We have a test-segment and a production segment with various JBoss 5 clusters on it. In test there is nothing wrong but in production clustering was not working (in the server.log we see on all the nodes that the cluster can only see itself as a node). Application settings are exactly the same, but when we configured the switches in the production segment to allow multicast-udp, the situation quickly becomes unstable. We saw the following:
It was still no working (each node could only see itself as part of a cluster).
All over the place, frequently nodes advertised themself with load -1 in the mod-cluster-manager application. After a reload of the apache webservers, things were normal again (positive load), but allas: not for long. This is unexpected and we have seen this literally all-over-the-place. So random nodes became unavailable to the webserver at random time (the nodes were sill reacheable via the direct IP address, though.
When we rolled back the solution (Allow multicast on the switches and on the network segment), things went back to normal ?
To my knowledge, this should have worked. Can someone suggest anything that points me in the right direction. Also, does there exist an application (war or ear file) to test multicast traffic in the application server itself (besides solutions as jgroups-3.3.3.Final.jar.