[Nottingham] Unkillable process?

[Graeme Fowler]

Hi

On 24/04/2006 17:06, Johannes Kling wrote:
> No luck, unfortunately :-/. SEGV, PIPE, QUIT, USR1, USR2 and ABRT were
> similarly ignored.

Ugh. This is a Bad State.

> The script's got its cwd open, a few libraries; stdout and stderr are
> attached to a logfile, stdin seems to be an unconnected pipe. The
> patch process has nothing open (as mentioned, though, it looks like a
> zombie in /proc).

Unfortunately I have a sneaking feeling that this is, in the best 
tradition of using completely inappropriate words to describe situations 
withing computers, broken ;-)

Are you getting support from RedHat for this machine? I ask as it could 
be a kernel problem and they might be able to help. You could search 
RedHat's bugzilla for similar conditions.

As a last resort (after rebooting the server, obviously), you could run 
the whole shebang under strace and pipe the output via "tee" to a text 
file. It's awfully noisy, and sometimes it helps stop the problems (thus 
making debugging even more difficult) but it could well show you what 
the process (and children) were doing at the time they snuffed it.

The fact that you get absolutely zero output after attaching strace to 
it makes me think that it is, indeed, stuck chit-chatting with a kernel 
thread that you can't see (you might if it interrupted and started 
again, as you'd get chance to attach to the thread then) *or* something 
spooky has gone on with the child process "patch" and similar unreadable 
chit-chat is going on there.

Have you tried signalling either process with strace attached, 
outputting to a file?

Graeme