Is this expected clustering behaviour and if so, why??

[moorec]

We have been doing some cluster tessting and noticed a rather unexpected result. We have defined 4 jboss nodes each with a local JNDI definition that uses port 1099. Each has it's own ip address. All 4 nodes belong to the same cluster which is contacted on port 1100.

Here's the problem. We run a client program that contacts node 3 on 1099 and which is shut down. Node 1, which is running, responds instead.

Is there an order in Jboss where a client tries to get to the local JNDI service and if not found, climbs a hierarchy until it reaches the cluster manager which redirects the request to a listening server??

Thanks c.

[slaboure]

more or less yes. When the client cannot reach a JNDI server, it will try to discover a HA-JNDI service. Which is why you see this behaviour.

Cheers,


sacha

[moorec]

Thanks for that. My collegue say "bugger". However, here's another question. If we have 2 clusters, one with the 4 nodes identified in the original question and another with 4 different nodes, how can we stop the second cluster's nodes from responding to a request destined for nodes in cluster 1. Our setup has all nodes in all clusters listening on port 1100 but different ip addresses.

Note: in this scenario, all nodes in cluster 1 are down.
Cheers

[slaboure]

Take a recent CVS snapshot: you will then be able to define the "jnp.partitionName" system property and set the appropriate partition name. Client side Automatic Discovery will then restrict its discovery to this partition only.

To entirely disable the AD feature, set the jnp.disableDiscovery setting.

Cheers,


sacha