[Gluster-devel] [Gluster-infra] Reduce regression runs wait time - New gerrit/review work flow

[Kaushal M]

On Thu, Jun 18, 2015 at 1:50 AM, Niels de Vos <ndevos at redhat.com> wrote:
> On Mon, Jun 15, 2015 at 04:19:14PM +0530, Kaushal M wrote:
>> Hi all,
> ...
>> I propose that we move back to the old model of starting regression
>> runs only once the maintainers are ready to merge. But instead of the
>> maintainers manually tiggering the runs, we could automate it.
>
> I think auto triggering regression tests is good. We should ask the
> developers to run regression tests before posting complex changes. If
> the parallelisation of regression tests is done, the wait time should
> reduce too.
>
> As a maintainer that spends quite some time reviewing patches, I prefer
> to see a +1 verified before I start to review something. With that, I at
> least have some confidence that there are no obvious mistakes I need to
> point out. If developers have to wait on me before regression testing
> gets started, I feel more like a block on road than helping them. There
> really are *many* patches that get a FAILED result where there is a
> problem in the code. Developers should get a response about that as soon
> as possible, and waiting for a maintainer to start the regression tests
> does not help.

The really obvious mistakes should get caught by the smoke tests. With
the proposed workflow, we will still run smoke whenever a change is
submitted. The result of this will still be communicated back to
Gerrit, it could be in the form of a Verified+1 flag, or could be
under another name.

If we feel that the current smoke tests aren't satisfactory enough to
build confidence, then we must improve them.

>
> I also had to ask maintainers for triggering regression tests for my
> first patches, it is not a nice experience. Anything we can do to
> improve the experience for (new) developers should be done, delaying
> (auotmated) feedback isnt a step in the right direction.
>
>> We can model our new workflow on those of OpenStack[1] and
>> Wikimedia[2]. The existing Gerrit plugin for Jenkins doesn't provide
>> the features necessary to enable selective triggering based on Gerrit
>> flags. Both OpenStack and Wikimedia use a project gating tool called
>> Zuul[3], which provides a much better integration with Jenkins and
>> Gerrit and more features on top.
>
> More intelligent triggering would be helpful. Unfortunately we have a
> stack of xlators and it is difficult to say if there are unintended
> side-effects in different, untouched pieces of the code.
>

I only started looking at Zuul, because I believed that the existing
Gerrit-trigger plugin for Jenkins didn't support triggering based on
flags. But your reply in another mail thread says otherwise. If it is
possible to achieve the workflow without using Zuul, I'm all for it.

When talking about intelligent triggering, I was referring to the
possibility of ordering several different types of jenkins jobs, and
not selection of a subset of regression tests.

>
>> I propose the following work flow,
>>
>> - Developer pushes change to Gerrit.
>>   - Zuul is notified by Gerrit of new change
>> - Zuul runs pre-review checks on Jenkins. This will be the current smoke tests.
>>   - Zuul reports back status of the checks to Gerrit.
>>     - If checks fail, developer will need to resend the change after
>> the required fixes. The process starts once more.
>>     - If the checks pass, the change is now ready for review
>> - The change is now reviewed by other developers and maintainers.
>> Non-maintainers will be able to give only a +1 review.
>>   - On a negative review, the developer will need to rework the change
>> and resend it. The process starts once more.
>> - The maintainer give a +2 review once he/she is satisfied. The
>> maintainers work is done here.
>>   - Zuul is notified of the +2 review
>> - Zuul runs the regression runs and reports back the status.
>>   - If the regression runs fail, the process starts over again.
>>   - If the runs pass, the change is ready for acceptance.
>> - Zuul will pick the change into the repository.
>>   - If the pick fails, Zuul will report back the failure, and the
>> process starts once again.
>
> It would be nice if Zuul, in its last step, can pick the change on top
> of the latest HEAD, run the build/smoke test again, and only push the
> change when all is OK. We have seen patch/merge races where a
> function/define was changed, and an other patch used that
> function/define. These caused much issues when the branch failed to
> compile. Being able to prevent that would be very good.
>
>
>> Following this flow should,
>> 1. Reduce regression wait time
>
> "wait time" for what or who? The merging of the patch would still only
> happen after all tests are done. If something fails the last test, more
> people (reviewers and maintainer) need to spend additional time.
>

The wait time for a change to have regression run on it once ready.
Now, we have changes which have been given +1/+2 reviews, but keep
waiting for ages for their regressions to finish, because of all the
other jobs in the queue most of which probably shouldn't have been in
the queue in the first place. The main intention for this workflow is
to reduce this kind of waits for changes that are near merge state.

>> 2. Improve change acceptance time
>> 3. Reduce unnecessary  wastage of infra resources
>
> We could, and should optimize that in our parallel testing and educating
> develpers to only re-run regressions when needed. Splitting up the
> regression tests also makes it possible to only re-run a small part of
> the tests.
>
>> 4. Improve infra stability.
>
> Not sure if adding an other component and (complex?) configuration adds
> to "Improve infra stability". It would be nice to have a very minimal
> set of tools, and many people understanding them. With the current
> Gerrit and Jenkins configuration we have, we seem to be already very
> limited on people that can investigate issues.

I agree. I believe if we were to move to using tools like Puppet/Salt
to perform the management of these systems the manual effort required
to manage and maintain the systems will be reduced. I know that misc
was working on setting up Salt configurations to handle all the
Gluster community infra, but I do not know the current state of the
effort. This is something that should be done.

>
>
>> It also brings in drawbacks that we need to maintain one other piece
>> of infra (Zuul). This would be an additional maintenance overhead on
>> top of Gerrit, Jenkins and the current slaves. But I feel the
>> reduction in the upkeep efforts of the slaves would be enough to
>> offset this.
>>
>> tl;dr
>> Current auto-triggering of regression runs is stupid and a waste of
>> time and resources. Bring in a project gating system, Zuul, which can
>> do a much more intelligent jobs triggering, and use it to
>> automatically trigger regression only for changes with Reviewed+2 and
>> automatically merge ones that pass.
>>
>> What does the community think of this?
>
> It is a good suggestion, but I would wait with spending time on this
> until we see some results from the parallel testing. Also, as Michael
> mentioned, we should aim for tools that are packaged for EPEL-7, so that
> we can have our infrastructure well managed on RHEL or CentOS systems.
>

Fixing up the current issues with spurious test failures due to
GlusterFS bugs or the infra issues is more important to this proposal.
Implementing this as is, will just lead to more work for the
maintainers. Getting parallel testing ready will help and is
definitely required, but it is not essential.

> Thanks,
> Niels
>
>
>> ~kaushal
>>
>> [1]: http://docs.openstack.org/infra/manual/developers.html#automated-testing
>> [2]: https://www.mediawiki.org/wiki/Continuous_integration/Workflow
>> [3]: http://docs.openstack.org/infra/zuul/
>> _______________________________________________
>> Gluster-infra mailing list
>> Gluster-infra at gluster.org
>> http://www.gluster.org/mailman/listinfo/gluster-infra