I'm working on trying to find a "reproducable" test case. At this
point, we've seen the problem multiple times under load conditions.
It appears to be much more reproducable if our server is running
w/ non-blocking I/O. [we've only seen the problem when we are
running the server VM on solaris].

I'm filing this now in the hopes that the pstack alone might provide
enough info to actually track the problem down

I'm attaching 2 sets of pstack traces which show the problem :

  set 1: w/ blocking I/O
          pstack -> pstack taken when the problem first showed up
          pstack2-> pstack taken when all connections had exited
                     In this case, the "client" threads used
                     by the broker (our server) were no longer process
                     read or write events
          pstack3 -> pstack taken several seconds after pstack2
                    (shows the threads that are spinning)

         Host info:
             java -fullversion (for client & broker)
                java full version "1.4.0-b92"
             hostname: sidewinder
             showrev -p : attached as showrev-1
             sysinfo : attached as sysinfo_1

   set 2: older pstack trace taken w/ non-blocking IO (was found
          running a different test case)

         Stack trace of the spinning thread (I no longer
           have the full pstack traces)

               pstack.bad_thread

         Host info:
             java -fullversion (for client & broker)
                java full version "1.4.0-b92"
             hostname: iclab2
             showrev -p : attached as showrev-2
             sysinfo : attached as sysinfo_2

I'll attach a test case as soon as I can find one