[Buildbot] #1001: start_instance on a latent BuildSlave called before stop_instance finished

[Buildbot trac]

#1001: start_instance on a latent BuildSlave called before stop_instance finished
---------------------------+--------------------
Reporter:  moschny         |       Owner:
    Type:  defect          |      Status:  new
Priority:  major           |   Milestone:  0.9.+
 Version:  master          |  Resolution:
Keywords:  virtualization  |
---------------------------+--------------------

Comment (by bshi):

 With GCE latent build slaves and a busy cluster of 8, we are running into
 this problem quite frequently now - a slave gets into a bad state at about
 1 per two days for us (8, 16-core slaves, low-hundreds of builds a day).
 There may be areas where we could be much more defensive with our use of
 the GCE API but it would be nice for buildbot's state management to be
 more protective.

--
Ticket URL: <http://trac.buildbot.net/ticket/1001#comment:9>
Buildbot <http://buildbot.net/>
Buildbot: build/test automation