[Gluster-devel] regressions on release-3.7 ?

[Atin Mukherjee]

On 04/20/2015 06:44 PM, Jeff Darcy wrote:
>> The same problems that affect mainline are affecting release-3.7 too. We
>> need to get over this soon.
> 
> I think it's time to start skipping (or even deleting) some tests.  For
> example, volume-snapshot-clone.t alone is responsible for a huge number
> of spurious failures.  It's for a feature that people don't even seem to
> know we have, and isn't sufficient for us to declare that feature
> supportable, so the only real effect of the test's existence is these
> spurious regression-test failures.  In other cases (e.g. uss.t) bugs in
> the test and/or the feature itself must still be fixed before we can
> release 3.7 but that doesn't necessarily mean we need to run that test
> for every unrelated change.
Commit 1efa50861b2cee68de9c9b751d9fc5eed08f5e5b (related to glusterd
urcu) introduced this problem. I would say test cases like this actually
unfolds the real problem which is hard to catch while coding. snapshot
clone consumes 16 threads in syncop at one point of time and this is why
when the execution begins after callback, its initiated by a different
thread and since urcu uses local thread variables unlocking from a
different thread crashes. We already have a patch
http://review.gluster.org/#/c/10285/ (review expected) to address this
issue. IMO, if we start skipping test cases we may not be able to
uncover bugs. Thoughts?

~Atin
> 
> The purpose of a regression test is to catch unanticipated problems with
> a patch.  A test that fails for its own unrelated reasons provides
> little or no information of that nature, and is therefore best treated
> as if no test for that feature/fix had ever existed.  That's still bad
> and still worthy of correction, but at least it doesn't interfere with
> everyone else's work.
> _______________________________________________
> Gluster-devel mailing list
> Gluster-devel at gluster.org
> http://www.gluster.org/mailman/listinfo/gluster-devel
> 

-- 
~Atin