Need for cleanup of idle winbindd child processes...

[Volker Lendecke]

On Thu, Mar 27, 2014 at 11:28:19AM -0700, Jeremy Allison wrote:
> On Thu, Mar 27, 2014 at 03:56:32PM +0100, David Disseldorp wrote:
> > Hi Ravindra,
> > 
> > On Thu, 27 Mar 2014 14:25:01 +0000, Ravindra Channabasappa wrote:
> > 
> > > If we enable trusted domain flag and there are 10 trusted domains, the number of processes increase (10x10) under load. Once started these processes do not exit. Also, it is observed in our setup that the memory consumed by these child processes keeps on increasing (for every load test run) without any sign of reducing. This has performance implications as we successively run load test on the winbindd daemon.
> > 
> > Do you have any "smbcontrol <pid> pool-usage" dumps, or valgrind logs
> > showing the memory leak?
> 
> Yep. Fixing the memory leak would be preferable :-).
> Finding it comes first..

Indeed.

One thing that might contribute to the memory shown in "top"
or equivalent tools is the tdb files winbind uses, most
notably the winbindd_cache.tdb and the gencache files. Those
are mmapp'ed by all winbind processes, thus shared. In my
experience it is very hard to correctly interpret the RSS
output in top in the light of large shared memory mappings.

Just for testing purposes, you might want to set

use mmap = no

which avoids memory mapping of tdbs at a significant
performance penaltly. But it will tell you whether you are
seeing a genuine winbind memory leak or whether the RSS
increase is due to large tdb files.

With best regards,

Volker Lendecke

-- 
SerNet GmbH, Bahnhofsallee 1b, 37081 Göttingen
phone: +49-551-370000-0, fax: +49-551-370000-9
AG Göttingen, HRB 2816, GF: Dr. Johannes Loxen
http://www.sernet.de, mailto:kontakt at sernet.de