On Tue, Jan 6, 2009 at 3:46 PM, Paul Burba <ptburba_at_gmail.com> wrote:
> On Tue, Jan 6, 2009 at 3:41 PM, <buildbot_at_mobsol.be> wrote:
>> Full details are available at:
>> http://www.mobsol.be/buildbot/win32-xp%2520VS2005/builds/1584
>>
>> Author list: julianfoad
>>
>> Build Slave: djh-xp-vse2005
>>
>>
>> Subversion Buildbot
>> http://www.mobsol.be/buildbot/
>>
>>
>> Last 100 lines of the build log (step: Test fsfs+ra_local ):
>>
>> Traceback (most recent call last):Failure: buildbot.slave.commands.TimeoutError: SIGKILL failed to kill process
>>
>> ------------------------------------------------------
>> http://subversion.tigris.org/ds/viewMessage.do?dsForumId=552&dsMessageId=1008403
>
> I'm seeing this on on my machine too, the test suite hangs on
> prop_tests.py 10, but only when run in parallel, and even stranger,
> only when run in --verbose mode. Looking into this now...
The hanging tests on my machine (and I hope DJ's buildbot) were caused
by Arferver's change to subprocess.Popen()/subprocess.Popen.wait() in
r34863. After that change, when running the tests in parallel there
was a deadlock when prop_tests.py 10 'try to set inappropriate props'
tries a whole slew of invalid propsets, producing a lot of stderr...at
least that's what appears to have been happening. Per the Python docs
this as a known/common problem with Popen.wait():
Warning - This will deadlock if the child process generates enough output
to a stdout or stderr pipe such that it blocks waiting for the OS pipe
buffer to accept more data. Use communicate() to avoid that.
So I switched from subprocess.Popen.wait() to
subprocess.Popen.communicate() in r35093 and that fixed the problem
(though it wasn't quite so simple as that).
Paul
Received on 2009-01-09 00:37:49 CET