Running CMS with many concurrent users leaves suspicious amount of sockets in FIN_WAIT2 state

When the CMS is used by many concurrent users (about 5+), after a while quite a number of sockets get into state FIN_WAIT2 (about 20+). This may indicate connections that are not properly closed.

These states disappear automatically over time due to timeouts, but may cause problems if the sockets and/or associated Tomcat threads are not released. With many users we may either run out of sockets or threads.

CMS7-6623 is related; in there we aligned the CXF versions used by CMS and site, because the CMS used a version (2.4.0) in which sockets were not properly closed (see https://issues.apache.org/jira/browse/CXF-3505), leading to FIN_WAIT2 states that kept Tomcat threads occupied, leading to a massive slowdown of the CMS once a certain threshold of FIN_WAIT2 states had been reached. With aligned CXF versions we still see many FIN_WAIT2 states appearing (although a lot less then before), so there may still be some problem left to tackle.