. I am playing around
> trying to implement a simple monitor for Dubbo (this one
> https://github.com/bitstorm/wicket-coreui). To start grasping what I have
> to do I've looked at what dubbo-monitor-simple (
> https://github.com/apache/incubator-dubbo-ops) already does. While the
> overall mechanism is quite clear I find difficult working with the current
> URL-based notification mechanism (I'm referring to interface
> NotifyListener). Extracting informations from url is not trivial and I
> think we might use a more complex and clear object for this task. In
> particular is difficult to be notified if a consumer application has been
> unregistered.
>
> Hope this could be helpful for future improvements.
>
> Thank you.
>
> On Mon, Jul 9, 2018 at 9:32 AM, Liujie Qin <li...@apache.org> wrote:
>
> >   Hi All, Dubbo has become a excellent RPC framework with the efforts of
> > various partners in the community. However, in the process of
> development,
> > we found some shortcomings of Dubbo 2.x, which is listed as follows:
> >
> >    - The maintainability is relatively poor. The code uses a lot of
> >    dynamically generated classes. It is difficult to understand and
> > difficult
> >    to debug. The code hierarchy is not clear enough, and the calling
> logic
> > is
> >    confusing. As an experienced Dubbo developer, sometimes it is
> difficult
> > for
> >    me to sort out the calling process of Dubbo 2.x; the more developed,
> the
> >    more difficult it is to maintain, so it needs to be refactored in
> order
> > to
> >    develop better.
> >    - Some extensions cannot meet current needs, such as service addresses
> >    push monitoring, performing Filter early interception requests before
> >    service addresses selection, printing the process of route selection,
> > etc.
> >    - The underlying Dubbo 2.X is based on synchronous calls. The
> >    asynchronous call is wrapped synchronously. This introduces
> inaccuracies
> >    such as RT during asynchronous calls.
> >    - At present, the service metadata information and the address of the
> >    Dubbo service are coupled together. Most of the metadata is the same,
> > and
> >    the metadata is redundant. At the same time, as the size of the
> machine
> >    increases, the Registry is under great pressure, especially when
> machine
> >    restart.
> >    - The service registry and the configuration center are coupled
> >    together, and after splitting, the management and storage of service
> >    governance rule data such as routing rules can be better handled.
> >    - The core is written based on JDK 1.6 and cannot use the newer JDK
> >    features. It is recommended to upgrade JDK to 1.8 and above.
> >
> > For the above reasons, in order to better develop Dubbo and attract more
> > people to participate in the development of the Dubbo community, I
> propose
> > to refactor the Dubbo 2.x code. The newly developed version number is
> Dubbo
> > 3.x. Refactoring Dubbo 3.x will have the following advantages:
> >
> >    - Refactoring core modules to make the code more concise and
> >    understandable
> >    - Refactoring the routing process, Making the route process to
> Pipeline
> >    style,Making the route process can be observed
> >    - Add new extension points to make it easier for developers to extend
> >    the framework
> >    - Add the Service Domain Model to enable developers to handle its
> >    lifecycle
> >    - Supports asynchronous interface, unified call process returns
> >    CompleteFuture, synchronous call process uses CompleteFuture wrapper;
> > Solve
> >    the problem of inaccurate statistics such as RT during asynchronous
> call
> >    - Splitting out the Configuration Center and Registry for better
> control
> >    and support for larger clusters
> >    - Prepare for new features such as Http2 stream、Reactive stream
> >    - Performance improvement
> >    - And so on
> >
> > Of course, refactoring Dubbo 3.x also faces the following challenges:
> >
> >    - A wide range of refactorings that can cause extension points to be
> >    incompatible
> >    - The Dubbo team may need to maintain two branches at the same time
> for
> >    a longer period of time
> >
> > For the following challenges, we can do this:
> >
> >    -
> >
> >    For major extension points, such as Fitler, etc., be sure to ensure
> >    compatibility and enable developers to smoothly upgrade
> >    -
> >
> >    For some extension points, the Dubbo team will listen to the community
> >    and selectively fix compatibility issues with some extension points.
> >
> > Comments welcome!
> >
> > thank you ~
> >
>
>
>
> --
> Andrea Del Bene.
> Apache Wicket committer.
>

