[Buildbot-devel] Possible SVNPoller bug with multiple SVN codebases (buildbot-0.8.10-pre)

Xavier Francoi xavier.francois3981 at gmail.com
Fri Aug 8 17:05:16 UTC 2014


Thanks for the info/links!  I had a look but wasn't able to use it to
effectively pinpoint what is happening in my environment (ie. why the
svnpoller is wedging itself so frequently).

As mentioned earlier, I've been restarting the buildbot processes to
resolve this issue.  However this approach has its problems because
buildbot will not grab the recent version for the broken SVN repo until a
new change is triggered for that repo.  For example, if after restarting
the buildbot processes a code checkin is done to the previously non-broken
SVN repo then buildbot will still use the old version for the previously
broken SVN repo.  A "fake" code checkin could be pushed to the broken SVN
repo to force buildbot to sync.  However this isn't practical since it may
trigger a plethora of processes in other buildbots systems monitoring this
repo, thereby needlessly exhausting compute resources.

What would be the easiest way to set up a test environment to debug the
issue?  Maybe there is a more effective way to tackle this than by just
looking at twisted.log, svnpoller.py and my buildbot systems that
eventually fail.

Thanks!



On Thu, Aug 7, 2014 at 7:28 PM, Charles Lepple <clepple at gmail.com> wrote:

> On Thu, Aug 7, 2014 at 4:54 PM, Xavier Francoi
> <xavier.francois3981 at gmail.com> wrote:
> > The issue: svnpoller stops detecting revision changes to one of the SVN
> > repositories after some time.  Initially the svnpoller functions
> correctly
> > (ie detect code checkins so test suites can be triggered) but after a
> week
> > or more the svnpoller stops detecting code check-ins to one of the
> > repositories.
>
> I'm not sure I ever saw SVNPoller wedge itself after only a week, but
> occasionally (after several months?) I would see SVN connections to
> alioth.debian.org stay open long past the poll interval. (Usually this
> was when Alioth was heavily loaded, so I would just restart the poller
> later.) Eventually we moved to Git, and switched to a push model, so I
> never had enough information for a complete bug report.
>
> There is some good background information here:
> http://trac.buildbot.net/ticket/2183
>
> --
> - Charles Lepple
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://buildbot.net/pipermail/devel/attachments/20140808/424e0b77/attachment.html>


More information about the devel mailing list