[jboss-jira] [JBoss JIRA] Created: (JBMESSAGING-1862) Enhancement - No duplicated messages when cluster is split up into to sub clusters

[Miroslav Novak (JIRA)]

Enhancement - No duplicated messages when cluster is split up into to sub clusters
----------------------------------------------------------------------------------

                 Key: JBMESSAGING-1862
                 URL: https://issues.jboss.org/browse/JBMESSAGING-1862
             Project: JBoss Messaging
          Issue Type: Enhancement
          Components: JMS Clustering
    Affects Versions: 1.4.0.SP3.CP12, 1.4.8.GA
            Reporter: Miroslav Novak
            Assignee: Miroslav Novak


For now the behaviour is following: 

Test configuration:
4 nodes EAP 4.3.8 in cluster, MySQL 5 for JBM, clustered queue deployed on all nodes.
Stand alone producer, receiver.

We use ConnectionFactory in following scenario for standalone producer and consumer (using auto_acknoledge mode) to be sure that client is connected to required node.
Test scenario:
1) Start all nodes, There is now cluster of four nodes A - B - C - D with clustered queue "testDistributedQueue"
2) Send 500 messages into each node
3) Check status of DB, you should see 500 messages for each node
4) Remove network connectivity between nodes A-C, A-D, B-C, B-D (so there are now connected only A-B,C-D)
5) Wait (5 min)
6) Disconnect all nodes from DB
7) Try read from A
8) Try read from B 
9) Look to database - each node have 500 messages in the queue
10) Reconnect all nodes to DB
11) Look to database you will see that fail-over occurs. Number of messages on each node - A: 1499 B:499 C:0 D:0 - two messages are lost - it's current behaviour
12) Read from A - 1998 messages is received and there are no messages in DB (it's ok)
13) Read from B - 0 messages is received (it's ok)
14) Read from C - 500 messages is received. Those are duplicated messages and the new regression.
15) Read from D - 500 messages is received. Those are duplicated messages and the new regression.

Also when A is reconnected to DB (step 10) then nodes C and D are removed from jbm_postoffice table. But those nodes still live.

As enhancement this should be fixed. 

An expected behaviour should be specified but those things should hold:
a) No duplicated messages should occur in step 14. and 15.  
b) No message loss after step 10.
c) Client should be able to read all the messages from clustered queue on any node.

Thanks,

Mirek

--
This message is automatically generated by JIRA.
For more information on JIRA, see: http://www.atlassian.com/software/jira