[Date Prev][Date Next] [Chronological] [Thread] [Top]

Re: Is it normal?

On Thu, 2 Dec 2004, Quanah Gibson-Mount wrote:

--On Thursday, December 02, 2004 11:31 AM +0100 Tony Earnshaw <tonye@billy.demon.nl> wrote:

ons, 01.12.2004 kl. 20.43 skrev Quanah Gibson-Mount:

> I am using Openldap 2.2.13 as LDAP backend for Samba.
> Total number of records in the database is about 150, but slapd for
> some reason consumes 140Megs of RAM and this number grows (last time I
> checked few days ago was 68). I was looking everywhere but never found
> what could actually cause this, so if this question is old, just let me
> know where I can find more information about this issue.
> Thanks in advance.

There were some memory and cache leaks fixed in the 2.2.19 release that
may  solve this problem.

People *must* give OS and distro for reference, plus supporting software.

 After months of uptime on RHAS3, self-compiled 2.2.17/Cyrus SASL
2.1.18/BDB4.2.52, openssl 0.9.7d (up to 80 concurrent users, total in
that time ~1.000 users) and after 5 days of uptime RHAS3 self-compiled
2.2.18/Cyrus SASL 2.1.20/BDB 4.2.52/Openssl 0.9.7d (only 2-3 concurrent
users) slapd doesn't even show up in top. ps auxwww shows 8368 RSS,
194024 VSZ on this (5 days up) rig.

This could have far more to do with the supporting system than with the
OL version alone.

Perhaps, yet both Howard & I were able to trigger the memory & cache leaks on both linux & solaris. However, it was not easily triggered, and didn't trigger consistently. It was something that was quite difficult to track down because of that. Out of my 9 production servers @ Stanford, sometimes none would exhibit the behavior for weeks at a time. Sometimes 3 would consume all the available memory available on the same day, and have to be restarted. There was no consistent pattern to it. Anyone running OL prior to 2.2.18 without the memory & cache leak patches is running the risk of having slapd chew up their resources.

I am seeing the same issue on Solaris 9 (OL 2.2.1[78]) with a very small directory. It is very unpredictable as to when it consumes all the resources.

I'll note that OL 2.2.18 was released October 23rd, so it would also be hard to have it up and running for "months". ;) The same issue was present in OL 2.2.17, and seems related to issues I had as far back as 2.2.15.


-- Quanah Gibson-Mount Principal Software Developer ITSS/Shared Services Stanford University GnuPG Public Key: http://www.stanford.edu/~quanah/pgp.html

-- Igor