[Buildbot-devel] trial buildbot.test failures on NetBSD

[Jeremy C. Reed]

On Tue, 7 Jul 2009, Dustin J. Mitchell wrote:

> Looking at another failure, this time
> buildbot.test.test_run.Disconnect.testBuild2 as run on your system by
> the metabuildbot:
> 
> 2009-07-07 15:31:52-0500 [Broker,0,127.0.0.1] acquireLocks(step
> <buildbot.steps.dummy.Dummy instance at 0x7f7ff6e62dd0>, locks [])
> 2009-07-07 15:31:54-0500 [-] doing shutdownAllSlaves
> 
> this test's intent is to start a build (at 15:31:52, and then 0.5s
> later kill it.  The build is only slated to take one second, yet in
> the timing above we see two seconds elapsed.
> 
> It looks like the twisted reactor is not doing well at getting the
> timing right.  From what you wrote earlier, this is also affecting the
> Twisted unit tests.  See what happens when you run these tests with a
> different reactor?  trial has the --help-reactors switch that lists
> the available reactors, and the --reactor= option to specify one.

Still had two seconds when using poll versus select. (I don't know what 
default is and kqueue is not working due to "No module named kqsyscall".)

But with --reactor=poll it failed with

Failure: twisted.spread.pb.PBConnectionLost: [Failure instance: Traceback 
(failure with no frames): <class 'twisted.internet.error.ConnectionDone'>: 
Connection was closed cleanly.

And with --reactor=select it failed with:

Failure: twisted.spread.pb.PBConnectionLost: [Failure instance: Traceback 
(failure with no frames): <class 'twisted.internet.error.ConnectionLost'>: 
Connection to the other side was lost in a non-clean fashion.

I am doing a complete run with --reactor=poll now.... and then I will do 
with --reactor=select ... but it takes awhile.

> I'm sure there's a Twisted bug on this topic somewhere, but google is
> not finding it for me.