[Buildbot-commits] [Buildbot] #2455: buildbot silently reuses existing run #s
Buildbot
nobody at buildbot.net
Wed Feb 20 22:13:13 UTC 2013
#2455: buildbot silently reuses existing run #s
--------------------+------------------------
Reporter: virgilg | Owner:
Type: defect | Status: new
Priority: major | Milestone: undecided
Version: 0.8.7p1 | Resolution:
Keywords: |
--------------------+------------------------
Old description:
> After upgrading from 0.8.5 to 0.8.7p1 we see these:
>
> 2013-02-20 12:52:03-0800 [-] Warning: Overwriting old serialized Build at
> /Users/buildbot/buildmaster/tests-prerelease/5011-log-
> Deleting__Users_buildbot_buildslave_tests-prerelease_build_Logs.-stdio
>
> and our run #s are all messed up.
> It seems to happen immediately after a reconfigure.
New description:
After upgrading from 0.8.5 to 0.8.7p1 we see these:
{{{
2013-02-20 12:52:03-0800 [-] Warning: Overwriting old serialized Build at
/Users/buildbot/buildmaster/tests-prerelease/5011-log-
Deleting__Users_buildbot_buildslave_tests-prerelease_build_Logs.-stdio
}}}
and our run #s are all messed up.
It seems to happen immediately after a reconfigure.
--
Comment (by dustin):
So, all of the bugs you've found are pointing to your having multiple
buildmasters running in the same directory.
In this case, there's only function which decides on the build number
(`determineNextBuildNumber`). Twisted is single-threaded, so that
function can't race with itself in the same process. So this could be
caused by two processes trying to create build pickles in the same
directory.
Another alternative is running the buildmaster on a lame network
filesystem where files do not appear immediately in the directory listing
after they are created.
Either option would explain duplicate files in the status_push queue, as
well as the DB corruption. The failure to receive signals might be caused
by two masters running at once, but not by bad NFS.
--
Ticket URL: <http://trac.buildbot.net/ticket/2455#comment:1>
Buildbot <http://buildbot.net/>
Buildbot: build/test automation
More information about the Commits
mailing list