[wildfly-dev] JBoss MSC Plans

[Richard Opalka]

Hi All,

   comments in-line:

On Fri, Jan 5, 2018 at 8:31 PM, Brian Stansberry <
brian.stansberry at redhat.com> wrote:

> Thanks, Richard. Comments in-line.
>
> On Thu, Jan 4, 2018 at 7:56 AM, Richard Opalka <ropalka at redhat.com> wrote:
>
>> Hi everyone,
>>
>>    I would like to outline a brief JBoss MSC Plans for Year 2018.
>>
>> Goals defined below address both:
>>  * Cloud First Effort
>>    - Reduce WildFly memory footprint
>>
>
> Do you have a sense of how much we're talking about here, e.g. from a
> standalone server using the standalone-full.xml config and no deployments?
>

If we'd replace injections, values and injectors with new MSC2 like
dependencies than I think we would be still talking about just small memory
improvements.
But such change might have better impact on MSC performance (from
concurrency point of view).
Anyway main reason of doing it is to approach MSC1 to MSC2 as much as
possible.
Then potential switch from MSC1 to MSC2 in the future would be more
straightforward process.

>
>
>
>>    - Speed up WildFly boot time
>>
>
> Same question here. A reload takes a couple hundred ms of which my guess
> is MSC is a quite small part.
>

The improvement here might be better comparing to memory improvements.
Simplified MSC state machine (and less MSC states) results in less context
switches.
I have a local prototype where we have been able to eliminate one third of
MSC states
(and I believe we could simplify MSC state machine even further in the
future).
I don't have numbers yet but this change should improve WildFly boot
process and deployments.


>
>
>>  * Potential Migration to MSC 2
>>
>> MSC Library Goals:
>>  * SHORT TERM GOALS (MSC 1.2.x, 1.3.x)
>>    [1] simplify MSC state machine - deprecate useless MSC internal states
>>        (e.g. REMOVED, WONT_START, WAITING, REMOVING, STOPPING maybe
>> others)
>>    [2] deprecate API exposing MSC internals
>>        (e.g. replace ServiceListener with LifecycleListener)
>>    [3] deprecate MSC Values, Injections & Injectors and provide
>> alternative API
>>        (alternative API should have same minimalistic memory requirements
>> like MSC 2)
>>  * LONG TERM GOALS (MSC 1.4.x)
>>    [4] remove deprecated APIs and all deprecated stuff
>>
>> WildFly MSC Integration Goals:
>>  + SHORT TERM GOALS (WildFly 12)
>>    * replace ServiceListeners with LifecycleListeners (LifecycleListeners
>> don't expose MSC internals)
>>    * eliminate MSC optional dependencies from WildFLy Core
>>
>
> Sounds good. The basic issue there is the modeling of optional module
> dependencies using MSC services, and IIRC you have an alternate approach in
> mind.
>

I didn't give it much thinking yet, but at the moment there are two
alternatives I am aware of:
1) Let JBoss Modules do handling of optional dependencies on its own and
don't map optional modules to MSC optional dependencies
    - the idea is present in this commit:
https://github.com/ropalka/wildfly-core/commit/c65841c1892f02e14a14101cca4035f89c932ce5
      (and all WildFly-Core and WildFly testsuites are passing with this
commit)
2) Workaround MSC optional dependencies for optional modules somehow
    - that's my next TODO task

DomainModelControllerService also has one for some reason but I figure that
> can be eliminated easily enough.
>

This I hope should be easy to workaround via capabilities (i.e. install
REQUIRED dependency iff some capability is present).


>
>
>>    * eliminate MSC Values, Injections & Injectors and migrate to
>> alternative API
>>
>
> This sounds like a very large task for WF 12, for which only some weeks of
> dev time remain.
>

This sounds more complex than it is. But of course there can be use cases
that might complicate this task.


>
>
>>    * WildFly Management Layer should not expose MSC APIs as its public API
>>      (complicates potential migration to MSC2 and breaks encapsulation)
>>
>
> I can't see this happening in WildFly 12. Let's talk about it when we meet
> up at the end of the month.
>

We definitely need to talk about it and come to a conclusion.


>
>  + LONG TERM GOALS (WildFly versions targeting next major EAP)
>>    * eliminate MSC optional dependencies from WildFly (probably via
>> Capabilities)
>>      (MSC Optional Dependencies have been fixed recently but there's
>> still performance issue
>>      - in worst case scenario if service A has N optional dependencies
>>      it may happen service A is restarted N times before it is stabilized
>> in UP state)
>>
>> Feedback and comments more than welcome!
>>
>> --
>> Rio
>>
>>
>> _______________________________________________
>> wildfly-dev mailing list
>> wildfly-dev at lists.jboss.org
>> https://lists.jboss.org/mailman/listinfo/wildfly-dev
>>
>
>
>
> --
> Brian Stansberry
> Manager, Senior Principal Software Engineer
> Red Hat
>

Thanks for asking.

-- 
Rio
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://lists.jboss.org/pipermail/wildfly-dev/attachments/20180107/5d9c5276/attachment.html