Re: Proposal for Dubbo 3.0

Posted by Andrea Del Bene <an...@gmail.com>.
Hi,

I'm very new to Dubbo but I'd like to share some ideas. I am playing around
trying to implement a simple monitor for Dubbo (this one
https://github.com/bitstorm/wicket-coreui). To start grasping what I have
to do I've looked at what dubbo-monitor-simple (
https://github.com/apache/incubator-dubbo-ops) already does. While the
overall mechanism is quite clear I find difficult working with the current
URL-based notification mechanism (I'm referring to interface
NotifyListener). Extracting informations from url is not trivial and I
think we might use a more complex and clear object for this task. In
particular is difficult to be notified if a consumer application has been
unregistered.

Hope this could be helpful for future improvements.

Thank you.

On Mon, Jul 9, 2018 at 9:32 AM, Liujie Qin <li...@apache.org> wrote:

>   Hi All, Dubbo has become a excellent RPC framework with the efforts of
> various partners in the community. However, in the process of development,
> we found some shortcomings of Dubbo 2.x, which is listed as follows:
>
>    - The maintainability is relatively poor. The code uses a lot of
>    dynamically generated classes. It is difficult to understand and
> difficult
>    to debug. The code hierarchy is not clear enough, and the calling logic
> is
>    confusing. As an experienced Dubbo developer, sometimes it is difficult
> for
>    me to sort out the calling process of Dubbo 2.x; the more developed, the
>    more difficult it is to maintain, so it needs to be refactored in order
> to
>    develop better.
>    - Some extensions cannot meet current needs, such as service addresses
>    push monitoring, performing Filter early interception requests before
>    service addresses selection, printing the process of route selection,
> etc.
>    - The underlying Dubbo 2.X is based on synchronous calls. The
>    asynchronous call is wrapped synchronously. This introduces inaccuracies
>    such as RT during asynchronous calls.
>    - At present, the service metadata information and the address of the
>    Dubbo service are coupled together. Most of the metadata is the same,
> and
>    the metadata is redundant. At the same time, as the size of the machine
>    increases, the Registry is under great pressure, especially when machine
>    restart.
>    - The service registry and the configuration center are coupled
>    together, and after splitting, the management and storage of service
>    governance rule data such as routing rules can be better handled.
>    - The core is written based on JDK 1.6 and cannot use the newer JDK
>    features. It is recommended to upgrade JDK to 1.8 and above.
>
> For the above reasons, in order to better develop Dubbo and attract more
> people to participate in the development of the Dubbo community, I propose
> to refactor the Dubbo 2.x code. The newly developed version number is Dubbo
> 3.x. Refactoring Dubbo 3.x will have the following advantages:
>
>    - Refactoring core modules to make the code more concise and
>    understandable
>    - Refactoring the routing process, Making the route process to Pipeline
>    style,Making the route process can be observed
>    - Add new extension points to make it easier for developers to extend
>    the framework
>    - Add the Service Domain Model to enable developers to handle its
>    lifecycle
>    - Supports asynchronous interface, unified call process returns
>    CompleteFuture, synchronous call process uses CompleteFuture wrapper;
> Solve
>    the problem of inaccurate statistics such as RT during asynchronous call
>    - Splitting out the Configuration Center and Registry for better control
>    and support for larger clusters
>    - Prepare for new features such as Http2 stream、Reactive stream
>    - Performance improvement
>    - And so on
>
> Of course, refactoring Dubbo 3.x also faces the following challenges:
>
>    - A wide range of refactorings that can cause extension points to be
>    incompatible
>    - The Dubbo team may need to maintain two branches at the same time for
>    a longer period of time
>
> For the following challenges, we can do this:
>
>    -
>
>    For major extension points, such as Fitler, etc., be sure to ensure
>    compatibility and enable developers to smoothly upgrade
>    -
>
>    For some extension points, the Dubbo team will listen to the community
>    and selectively fix compatibility issues with some extension points.
>
> Comments welcome!
>
> thank you ~
>



-- 
Andrea Del Bene.
Apache Wicket committer.

