[collectd] collectd dieing without error?

[Ulf Zimmermann]

> -----Original Message-----
> From: Florian Forster [mailto:octo at verplant.org]
> Sent: Thursday, May 17, 2007 01:28
> To: Ulf Zimmermann
> Cc: Sebastian Harl; collectd at verplant.org
> Subject: Re: [collectd] collectd dieing without error?
> 
> Hi Ulf,
> 
> > > On Wed, May 16, 2007 at 02:42:58PM -0700, Ulf Zimmermann wrote:
> > > > I currently run an own compile of 3.11.2 which dies without
error,
> > > > no message in syslog, even when recompiled with debug option.
This
> > > > only happens on the RedHat EL4 U4 machines I rolled it out on.
> 
> this happened to me, too. The coredump blames the processes plugin,
but
> it didn't have debugging symbols, so take this with more than just one
> grain of salt ;)
> 
> > No, it runs for a few days and then just is gone. No message in
syslog
> > about seg fault.
> 
> I don't think that segfaults get logged to syslog. We might be able to
> catch SIGSEGV and work out a stack backtrace.. I've just found that
the
> GNU libc provides a `backtrace' function in <execinfo.h> (but no
> manpage, *sigh*). That might be a good feature to implement.
> 
> > It seems often to happen when load is a big higher but not always.
> 
> I had the impression that it happens after the daemon has been running
> about 7 days. If this holds true, my installation with debugging
symbols
> should die in 2-3 days..

I have several instances running where it is longer then that:

app36 ulf /home/ulf > ps -aef | grep collect
root     10326     1  0 May05 ?        00:20:14 /usr/sbin/collectd

A number of servers were upgraded May 04/05, only 1 of them so far had
it die.

> 
> > Haven't found anywhere, console, syslog, coredump file on the
systems
> > it has happened, be it normal compile or debug enabled.
> 
> Under most distributions the creation of a coredump is disabled by
> default. You can enable it for your current shell (bash, other shells
> have similar commands) and programms started by it using
>  $ ulimit -c unlimited
> The coredump should be in the same directory as the .rrd-files.
> 
> Regards,
> -octo
> --
> Florian octo Forster
> Hacker in training
> GnuPG: 0x91523C3D
> http://verplant.org/