[Samba] Can one set limits on new core dump?
roamdad at sonic.net
Mon May 15 11:40:04 GMT 2006
James Peach wrote:
> On Sat, 13 May 2006 12:16 am, Gerald (Jerry) Carter wrote:
>> This was your change right ?
> Yup. It's deliberately not configurable so that we can always get
> *something* that might help with fault diagnosis.
Is there a chance for some kind of compromise?
winbindd cranked out hundreds of core dumps in less time than
it took to get a cup of coffee.
My vmware machines all died for lack of temporary file space.
Ultimately, it required a reboot to get back to normal
because a lot of daemons require var space.
If it's repeatable, the common process is to re-enable core
dumps and run a monitored test.
Barring a compromise, I'll have to investigate and probably
recommend hard limits be inherited in the startup files.
Otherwise, run the risk of having samba take down the entire
machine for the benefit of the developers on a Murphey.
The way I've done it for 30 years is limit core dumps for
normal day to day, re-enable it during problem determination.
I long for the days long, long ago and far, far away
where there was a presumption of intelligence. Maybe it's
better this way and I need to just fade away. I don't know.
>> Doug, I'm more interested in why winbindd is seg
>> faulting in the SAMBA_3_0 tree. Can you give me more
> Agreed. Please let's get a backtrace at least:
> gdb `which winbindd` /path/to/core/file
> (gdb) where
It was an old xos idmap_ad ad.so in samba/lib which I deleted.
Still, why did samba load it instead of the internal ad module?
Still interested? If so, I have to find a copy on
an old DVD backup disc.
>> Doug VanLeuven wrote:
>>> Sorry Jeff, been there, done that, if you'd read the whole post.
>>> Jeff Saxton wrote:
>>>> man ulimit
>>>> hint: ulimit -c
> This probably won't work because in fault.c we explicitly set the core
> size to 16MiM (IIRC).
>>>> Doug VanLeuven wrote:
>>>>> Hi all,
>>>>> Is there anyway to limit the new core dumping panics?
>>>>> Can't find anything on it. (If I'd only looked in that ...)
>>>>> Was my mistake, but winbindd filled up an entire volume
>>>>> and froze out every process writing to that drive.
> You should only get 1 core file per daemon unless you have some
> system-specific core file naming facility enabled. If winbind is dumping
> core often it should always be in LOGBASE/cores/winbindd/core.
>>>>> I started it from a shell and my soft limit is
>>>>> already zero. (ulimit -S -c 0)
>>>>> FC4 2.6.16-1.2069 smp, gcc 4.0.2-8
>>>>> samba 3.0.23pre2-SVN-build-15162
>>>>> Regards, Doug
More information about the samba