Javier Fernández-Sanguino Peña schrieb: Hm...
> > As I suspected the bug occured again after I updated my system with
> > aptitude.
> Strange..
I am not so sure of this connection any more... I just restarted cron
by accident (while trying to attach a debugger to it, because I
couldn't think of a better method to catch the child than to break on
"getpid" in gdb, I inadvertently killed and restarted cron). After
that, cronjobs worked again and to reproduce the bug, I did an update
with aptitude. Now cron is still working (unfortunately).
Other candidates for succesful reproduction of the bug are
- wait a few days (sounds funny, I know)
- Crashes of the machine (we had a powerfailure in our flat
yesterday, and I had some problems with broken memory lately)
> - monitor when cron launches its children (pid A)
> - send kill -SIGSTOP to pid A as soon as you see it
> - run 'gdb /usr/sbin/cron A'
> - send 'kill -SIGCONT A' and 'run' in gdb.
> - when you get the sigsegv in the cron job run 'bt'
The child segfaults very quickly. Can you think of a way to stop the
child automatically?
I tried
strace -f -p `pidof cron` -o '|sh -c "grep getpid|awk \"{print $5}\"|while read
pid; do echo PID: $pid; kill -STOP $pid; done"'
But there's a bug in that line that I can't spot.
Would it work to simply stop cron and attach gdb to it and then
"cont"?
Well, I'll see when this bug reappears.
--
Friedrich Delgado Friedrichs <[EMAIL PROTECTED]>
signature.asc
Description: Digital signature
