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

slapd locked up

Hash: SHA1


I have a problem with openldap (of course I searched google, the faqs an
the bug-database).

I am using OpenLdap 2.0.11 as a NIS-Replacement serving user-accounts of
1500 users to 100 workstations. My workstations use libnss-ldap for
accessing these information.

The database is ldbm and app. 5 MB in size.

I am using one primary and two replica-Servers.

Since recently all three servers show a very ugly behaviour.

After some time of operation the load on one server rises and slapd
consumes >90% CPU-Power.

My first guess was a corrupted database. The server runs since almost
two years without interruption. So I dumped the database, erased all
files and reloaded everything. (slapcat, rm, slapadd) -> Problem not

Next I increased cachesize to 2000 and dbcachesize to 5000000 (5mb).
Except that slapd now uses 60 MB RAM (;-) the Problem still exists.

There is still enough free RAM and abundant unused file-handles.

Using logging and stracing I found out, that there are lots of writes to
the clients which seem to "block". I know slapd uses nonblocking writes
so there is error EAGAIN.

It seems that slapd tries to send the data for ever so no further requests
are handeled. Shouldn't ldap stop try sending data after some time and
work on other requests?

What can I do now?

Would an update to 2.1.x help? Is there a config switch which allows a
little bit more parallelism in the work? Or can I tell slapd to stop
sending data after some time? Is there a forking or multithreading slapd

Thank you very much in advance

Chris, hating having software-trouble on Fridays ;-)
- -- 
Chris Huebsch    www.huebsch-gemacht.de | TU Chemmnitz, Informatik, RNVS
GPG-Encrypted mail welcome! ID:7F2B4DBA |   Str. d. Nationen 62, B204
   Chemnitzer Linux-Tag 2004, 6.-7.3.   |       D-09107 Chemnitz 
  http://www.tu-chemnitz.de/linux/tag/  |  +49 371 531-1377, Fax -1803
Version: GnuPG v1.2.1 (GNU/Linux)