ClusterPartition's view members intermittently contain unexp

[asaunders]

I'm running JBoss3.2.3 on Redhat 9. My config contains 3 clustered nodes on the same machine. At startup the nodes appear in the ClusterPartition MBean's View attribute containing each node's ID where the ID's are the URLs of the JNDI providers of the nodes - each node has a JNDI provier. This is what I expect to see, e.g. View = [192.168.1.100:1099, 192.168.1.100:1199, 192.168.1.100:1299]

Intermittently the node ID of one or more of the nodes appear in the View of another node with a URL which is not the JNDI provider of the node. This happens despite the fact that all JNDI providers continue to run. There's nothing obvious in the logs other than the server reporting a membership change notification (although no servers have been stopped/started). So the view on one node will appear to be sometheing like [192.168.1.100:1099, 192.168.1.100:1199, 192.168.1.100:33354], where the 1299 node has been replaced. However, the ClusterPartition MBean of the 1299 node still shows its NodeName attribute as 192.168.1.100:1299 and it's View is [192.168.1.100:1099, 192.168.1.100:1199, 192.168.1.100:1299]

Once in this state it requires a restart of one or more of the nodes to rectify the situation.

Has anyone seen this problem and/or can suggest a solution?

Thanks.

[asaunders]

Hmmm, no replies:-(

Is anyone from the JBoss group monitoring this forum? If so, perhaps you could suggest a better place to raise this issue. It has pretty serious consequences. Apps that rely on the cluster view information to locate JNDI providers of peer nodes stop working. The only way I've found to fix this is to restart the node with the corrupted view. This is less than ideal for any app that has high availability requirements.

[slaboure]

Please try with 3.2.5, I don't think you will get much help in the forums if you don't have a recent version of JBoss.

Cheers,


sacha