[Buildbot-devel] Buildbot-0.7.12rc2

[Dustin J. Mitchell]

On Mon, Jan 11, 2010 at 3:37 AM, Axel Hecht <l10n.moz at googlemail.com> wrote:
> I don't think it's testNoMerges, tbh. The test alone runs fine, even though
> I'm able to reproduce the failure if running the whole test suite. It
> complains about the reactor not being empty, so I blame a previous test for
> not cleaning up. My uneducated guess would be test_maildir, though I have no
> clue why, that test didn't change at all.
>
> trial doesn't let me run a subset of tests in order, so it's hard to figure
> out which combination of test cases actually triggers it.

Nicolas Alvarez did this analysis (by patching trial to run things in
order). This was what he found:

buildbot.test.test_limitlogs.MasterLogs.testLog
buildbot.test.test_locks.Locks.testLock1
buildbot.test.test_locks.Locks.testLock2
buildbot.test.test_locks.Unit.testRandomCounting
buildbot.test.test_mergerequests.MergeRequestsTest.testDefault

either of the testRandom* tests were able to trigger this.

There's been some talk about this in IRC, and the suggestion has been
to run each test in its own process, for better isolation.  This would
probably mask whatever environment-pollution is occurring here, but at
least it would help us keep the test suite universally functional.  As
it is right now, it mysteriously fails, and nobody's been able to
figure out why.

Does anyone know how to do that with trial?

Dustin

-- 
Open Source Storage Engineer
http://www.zmanda.com