Re: Proposal for Dubbo 3.0

Posted by Ian Luo <ia...@gmail.com>.
Dubbo is an excellent RPC framework, but it does need keep evolving. I
think this proposal sends one strong message to the community that we are
actively thinking in Dubbo's future in order to make it better.

Community, pls. share your thoughts on this proposal as soon as possible
and as much as possible.

Thanks,
-Ian.

On Mon, Jul 9, 2018 at 3:32 PM Liujie Qin <li...@apache.org> wrote:

>   Hi All, Dubbo has become a excellent RPC framework with the efforts of
> various partners in the community. However, in the process of development,
> we found some shortcomings of Dubbo 2.x, which is listed as follows:
>
>    - The maintainability is relatively poor. The code uses a lot of
>    dynamically generated classes. It is difficult to understand and
> difficult
>    to debug. The code hierarchy is not clear enough, and the calling logic
> is
>    confusing. As an experienced Dubbo developer, sometimes it is difficult
> for
>    me to sort out the calling process of Dubbo 2.x; the more developed, the
>    more difficult it is to maintain, so it needs to be refactored in order
> to
>    develop better.
>    - Some extensions cannot meet current needs, such as service addresses
>    push monitoring, performing Filter early interception requests before
>    service addresses selection, printing the process of route selection,
> etc.
>    - The underlying Dubbo 2.X is based on synchronous calls. The
>    asynchronous call is wrapped synchronously. This introduces inaccuracies
>    such as RT during asynchronous calls.
>    - At present, the service metadata information and the address of the
>    Dubbo service are coupled together. Most of the metadata is the same,
> and
>    the metadata is redundant. At the same time, as the size of the machine
>    increases, the Registry is under great pressure, especially when machine
>    restart.
>    - The service registry and the configuration center are coupled
>    together, and after splitting, the management and storage of service
>    governance rule data such as routing rules can be better handled.
>    - The core is written based on JDK 1.6 and cannot use the newer JDK
>    features. It is recommended to upgrade JDK to 1.8 and above.
>
> For the above reasons, in order to better develop Dubbo and attract more
> people to participate in the development of the Dubbo community, I propose
> to refactor the Dubbo 2.x code. The newly developed version number is Dubbo
> 3.x. Refactoring Dubbo 3.x will have the following advantages:
>
>    - Refactoring core modules to make the code more concise and
>    understandable
>    - Refactoring the routing process, Making the route process to Pipeline
>    style,Making the route process can be observed
>    - Add new extension points to make it easier for developers to extend
>    the framework
>    - Add the Service Domain Model to enable developers to handle its
>    lifecycle
>    - Supports asynchronous interface, unified call process returns
>    CompleteFuture, synchronous call process uses CompleteFuture wrapper;
> Solve
>    the problem of inaccurate statistics such as RT during asynchronous call
>    - Splitting out the Configuration Center and Registry for better control
>    and support for larger clusters
>    - Prepare for new features such as Http2 stream、Reactive stream
>    - Performance improvement
>    - And so on
>
> Of course, refactoring Dubbo 3.x also faces the following challenges:
>
>    - A wide range of refactorings that can cause extension points to be
>    incompatible
>    - The Dubbo team may need to maintain two branches at the same time for
>    a longer period of time
>
> For the following challenges, we can do this:
>
>    -
>
>    For major extension points, such as Fitler, etc., be sure to ensure
>    compatibility and enable developers to smoothly upgrade
>    -
>
>    For some extension points, the Dubbo team will listen to the community
>    and selectively fix compatibility issues with some extension points.
>
> Comments welcome!
>
> thank you ~
>

