Jean-Francois,
Thank you very much for responding to our inquiry so quickly and we apologize for taking a while to respond.
We finally found the issue.
CLOSE_WAIT was a symptom of GF hanging due to being low on memory (near out of memory). Memory was being consumed due to the following reasons:
1) certain type user request caches (relatively) larger amount of data in user session.
2) session expiration was configured to one day.
We didn't observe this earlier because a) our stress test was sending type of requests which didn't cache much data in session, and b) we found out recently that this application had a regularly scheduled re-start while it was deployed on WLS.
As a short-term solution, by adjusting initial/max heap size and reducing the session expiration, we hope to address the issue. Code refactoring is in order but that will be our longer-term solution.
Obviously, the issue has nothing to do with GF but we appreciate your comments/feedback. We hope we didn't take up your time too much...
Best regards.
[Message sent by forum member 'ur_afroinu' (ur_afroinu)]
http://forums.java.net/jive/thread.jspa?messageID=342178