[Buildbot-devel] buildbot from git hangs during setup until slave connects?

[Dustin J. Mitchell]

On Tue, Sep 11, 2012 at 3:41 PM, Dan Kegel <dank at kegel.com> wrote:
> Murphy has truly blessed me with his presence today.
>
> On an Ubuntu 10.04 host, running VMWare player 5.0.0, and inside that
> Ubuntu 12.04, and installed on that buildbot from git (fresh from git
> this morning), my buildbot master gets part of the way through
> startup, then hangs on adding the first scheduler (a force scheduler, hmm!)
> before it starts up the web status listener.
>
> Two things will unstick it:
> 1) if a slave connects (or if I telnet to the slave port)
> 2) if I strace the process (with strace -p)
>
> W, I say, TF?
>
> I don't recall this happening before.

Oh dear, this sounds like a Python or Twisted bug of some sort.
Generally when strace will "fix" things, it means that some change is
not being properly communicated to Twisted via the select() call.  In
older versions, this was sometimes due to missing a SIGCHLD signal
when a child exited before the signal handler was installed.  But
schedulers don't use subprocesses.  So that's odd.

When you do strace it, what do you see?

Dustin