[TreeCache] viewAccepted()

[ls1528]

I have several Jboss clusters running on a dev/test Unix Server. Each cluster has a distinct partition name and a different multicast address and port. I modified the ../deploy/cluster-service.xml and changed the multicast address and port in the declaration of the JGroupsUDP protocol. The servers start without any errors.

I recently created a Jboss cluster on my Unix production server. It has a distinct partition name and a different multicast address and port than the ones I'm using on the dev/test Unix server. The server starts without any errors.

However, to my surprise, when I start the production Jboss server, the Tree Cache can see and knows about the Jboss servers on the dev/test Unix server. I see the message below in the production startup log file:

10:26:47,389 INFO [TreeCache] viewAccpeted(): [serverdev:61285|101] [serverdev:61285, serverdev:61290, serverdev:61325, serverdev:52951......

On the dev/test Unix server, I see the same message in all the Jboss server log files, however, the dev/test log files also contain a reference to the JBoss production server - "serverprod:56345".

How can I keep the Tree Cache for the two environments (production and dev/test) separated from each other? The Unix servers are on the same subnet. I am running Jboss V4.0.4.GA.

[belaban]

I suggest setting UDP.ip_ttl to a sufficiently low value, so that packets crossing the subnets are discarded by the router.
I would also recommend to change
- partition names
- mcast_addr and
- mcast_port
on both the prod and test lans

[ls1528]

OK. I decreased the value of UDP ip_ttl from 8 to 2 in the ../deploy/cluster-service.xml file on both the production JBoss servers and dev/test JBoss servers. I stopped and restarted all the servers.

The [TreeCache] viewAccepted() messages still appear in the JBoss startup log file on the Unix production server and dev/test server. The partition_names, mcast_addr and mcast_ports are different values on the production JBoss server and dev/test JBoss servers.

Do you have any other suggestions?

[belaban]

Those values are overridden with system properties, are you sure you don't do that ?

[brian.stansberry]

The cluster-service.xml file doesn't control any TreeCache instance. Most likely the one you're seeing logging from is the cache used for HTTP session replication. That's configured in deploy/tc5-cluster.sar/META-INF/jboss-service.xml. Basically the same type of process of using a different cluster name, mcast_addr and mcast_port should be applied to that file as well. The ip_ttl change can be done there too.