3.2.5 ... a long time ago.
I suppose that you start the master after you stop the last node from the cluster.
In that case the client loose informations about the cluster (later there is a auto recovery imlementation for the client invoker configuration, I'm not sure whether this is also in 3.2).
The problem should solved if the client remove the JNDI context and try to connect from scratch.
Or do you mean something different with 'halted state' ?
Yes I meant the same, thanks much for your help
I faced a similar issue last week in production, the two nodes in cluster were working fine for a long time, we applied a patch having change in a .class file. We restarted master nodeafter applying patch but it stopped receiving live traffic, however the child node was receiving live traffic, then we stopped child node to apply patch; as child stopped; the master node started receiving traffic, later on when we started child node, but then child node went into halted state i.e. did not received any live traffic while master was receiving. Then we stopped both nodes and started both at the same time, but still one node was receiving traffic and other did not. Please help me out on what is the valid procedure to start nodes in a JBoss cluster. In case we need to apply a patch having changes in binaries, how we stop the instances and start again. Please note that we are not using Form Deployment, we have deployed our application in exploded .ear form in the JBOSS_HOME/server/all/deploy directory. The JBoss 3.2.5 is running on Windows 2008 Server and JBoss is installed as Windows service using Java Wrapper. The JDK is 1.4.2.