[horde] postresql sessionhandler problem

[COMBES Julien - CETE Lyon/DI/ET/PAMELA]

Hello,

We have a horde server in production with horde 3.1.3, imp, turba ingo, 
mnemo, nag, mimp in the last stable version and kronolith 2.1.3. Horde 
is running on a debian Sarge server with apache 2.0.54 and php 4.3.10; 
and for the database we are using a postgreSQL 7.4.7 on a separated server

As the size of the horde's farm servers is going to increase, the 
sessionshandler is configure on 'pgsql' so that it can be shared between 
the servers with active/active high-availability . At this  time, we 
have only one horde server.

Sometimes, maybe under connexions peak, horde become more and more slow 
to finally become inaccessible for some users and then for all users. 
Restarting apache doesn't help. The only thing which help is to restart 
postgresql.

During the problem :
  - it seems that there is many lock on the horde_sessionhandler database.
  - all the tables on the postgreSQL's server (of horde and others 
applications which use this server) are working fine and respond 
correctly except the horde_sessionhandler table.
  - the number of apache's processes increase quickly.

If, we come back to "$conf['sessionhandler']['type'] = 'none';", the 
server works correctly.

The MaxClient of apache and postgreSQL are equal.

Now, we have returned to php sessionhandler to make it work but it's 
freeze the deployement of our other servers.

We haven't found a way to correct that in horde's documentations or on 
horde's mailing list archive. Is someone has any idea which could help 
us to correct this problem ?
Or, Is there a better/other solution to share sessionhandler between 
server ?

Regards,
Julien