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

Re: slapd memory usage

On Mon, 30 May 2005, Petri Laakso wrote:


I'm a bit worry about that slapd will use all of the memory
and kernel kills it violently (=> database corruption etc)
It seems slapd is reserving 1.6G of memory.
ldif dump of whole database is 84 megabytes of disk.
At this time there is 340 TCP connections to server.
Is this normal for slapd-2.1.30?
If it is we just have to buy more RAM :)

I found OL 2.1.30 slapd running on RHEL 3 to have a memory leak that was exercised by searches against the directory. We had to restart slapd every couple of days to prevent it from eventually consuming all memory on the host. The leak is not present (but a lot of wonderful features are) since OL 2.2.23 slapd, so you might try upgrading to the latest stable version.

For backend we use bdb without DB_CONFIG (for defaults)

You *definitely* need a DB_CONFIG tuned to your particular circumstances. Don't rely on the default settings, which are inadequate for almost every reasonable OpenLDAP use case. There is a wealth of good information about proper DB_CONFIG tuning in the OpenLDAP FAQ and in the list archives. Be sure to follow the links from the FAQ articles to the SleepyCat BDB docs and read those too.

slapd.conf ... threads 25

I recommend that you leave your threads setting at the default until you've properly tuned a DB_CONFIG file, then experiment carefully (if needed) to get the best setting.

checkpoint 128 15
sizelimit 1000000
cachesize 90000

-- Kirk Turner-Rustin | Programmer/Analyst Ohio Wesleyan University | Libraries and Information Services http://www.owu.edu | http://lis.owu.edu