Re: Proposal for Dubbo 3.0

Posted by Huxing Zhang <hu...@apache.org>.
Hi,

On Mon, Jul 9, 2018 at 3:32 PM, Liujie Qin <li...@apache.org> wrote:
>   Hi All, Dubbo has become a excellent RPC framework with the efforts of
> various partners in the community. However, in the process of development,
> we found some shortcomings of Dubbo 2.x, which is listed as follows:
>
>    - The maintainability is relatively poor. The code uses a lot of
>    dynamically generated classes. It is difficult to understand and difficult
>    to debug. The code hierarchy is not clear enough, and the calling logic is
>    confusing. As an experienced Dubbo developer, sometimes it is difficult for
>    me to sort out the calling process of Dubbo 2.x; the more developed, the
>    more difficult it is to maintain, so it needs to be refactored in order to
>    develop better.
>    - Some extensions cannot meet current needs, such as service addresses
>    push monitoring, performing Filter early interception requests before
>    service addresses selection, printing the process of route selection, etc.
>    - The underlying Dubbo 2.X is based on synchronous calls. The
>    asynchronous call is wrapped synchronously. This introduces inaccuracies
>    such as RT during asynchronous calls.
>    - At present, the service metadata information and the address of the
>    Dubbo service are coupled together. Most of the metadata is the same, and
>    the metadata is redundant. At the same time, as the size of the machine
>    increases, the Registry is under great pressure, especially when machine
>    restart.

+1 to separate service registration data and service configuration data.

>    - The service registry and the configuration center are coupled
>    together, and after splitting, the management and storage of service
>    governance rule data such as routing rules can be better handled.
>    - The core is written based on JDK 1.6 and cannot use the newer JDK
>    features. It is recommended to upgrade JDK to 1.8 and above.

+1 to rewrite Dubbo using 1.8

>
> For the above reasons, in order to better develop Dubbo and attract more
> people to participate in the development of the Dubbo community, I propose
> to refactor the Dubbo 2.x code. The newly developed version number is Dubbo
> 3.x. Refactoring Dubbo 3.x will have the following advantages:
>
>    - Refactoring core modules to make the code more concise and
>    understandable
>    - Refactoring the routing process, Making the route process to Pipeline
>    style,Making the route process can be observed
>    - Add new extension points to make it easier for developers to extend
>    the framework
>    - Add the Service Domain Model to enable developers to handle its
>    lifecycle
>    - Supports asynchronous interface, unified call process returns
>    CompleteFuture, synchronous call process uses CompleteFuture wrapper; Solve
>    the problem of inaccurate statistics such as RT during asynchronous call
>    - Splitting out the Configuration Center and Registry for better control
>    and support for larger clusters
>    - Prepare for new features such as Http2 stream、Reactive stream

Strong +1

>    - Performance improvement
>    - And so on
>
> Of course, refactoring Dubbo 3.x also faces the following challenges:
>
>    - A wide range of refactorings that can cause extension points to be
>    incompatible
>    - The Dubbo team may need to maintain two branches at the same time for
>    a longer period of time

If there were 3.x, we will have:

* 2.5.x: Java 6+
* 2.6.x: Java 6+
* 2.7.x: Java 8+
* 3.0.x: Java 8+

What is the difference between 2.7.x and 3.0.x?

>
> For the following challenges, we can do this:
>
>    -
>
>    For major extension points, such as Fitler, etc., be sure to ensure
>    compatibility and enable developers to smoothly upgrade
>    -
>
>    For some extension points, the Dubbo team will listen to the community
>    and selectively fix compatibility issues with some extension points.

I suggest we note down all the incompatible changes and provide a
migration guide for them.


>
> Comments welcome!
>
> thank you ~



-- 
Best Regards!
Huxing