[Buildbot-devel] Buildbot under Centos 4.2 causes exit code 129 when running ant build step

[Brian Warner]

> Thanks for the pointer to nohup Brian,  using nohup results in a
> successful build, however no logs and no status can be extracted as
> everything gets written to hohup.out.

Actually I was thinking of running the entire buildslave under nohup, but
really I don't know if that would help matters or make them worse.

> I have been scouring through the buildbot code attempting to find out
> why the ant task succeeds on suse 9.1 with kernel 2.4.21 but dies on
> Centos 4 with kernel 2.6.9-22.  It is definitely linked to SIGHUP, but
> why is the ant command receiving a SIGHUP signal ?  I've found a comment
> within buildbot/slave/commands.py<Line 80> that talks about a bug in
> recent debian glibc and SIGHUP.  Could this be related ?

Hmm, yes, I think so. (I'd completely forgotten about that comment). I
checked my notes, and couldn't find any further information behind that
remark.. as best I can remember, I stopped using PTYs and the problem went
away (set usepty=0 in the buildslave's buildbot.tac file). I think it only
affects processes that use threads, and most of mine don't these days, so
that might be a second reason I haven't seen that problem in a while.

It may be that closing stdin (just after that comment) on a PTY causes it to
send SIGHUP to the child process.. but only when using threads?? I dunno. Try
commenting out the 'self.transport.closeStdin()' and see if that affects
anything.

mystified,
 -Brian