More turd wrangling.  Be sure to kill off child processes when the parent
receives a signal.  This should prevent the slow, but steady buildup of
stuck ssh processes on boss resulting from the monitor daemon trying to
bring back nodes.