That looks like the same problem I'm having in
with JBoss 3.2.3. Unfortunately no one answered.
I don't use JMS, but the behaviour seems similar: if I kill one of the servers, the cluster is not always formed back. If I gracefully shut it down, the cluster is formed back.
It sounds like the merge processing is not working correctly.
Are you seeing messages on the console saying it is attempting to merge
Post the steps and the logs as a bug at www.sf.net/projects/jboss
Enable the example cluster TRACE logging found at the bottom of conf/log4j.xml
to get a complete log.
I do see the following messages (sometimes)
(on the Node which was not the Singleton before the network error)
12:00:47,920 INFO [DefaultPartition:ReplicantManager] Start merging members in DRM service...
12:00:48,045 INFO [HAILServerILService] Notified to stop acting as singleton.
12:00:48,061 INFO [DefaultPartition:ReplicantManager] ..Finished merging members in DRM service
It does not always try to merge and even when it says that it is merging it doesn't seem to merge the cluster state correctly.
I will repeat the test with TRACE on and submit it as a bug.