[OpenStack-Infra] [Zuul] Understanding dequeue-on-new-patchset

James E. Blair corvus at inaugust.com
Thu Aug 21 23:17:05 UTC 2014


Jeremy Stanley <fungi at yuggoth.org> writes:

> On 2014-08-21 15:11:13 -0700 (-0700), James E. Blair wrote:
> [...]
>> Strictly speaking, that will mean that every patchset will go through
>> the merger and Jenkins.  But if testing for a patchset is in progress
>> when a new patchset is uploaded, the tests will abort.
> [...]
>
> One corner case I think we've seen which could lead to confusion is
> that once a job enters the Gearman queue it doesn't get removed even
> if Zuul removes it from the pipeline... so if your worker volume is
> backed up enough that some jobs for a patchset haven't been assigned
> before another patchset causes the current one to get dequeued,
> Jenkins can end up running those jobs anyway.
>
> Did we ever end up implementing a solution for that, or does that
> behavior still exist?

Jobs in the Gearman queue are definitely supposed to be canceled; if
not, that's a bug (and not a known one).

-Jim



More information about the OpenStack-Infra mailing list