Symptoms

Every 30 minutes, CPU spikes and/or performance suffers. The CacheStatisticsEventPublisher class is determined to be at fault (see Diagnosis below). 

Diagnosis

You can use inbuilt VM or server monitoring tools, or a third party tool like New Relic, to pinpoint problematic requests/threads. You could also analyse a series of external thread dumps to determine long running threads, or to look for patterns of threads running during slow downs.

Workaround

The CacheStatisticsEventPublisher is currently not used, so it is safe to disable it from running. To do this, enter this JVM parameter:

-Dcom.atlassian.confluence.CacheStatisticsAnalyticsJob.repeatPeriodMinutes=-1

Resolution

This issue will be resolved in Confluence 5.6. Please upgrade to that version when it is available.