[WICKET-6522] ThreadLocal leak in PageStoreManager

• Type: Bug
• Status: Resolved
• Priority: Major
• Resolution: Fixed
• Affects Version/s: 7.9.0
• Fix Version/s: 7.10.0, 8.0.0-M9
• Component/s: wicket
• Environment: Wildfly 10 on CentOS 7 Linux.
• Assignee: Emond Papegaaij
• Reporter: Klaasjan Brand
• Created: 22/Jan/18 10:50
• Updated: 06/Feb/18 12:55
• Resolved: 30/Jan/18 13:27

Description

One of our heavy-used Wicket production applications has developed 'interesting' behaviour since upgrading to 7.9.0: after a while the system becomes slower and seems to need more CPU time to handle the same amount of requests. Threaddumps exposed that a lot of CPU time was consumed by java.lang.ThreadLocal$ThreadLocalMap.getEntryAfterMiss, pointing to a ThreadLocal leak.

After making a heapdump (20g) of one of our servers we found lots of ThreadLocalMap$Entry objects referenced by org.apache.wicket.page.PageStoreManager$SessionEntry$1 containing the value Boolean.FALSE.

Seems that boolean is new behaviour introduced by the fix for WICKET-6465 and is causing complications on servers like Wildfly we use which don't recycle threads.

 

 

 

Issue Links

• links to: GitHub Pull Request #259