Thank you for the anecdote :)
>Then again, I don't think buildbot is
exactly designed to handle the vagaries of being reconfigured as often
as ours is, or having its database cleared as often.
Well it is designed to be reconfigured as often as possible.
Having the database cleared is however not part of the deal :)
I am not sure of a scenario were master will succeed to start but not have
the slave configuration loaded yet, and a sqlite remove which fixes the
Would have been a good idea to keep that sqlite for forensics. You might
indeed have an sqlite corruption, which we could detect and try provide a
What is sure is that it is not recommended to use sqlite for bigger prod
like you have.
Le lun. 19 sept. 2016 à 17:36, Neil Gilmore <ngilm...@grammatech.com> a
> Good morning everyone,
> Since this isn't really the place for bug reports, consider this an
> When I came in this morning, there was a message that buildbot's UI was
> unresponsive, which it was. Looking at the most recent twistd.log, the
> only thing I was was that the irc client had been bouncing up and down
> for the last 8 hours or so. Buildbot stop failed to stop the process.
> Killing it did. But restart wasn't very successful -- every worker was
> an 'unauthorized login'. Again buildbot stop failed to stop the process,
> so I killed it again. Deleted state.sqlite, ran buildbot upgrade-master,
> and started again. This appears to have worked.
> One problem we see when we have to do this is that worker indexes shift
> around when we have to delete the database. This makes any URLs in build
> emails that indexed into the previous database invalid. Not the worst
> problem ever, but inconvenient. Then again, I don't think buildbot is
> exactly designed to handle the vagaries of being reconfigured as often
> as ours is, or having its database cleared as often.
> Our current production system has 0.9.0rc1 for the master, and mostly
> the same for workers. Twisted is 16.2.0. The only changes I know of are
> a couple custom build steps we have in shell.py, a fix for a problem
> with __cmp__() in ComparableMixin (we'd been somehow getting a set in an
> object, and __cmp__() would fail during some reconfigs), and a slight
> change to the UI to always show the last 4 builds instead of the default
> When we do multi-master, it'll be rc2, at least to start. The troubles
> I'm having with that currently are not buildbot problems.
> As always, thank you for your time.
> Neil Gilmore
> users mailing list
users mailing list