I have four JBoss 3.2.5/Tomcat servers that are clustered to load balance traffic. The traffic comes from IIS/JK2 with a thread pool size of 200. I have been monitoring the status of the JK threads in JBoss using the MBean jboss.web:name=jk-8009,type=ThreadPool and it's attributes currentThreadCount and currentThreadsBusy.
The anomally is that one server acts normally, being that under load the currentThreadsBusy count jumps up, but quickly recovers when load drops off. This is optimal for monitoring.
The other three servers increase the currentThreadsBusy count under load, but then it never drops off. The result is that currentThreadsBusy becomes a high water mark rather than a proper monitor.
I have looked at the configurations and they are configured the same, as far as I can see. Can anyone explain this behaviour ?