We're having a problem where the load-balancer plugin opens up connections to a cluster instance that is non-responsive due to an out of memory problem. The plugin appears to detect that it is not healthy but does not close the connections. Netstat shows them in a FIN_WAIT_2 state.
The web server (7.0 update 8) then hangs once the number of open connections in FIN_WAIT_2 is large enough.
Anyone know of a way to prevent a single cluster instance from taking out the entire web server? Yes, we do have support and have opened a case with Oracle but they are so far unable to provide a solution.
[Message sent by forum member 'davideagen']
http://forums.java.net/jive/thread.jspa?messageID=471413