build-manager loses track of buildds sometimes

[LaMont Jones]

Sometimes, build-manager decides to ignore a buildd until at least tickled. Frequently, marking the buildd manual and then returning it to auto is sufficient to make buildd-manager see it again, other times nothing seems to make buildd-manager
notice the buildd again. In today's case, the last log entry for berkelium is:

2010-03-08 12:11:42+0000 [-] Builder 'berkelium' rescued from '1550051-3165289': 'No job assigned to builder'

And then nothing. -->manual-->auto did nothing, nor did marking it !OK, !active, and back.

recycled berkelium to see if that makes any difference, will follow up after that finishes.

lamont

[LaMont Jones (lamont)]

recycling the ppa host worked around the issue.

lamont

[Michael Nelson (michael.nelson)]

This looks a lot like bug 463046?

[William Grant (wgrant)]

My initial response was the same as Michael's.

[William Grant (wgrant)]

LaMont: if this happens again, please run "import xmlrpclib; xmlrpclib.ServerProxy('http://somebuildd:8221/rpc').status()" in a Python interpreter and take note of the returned string. I suspect it will say "BuilderStatus.ABORTING", which would confirm that this is bug #463046.