Thank you for your answer. We actually could not wait for a solution or workaround anymore, due to the project planning and the time we spent trying to find a solution with OpenLDAP. So, we decided to test replacing OpenLDAP by OpenDJ, and the tests give pretty good results: we don't have the problem anymore.

Howard Chu a écrit :

cgrosjean@janua.fr wrote:

...

Full_Name: Cyril Grosjean Version: 2.4.26 OS: Red Hat Enterprise Linux Server release 5.4 (Tikanga) URL: ftp://ftp.openldap.org/incoming/ Submission from: (NULL) (82.66.198.206)

My OpenLDAP server freezes, after a few requests, and I can't understand why. I have a few LDAP clients (ldapsearch, a custom Java app. and ApacheDirectoryStudio), running from different systems, either locally on the OpenLDAP server, or on another OpenLDAP server, or on a remote workstation, and none manages to get an answer from OpenLDAP. The connection is established but each client gets stuck waiting for any result.

We can reproduce the freeze, after a few requests coming from the custom Java application. The problem doesn't occur if we don't use that application.

The last time, it occured only a few minutes after the OpenLDAP daemon was restarted, under very low load. I use OpenLDAP 2.4.26, over Red Hat Enterprise Linux Server release 5.4 (Tikanga), Kernel version 2.6.18-164.el5, amd64 architecture over VMWare. I also use BDB 4.8.30, and OpenLDAP's been compiled with the following options:

./configure --prefix=/opt/openldap-2.4.26 --enable-dynamic=yes --enable-dynlist=yes --enable-memberof=yes --enable-ppolicy=yes --enable-accesslog=yes --enable-sssvlv=yes

I noticed only 60 to 70 established connections on port 389 (according to netstat) when the freeze occurs.

The OpenLDAP server is set in syncrepl mirror mode replication, and the other OpenLDAP server (only used as a hot stand-by) works fine (but doesn't receive any load). However, one can reproduce the problem with replication disabled.

Looks like you have a thread blocked trying to modify a record, while all the other writer threads are blocked waiting for that writer to complete. The output of db_stat -CA would help to indicate why the writer is blocked.

It also looks like you've configured your replication with mirrormode and delta-syncrepl. This combination is not yet supported in OpenLDAP, certainly not in 2.4.26. It may be supported in 2.4.27, depending on how our ongoing testing turns out.