[Buildbot-commits] [Buildbot] #2022: Stale Broker errors not caught on reconfig

[Buildbot]

#2022: Stale Broker errors not caught on reconfig
-------------------+----------------------
Reporter:  dustin  |      Owner:
    Type:  defect  |     Status:  new
Priority:  major   |  Milestone:  0.8.5
 Version:  0.8.2   |   Keywords:  reconfig
-------------------+----------------------
 We're seeing reconfigs "wedged", meaning they never complete.  As I
 understand it, here's what's happening:

 Due to bad luck, a !BuildSlave.slave reference is stale when the reconfig
 begins.  I think that this could be minimized, but not avoided entirely,
 since the attach/detach process is not quite atomic at this point.  Given
 that we reboot a slave every few seconds, it's not surprising.

 When this happens, the {{{setBuilderList}}} remote call fails with
 {{{twisted.spread.pb.DeadReferenceError}}}, which is not handled anywhere.
 As a result, the {{{setBuilderList}}} call effectively never completes,
 and the reconfig stalls.

-- 
Ticket URL: <http://trac.buildbot.net/ticket/2022>
Buildbot <http://buildbot.net/>
Buildbot: build/test automation