My xmx is 11g right now, I need it this way or production environment will end up doing full gc. The heapdump is taken after a GC.
I think the problem comes from eclipselink. We have some caches. Those caches are storing some entities which are in fact storing lazy relations. In case of a batchfetch (or joinfetch), eclipselink needs to store the original query that was performed to get the result. I guess that the problem comes from this, as I can see a lot of QueryBasedValueHolder objects.... -- Sent from: http://tomee-openejb.979440.n4.nabble.com/TomEE-Users-f979441.html
