Bug#779078: apache2-bin: event mpm: child segfault in notify_suspend causes parent to exit during log rotation

[Stefan Fritsch <sf@sfritsch.de>]

Hi,

On Tuesday 05 May 2015 14:01:56, Michael Ablassmeier wrote:
> we recently migrated over an OTRS instance from another system to
> debian jessie and it appears we may have the same issue, sporadic
> segfauls, but we do not use the event_mpm but the default worker:
> 
>  [Tue May 05 13:02:19.929973 2015] [mpm_worker:notice] [pid
> 41460:tid 140289534666624] AH00292: Apache/2.4.10 (Debian)
> OpenSSL/1.0.1k mod_perl/2.0.9dev Perl/v5.20.2 configured --
> resuming normal operations

This can't really be the same bug, because the original issue was in 
mpm_event specific code.


>  root@support:/opt/otrs# grep -c Segmentation
> /var/log/apache2/error.log 36
> 
> find dump output attached. It seems the problem went away after
> switching to prefork worker.

I have trouble finding out which thread caused a crash in that dump 
and why. Do still have the core and see if gdb gave any interesting 
info when loading the core dump, before the bt command? And please 
open a new bug report for this one.

There is one thread in

Thread 1 (Thread 0x7f3ff08ce700 (LWP 41765)):
#0  0x00007f3ff9c7dac2 in modperl_interp_unselect () from 
/usr/lib/apache2/modules/mod_perl.so
No symbol table info available.
#1  0x00007f3ffd5819be in run_cleanups (cref=<optimized out>) at 
/tmp/buildd/apr-1.5.1/memory/unix/apr_pools.c:2352
        c = <optimized out>

This could mean that there are some thread safety issues in mod_perl. 
If this turns out to be the culprit, having a dump with an unstripped 
mod_perl would be nice. Though the bug report should go to mod_perl, 
then.

Cheers,
Stefan