It seems that OpenFire statistics, even for a clear server, can cause real problems to an enterprise database server.
How to reproduce the problem: integrate OpenFire with LDAP so it will get all the users and groups from the Active Directory. In my case this translated to about 15000 users and 20000 groups.
It seems that openfire is keeping the entire rrd database in a field in the database. In my case I found that the lengh of the field reached 180 M.
Our database is PostgreSQL 9.1, but I observered others reporting the same kind of problem for MySQL and I guess this will be reproductible with any database.
Due to how SQL works, the the database server will try to encode the response from binary to text, which will need a LOT of memory.
I checked the logs and discovered that PostgreSQL failed to allocate a continous block of memory of about 512MB only for these. The server has plenty of memory free, but finding a continous block of 512MB is not easy, even if you run your machines at 50% memory load
Not related to this, but I was unpleasantly surprised to discover that Jira bug tracker from ignite is read-only. Do I have to “bribe” someone to get the change of rising a bug?
Here are few details http://dba.stackexchange.com/questions/46675/postgresql-out-of-memory-while-tryi ng-to-a-request-of-500mb