Hi Howard,

I tried following test scenario:

1. Start cluster of 4 nodes - A,B,C,D
2. Start 4 publishers and 4 subscribers on one clustered topic - clients use ClusteredConnectionFactory with transaction acknowledge session.
3. Kill server A
4. Kill server B
5. Kill server C
6. Start A
7. Kill D
8. Stop producers

But during killing EAP servers sometimes happens that a message is delivered twice. In this case "subscriber1" got one message twice - log from perf06:

12:21:47,956 INFO [FailoverCommandCenter] JBoss Messaging server failure detected - waiting for failover to complete...
12:21:49,937 INFO [FailoverCommandCenter] JBoss Messaging failover complete
2011/06/09 12:21:49:945 EDT [INFO ][Thread-2] HOST perf21.mw.lab.eng.bos.redhat.com:rootProcess:jbmFailoverCase3 - ID:JBM-89dde43d-e333-4d35-8f0d-619ac019eaf6----------delegator>JBossMessage[77413672871922117]:PERSISTENT, deliveryId=8286
2011/06/09 12:21:49:956 EDT [INFO ][Thread-2] HOST perf21.mw.lab.eng.bos.redhat.com:rootProcess:jbmFailoverCase3 - ID:JBM-89dde43d-e333-4d35-8f0d-619ac019eaf6----------delegator>JBossMessage[77413672871922117]:PERSISTENT, deliveryId=8286

All logs are in QA Lab: /home/hudson/users-tmp/mnovak/duplicated_messages.zip

Hudson job:
https://hudson.qa.jboss.com/hudson/view/EAP5-JBM-Failover/job/eap-5-topic-trans-kill/19

Logs:
perf21 - deploy host + all producers + db status

perf23 - node A
perf24 - node B
perf25 - node C
perf26 - node D

perf06 - consumer "subscriber1"
perf07 - consumer "subscriber2"
perf08 - consumer "subscriber3"
perf22 - consumer "subscriber4"

Could you look at it, please?

Thank you,

Mirek