bind_addr ignored?

[peterb]

I have a clustered machine with 2 interfaces that's using TCP and TCPPING to connect and have specified the bind_addr both in the TCP protocol and through the --host switch. However the DefaultPartition keeps reporting the other interface address as the node name.

This results in faulty HAJNDI lookups because it reports a different address than the one on which JBoss is listening.

Any one have any ideas? Have I missed some configuration parameter?

[peterb]

I figured it out myself. Apparently what I wanted is not possible in 3.2.2 but it is in 3.2.5. In 3.2.5 (and possibly earlier version too, but certainly not 3.2.2) The ClusterPartition mbean has a NodeAddress attribute which can be set to the correct bind address.

[belaban]

I fixed this in the CVS. Did it for UDP, but apparently forgot TCP...
Bela

[belaban]

BTW: the --host switch suffices here, no need for bind_addr in the parameters for JGroups.

Bela