[Buildbot-commits] [Buildbot] #1992: master hangs on startup

[Buildbot]

#1992: master hangs on startup
-------------------+----------------------
Reporter:  dustin  |       Owner:
    Type:  defect  |      Status:  new
Priority:  major   |   Milestone:  0.8.4p1
 Version:  0.8.4   |  Resolution:
Keywords:          |
-------------------+----------------------

Comment (by dustin):

 I've recorded some notes here:
  https://gist.github.com/1033514
 basically, it looks like sometimes select() doesn't notice that a file
 descriptor is readable, which makes the selfpipe trick that Twisted uses
 to support callFromThread nonfunctional.  This seems like an enormous bug
 either in the kernel or in Python.  I can't narrow it down further because
 dtrace doesn't work very well on a Mac - in particular, it doesn't have a
 way to actually follow child processes (-f doesn't work).

 So I think that the fix, for now, to get 0.8.4p1 out the door, is to just
 add a busyloop into the DBThreadPool that runs every second - that will
 wake up select() and cause things to proceed as desired.  It ain't pretty,
 but without more data, that's all I can do.

 Once that fix is in on the branch, I'll leave this bug open, to start to
 collect system statistics about where this occurs.  So far, the common
 thread seems to be *BSD - I know of one NetBSD and two Mac OS X
 reproductions.  I can't reproduce it on my mac.  I can reproduce it on
 redheadphones' mac using Twisted-8.2.0, 9.0.0, and 11.0.0, all with
 Python-2.6.1.  I've seen it replicated with Python-2.7, as well.

-- 
Ticket URL: <http://trac.buildbot.net/ticket/1992#comment:3>
Buildbot <http://buildbot.net/>
Buildbot: build/test automation