[Buildbot-devel] no response from shell command during update

[Brian Warner]

brett <bneely at gmail.com> writes:

> During the update phase of my builds, my step.SVN source calls rm to
> remove the slave's build directory.  The command doesn't return
> anything to the master, the command times out, and the build fails.
> Here's the stdio log for the update task:
> ...
> command timed out: 1200 seconds without output, killing pid 571

> Does this look familiar to anyone?  I tried launching twistd with the
> --spew parameter but didn't know how to filter the output.

Weird. Take a look at 'ps' to see if the rm command has been reaped by the
buildslave.. maybe the SIGCHLD (sent by the kernel to the buildslave to tell
it that the child process has died and its exit status needs to be read) got
lost somehow. Try adding some debug messages (either log.msg() calls to write
things into the twistd.log in the buildslave's directory, or
self.sendStatus({'header': text}) calls to send text up to the buildmaster)
in buildbot/slave/commands.py in the ShellCommand._startCommand method, and
set ShellCommandPP.debug=True to enable a lot of slave-side logging about
child processes starting and stopping.

Finally, you might try using 'rm --verbose' to get more text out of the
step.. that might make it easier to tell the difference between the command
running normally (but not producing any output) and the command not getting
started at all.


puzzled,
 -Brian