Previous versions of JBoss EAP 6 contained a bug that could lead to an OutOfMemoryException in distributed web sessions. The exception was encountered if a web session expired without the lock objects created by the session manager being released or destroyed. As web sessions continued to expire, the residual lock objects accumulated in memory. Eventually, this would lead to an OutOfMemoryException. The only recourse was to redeploy the web application.
In this release of the product the lock objects are properly released and the OutOfMemory no longer presents
Stale sessions are not getting removed from memory.
SharedLocalYieldingLockManager.LocalLock hogs an increasing amount of heap space.
original case data https://bugzilla.redhat.com/show_bug.cgi?id=1030681
Possible source is https://issues.jboss.org/browse/WFLY-898, which was closed when https://issues.jboss.org/browse/WFLY-406 was submitted
Looking to port a fix for EAP 6 , but the fix for WFLY-406 is a complete rewrite and involves 321 files.
Is there a smaller scope fix that could be ported or a possible workaround?
Might this just be a product of the number of sessions in your application?
The SharedLocalYieldingClusterLockManager hold a LocalLock for every session "owned" by the local node.
However, if a session passivates, the local lock for that session stays in memory. Thus is it possible to exhaust the heap via the number of session even when using passivation. It might be a good idea to release/destroy the local lock on passivation too.
The LocalLock is meant to be destroyed when a session invalidates, expires, or when some other node takes the lock. While it's certainly possible that one of these cases is not correctly destroying the LocalLock - I want to rule out the above cause first.
This is definitely a bug in EAP. It can be easily reproduced, and the leak occurs every time a session times out. (It does not occur if session.invalidate is called).
fix checked in for 6.2.0
Verified for 6.3.0.DR0.
I'm working with the Linux performance teams on their efforts to do proactive performance analysis based on collection of sosreports. One of the things they hi-lighted is that our nodes are using a large amount of memory in Slabcache, in particular negative dentry cache entries.
We wrote a systemtap script and we've found that JBOSS is responsible for all of the negative dentry items, could this be related to this bug as well? What we see is JBOSS attempting to look for thousands of files that do not exist. My thinking is that perhaps it's looking for files associated with expired sessions
[root@gss-webjava01 ~]# stap dentry.stp
> My thinking is that perhaps it's looking for files associated with expired sessions
That is the directory where the sessions are passivated.
So expired session IDs is the most likely cause.
That wouldn't be related to this BZ though.
That would just be the normal expected behavior for an expired session ID.
Paul, can you provide some details in the Doc Text field as to what was causing this leak, and how you fixed it, so we can get it into the 6.3.0 Release Notes.
Added release note text.
Refactored Doc Text into prose form.