Yes ;-)............


Upgraded locally and initial tests are good, the not-shared mode is not bombing out like previously and its collecting as expected which is good.

Will upgrade the rest of our suite with the latest libs and will keep you posted on it!!

Thanks Sanne,
LL

thanks for all info, and yes please keep us posted.

Ok spoke to soon.......

In not-shared mode it runs fine but with shared mode it still crashes.

Both modes where run with the adapted SlaveDirectory which i posted earlier, i.e if the marker changes, close the old and open the new one, repeat the cycle everytime marker changes.

not-shared mode has frequent GC cycles but it frees up the memory and keeps going

shared-mode- everytime i change the marker file and it closes the old and opens the new, the VM size keeps growing until finally it cant take no more and crashes.

ok now I think I understand what's going on. Please use the non-shared mode until we fix the issue; still I'm not sure that it's leaking, is it possible that it's just taking more memory than what you have available? I know we can improve on that, but I need the info to understand what's happening.

Would you be willing to try a smaller index in you same setup, and see if the memory keeps growing?

Opened an issue, please comment there:
http://opensource.atlassian.com/projects/hibernate/browse/HSEARCH-557