[jboss-jira] [JBoss JIRA] Created: (JBMESSAGING-1547) Potential Dead Lock in MessageSucker.stop()

[Howard Gao (JIRA)]

Potential Dead Lock in MessageSucker.stop()
-------------------------------------------

                 Key: JBMESSAGING-1547
                 URL: https://jira.jboss.org/jira/browse/JBMESSAGING-1547
             Project: JBoss Messaging
          Issue Type: Bug
          Components: Messaging Core Distributed Support
    Affects Versions: 1.4.2.GA.SP1, 1.4.0.SP3.CP07
            Reporter: Howard Gao
            Assignee: Howard Gao




The stop() is synchronized on MessageSucker, in this method, it tries to unregister itself from the localQueue, as 

...
localQueue.unregisterSucker(this)
...

the unregisterSucker(MessageSucker) requires synchronization on 'lock' object in MessagingQueue.


On the other hand, the MessagingQueue.deliverInternal() method is synchronized on 'lock' object and inside the method the informSuckers() is called. If there are some suckers, the sucker.setConsuming() will be called upon the sucker. The setConsuming() method requires locking on the messaging sucker.

So dead lock may happen when a message sucker is being stopped due to remote queue leaving (the hosting node left the cluster) while the corresponding localQueue is delivering messages and trying to push messages to its suckers. Take two node cluster as an example,

1. A Node with a distributed queue (say remoteQueue) has left the cluster, causing the message sucker (remoteQueue-sucker) on the other node to be stopped and unregistered from localQueue.
2. remoteQueue-sucker.stop() acquires the lock on remotQueue-sucker, and then goes on till localQueue.unregisterSucker(this), at which point it tries to acquire the localQueue.lock object.
3. At that time the calling thread of localQueue.deliverInternal() has already held the localQueue.lock object, which prevents thread of step2 from getting the lock. Then it proceeds to informSuckers() where the remoteQueue-sucker.setConsuming() is trying to get the lock on remoteQueue-sucker. But this lock has been held by thread of step 2. So dead lock.


-- 
This message is automatically generated by JIRA.
-
If you think it was sent incorrectly contact one of the administrators: https://jira.jboss.org/jira/secure/Administrators.jspa
-
For more information on JIRA, see: http://www.atlassian.com/software/jira