(1:00:30 PM) mazz: ips: I can't remember if there is "pull the pin" suicide code in that shutdown hook
(1:00:48 PM) mazz: by that I mean, it tries to kindly wait, but if the threads don't die, we eventually call System.exit()
(1:00:56 PM) mazz: is that the 100s timeout you are talking about?
(1:01:18 PM) mazz:             // set our timebomb to ensure the agent dies
(1:01:18 PM) mazz:             spawnKillThread(1000L * 60 * 5);
(1:01:41 PM) mazz:     /**
(1:01:41 PM) mazz:      * If something goes wrong and the VM does not die on its own when we think it should,
(1:01:41 PM) mazz:      * the thread created by this method will explicitly kill the VM.
(1:01:41 PM) mazz:      * Calling this method sets a timebomb that will kill the VM after a timer runs out.
(1:01:41 PM) mazz:      * There is no way to stop this timebomb short of performing 007-type heroics.
(1:01:42 PM) mazz:      * 
(1:01:42 PM) mazz:      * @param doomsday the number of milliseconds that the VM has left to live; once this
(1:01:43 PM) mazz:      *        time expires, a kill thread will execute System.exit(0).
(1:01:43 PM) mazz:      */
(1:01:44 PM) mazz:     public void spawnKillThread(final long doomsday) {
(1:01:50 PM) mazz: so after 5 minutes, the agent should eventually die
(1:02:28 PM) mazz: during shutdown, we should see a daemon thread named "RHQ Agent Kill Thread" in the thread dump
(1:03:00 PM) mazz: so it seems to me that no more than 5m should elapse before the agent VM process dies

spawnKillThread(1000L * 60 * 5) should be configurable if people want to change this behaviour, but keep its current default.