Re: RE24 (2.4.15) slapd consuming large amounts of CPU

[Quanah Gibson-Mount <quanah@zimbra.com>]

--On Wednesday, March 04, 2009 1:55 PM -0500 John Morrissey <jwm@horde.net> 
wrote:

> I upgraded one of our consumers to 2.4.15 recently, from RE24 as of
> 20090113. Its configuration stayed the same, save for bumping the BDB
> cache size from 256MB to 2GB.
>
> After a few hours of uptime, slapd suddenly begins consuming enormous
> amounts of CPU (a three- to four-fold increase over less than a few
> minutes, and is sustained until slapd is restarted).
>
> A backtrace (below) shows many threads waiting on a mutex in
> bdb_cache_*().
>
> This is with BDB 4.7 (with all patches applied) on Debian lenny.
> 'db_stat -c' output is below (and looks sane), in case that's useful.
> Thinking thread contention, I bumped the worker thread count (slapd.conf
> threads directive) down from 32 to 8, which has made no difference.

Interseting.  I've seen this sort of CPU usage issue with OpenLDAP 
2.3/BDB4.2.52 on Ubuntu 8.  Do you know how BDB 4.7 was built?  I.e., 
specifically, did it have the:

--enable-posixmutexes --with-mutex=POSIX/pthreads

options enabled?

Also, are you using a shared memory key?  I've found that has a significant 
(positive) impact with BDB 4.7 on linux boxes.

--Quanah

--

Quanah Gibson-Mount
Principal Software Engineer
Zimbra, Inc
--------------------
Zimbra ::  the leader in open source messaging and collaboration