[Buildbot-devel] process killed by signal 1

[Brian Warner]

> The problem we have, (as you can see from the logs), is that the debian
> buildbot is dying with a:
> process killed by signal 1
> program finished with exit code -1
> 
> Now this is very strange, becaues if I run trial manually, with the same
> environment as buildbot is running in, all the tests run successfully
> and I get no 'process killed' problems.

That *is* weird. Since signal 1 is SIGHUP, my hunch is that the buildslave is
running the test under a PTY, and the somehow the PTY is closing slightly
early (after the child process has emitted the final test-count line, but
before it actually terminates). It's like logging into a remote host,
starting a process, then logging out.. the process may or may not be touched,
depending upon how it was started (with 'nohup', with '&', depends upon the
shell, etc). I think the shell uses SIGHUP to tell the process that its
terminal just went away.

These days I'm leaning away from using PTYs for child-process management, but
that's mostly because I don't have to deal with the environment which
prompted their use in the first place (unit tests which created several
non-daemonized child+grandchild processes, where a test failure could leave
those children running, so using a PTY provided a process group that made it
easy to kill them all off at the same time, at least under solaris).

If the problem persists, try telling the slave to not use a pty (by setting
'usepty=0' in the buildbot.tac file) and see if that affects it.

let me know if that helps,
 -Brian

PS: it looks like your buildbot is hitting the same problem that Nick patched
a little while back, where a restarted buildmaster is overlapping build
numbers with the earlier invocation, and an assert() statement is killing the
logfile creation.

PPS: you might also want to change your Trial step to use --bwverbose instead
of --verbose, so you don't get the ANSI color escape sequences in your
logfiles.