[sr-dev] SER crash : Segmentation fault

[Andrei Pelinescu-Onciul]

On Oct 09, 2009 at 20:33, inge <inge at legos.fr> wrote:
> Hello Andrei,
> 
> We are now running to 0.9.9+cvs20090925 since 10 days.
> 
> Today SER crash/stop without coredump. Do you know if we need to
> configure something for enable this option ?

No, coredump it's enabled by default since 0.9.3.
However note that if you don't start ser as root, it cannot enable
core-dumping (and you have to do it b by hand before starting ser).
Note also that even if started as root, if it's supposed to change its
uid (e.g. started with -u <some_user> or with uid in the .cfg) it won't
be able to dump core on any modern linux kernel (in this case you would
need to set /proc/sys/fs/suid_dumpable to 1 or remove the -u from ser
 command line).

You can check if it dumps core, by sending SIGABRT to one of the ser
processes (e.g kill -SIGABRT <pid_of_ser>).
> 
> With the previous crash, we got in /var/log/messages something like a
> CHILD which firstly crashed and then all the processes are followed. But
> here SER stop like "service ser stop" by printing only in the log
> "INFO : signal 15 received..."
> 
> Do you have any idea ?

Are you sure somebody hasn't stopped it?
If it crashed and couldn't dump core, there should be a message logged
(something like ... core was not generated...). Also you should see
messages about the signal that caused the first child process to
terminate and if it's really a problem it will be different from 15.

Another possibility is that the kernel killed some ser processes due to
 low memory (check dmesg for OOM).


Andrei