Atomikos Forum

We have a stress tool we use which generates a great deal of sql statements and executes them (each thread does this).

I have modified this tool so that there is now an XA transaction object being called (using AtomikosDataSourceBean), along with XA connections, and we use these connections instead of the previous regular JDBC ones.  The commit at the end becomes a utx commit.

When I do this minor change, I see a memory leak.  A java profiler tells me massive amounts of HashMap$Entry are increasing.

If I merely create XA connections and never use them, I never see the leak.  Only when I use these conncetions to create the preparedStatement objects (up to 500 per commit), there is a memory leak.  This persists across threads.

Any thoughts on what might be causing this?  Could this be an issue with Atomikos?

We've found out that the memory is being taken by com.atomikos.persistence.imp.FileLogStream  .  It seems to have many, many children "CoordinatorImp" classes, all of this is what is increasing the memory.

Is there anything we can do about this ?

Try setting a lower value for com.atomikos.icatch.checkpoint_interval in jta.properties

See http://www.atomikos.com/Documentation/JtaProperties

HTH