Message7469

I recently upgraded from jython 2.5.2 to jython 2.5.3 standalone jars. 
After this upgrade, we found that threading is broken in 2.5.3, when a 
process tries to join an existing thread.

Logs from 2.5.2:

Oct  1 15:18:27 zre-ldap002 zmconfigd[28096]: Found 3 threads
Oct  1 15:18:27 zre-ldap002 zmconfigd[28096]: Active Thread listener found
Oct  1 15:18:27 zre-ldap002 zmconfigd[28096]: Active Thread Thread found
Oct  1 15:18:27 zre-ldap002 zmconfigd[28096]: Active Thread MainThread found
Oct  1 15:18:27 zre-ldap002 zmconfigd[28096]: Fetching All configs

Logs from 2.5.3:

Oct  1 15:04:06 zre-ldap002 zmconfigd[23938]: Found 3 threads
Oct  1 15:04:06 zre-ldap002 zmconfigd[23938]: Active Thread listener found
Oct  1 15:04:06 zre-ldap002 zmconfigd[23938]: Active Thread Thread-8329 
found
Oct  1 15:04:06 zre-ldap002 zmconfigd[23938]: Attempting to join() 
Thread-8329
Oct  1 15:04:11 zre-ldap002 zmconfigd[23938]: join() Thread-8329 FAILED
Oct  1 15:04:11 zre-ldap002 zmconfigd[23938]: Hung threads detected (3 
total), exiting


This is infinitely repeatable.


Our loop has:

while True:

        Log.logMsg (4, "Found %d threads" % threading.activeCount())

        for th in threading.enumerate():
                Log.logMsg(4, "Active Thread %s found" % th.getName())
                if (th.getName() != "listener" and th.getName() != "MainThread" and th.getName() != "Thread" and th.getName != "SIGUSR2 handler"):
                        Log.logMsg(4, "Attempting to join() %s" % th.getName())
                        th.join(5);
                        if (th.isAlive()):
                                Log.logMsg(1, "join() %s FAILED" % th.getName())
                                Log.logMsg(1, "Hung threads detected (%d total), exiting" % threading.activeCount());
                                sys.exit(1)

        t1 = time.clock()

        try:
                # read all the configs
                myState.getAllConfigs(myConfig)
        except Exception, e:
                [Log.logMsg(1,t) for t in traceback.format_tb(sys.exc_info()[2])]
                if myState.forced:
                        Log.logMsg(0, "Key lookup failed.")
                Log.logMsg(1, "Sleeping...Key lookup failed (%s)" % (e,))
                time.sleep(60)
                continue

        try:
                # read zmconfigd config
                myState.getMtaConfig(myConfig.configFile)

                # watchdog restarts apps if they are not running
                watchdog()

                # check for config changes
                myState.compareKeys()

                Log.logMsg (5, "LOCK myState.lAction requested")
                myState.lAction.acquire()
                Log.logMsg (5, "LOCK myState.lAction acquired")
                myState.compileActions()
                myState.requestedconfig = {}
                myState.doConfigRewrites()
                myState.lAction.notifyAll()
                myState.lAction.release()
                Log.logMsg (5, "LOCK myState.lAction released")

                # executes rewrites/postconf/restarts
                if myConfig.restartconfig:
                        myState.doRestarts()
        except Exception, e:
                [Log.logMsg(1,t) for t in traceback.format_tb(sys.exc_info()[2])]
                if myState.forced and myState.forced < 100:
                        Log.logMsg(0, "Configuration inconsistency detected (%s)" % (e,))
                Log.logMsg(1, "Sleeping...Configuration inconsistency detected (%s)" % (e,));
                time.sleep(60)
                continue

        if myState.forced:
                break

        # start the listener after we have the lock, or an early request can cause problems
        # start the listener after the rewrites, so the start script doesn't return before they're complete
        if myState.firstRun and not myState.forced:
                requestListener()

        Log.logMsg (5, "LOCK myState.lAction released")
        myState.firstRun = False
        lt = time.clock()-t1
        Log.logMsg(4, "Loop completed in %.2f seconds" % (lt,));
        Log.logMsg(4, "Sleeping for %d." % (myConfig.interval,));

        # Jython won't wake up from time.sleep() when a signal is received and caught.
        # Uncaught signals seem to cause the JVM to exit
        myState.sleepTimer = myConfig.interval
        Log.logMsg (5, "Sleeping for %d" % myState.sleepTimer)
        while myState.sleepTimer > 0:
                time.sleep(sleepinterval)
                myState.sleepTimer -= sleepinterval
        Log.logMsg (5, "Waking up")

dt = time.clock()-t0
Log.logMsg(5, "%s completed in %.2f seconds" % (myConfig.progname,dt));
sys.exit(0)