Exceptions in client on cluster node shutdown

[michael.jacobs]

In our environment we have multiple clustered JBoss AS 4.2.3.GA instances serving ejb's to clients (also running within 4.2.3.GA).  When a cluster node gets shutdown we sometimes see the following exception in the client log:

Caused by: org.jboss.remoting.ServerInvoker$InvalidStateException: Can not process invocation request since is not in started state.

        at org.jboss.remoting.ServerInvoker.invoke(ServerInvoker.java:820)

Caused by: org.jboss.remoting.ServerInvoker$InvalidStateException: Can not process invocation request since is not in started state.

        at org.jboss.remoting.ServerInvoker.invoke(ServerInvoker.java:820)

Based on the timing it looks like probably the client gets the stub, then just before it's invocation, the shutting down node begins to undeploy its beans.  I am somewhat new to AS clustering, but I am guessing the stub invokes against this server since it just has not yet be notified that that node is not longer available for business.  We are using a round robin load balance policy.  My question is there any way within the framework to make the stub "try again" to another node, or should be code this ourselves.  I think using an Interceptor would be an option, but we are using EJB 2.

Thanks, MJ

[wdfink]

Do you use EJB2?

The it might be a solution of 'clustered-stateless-unified-invoker', see standardjboss.xml configuration.

Here a SingleRetryInterceptor is in the chain. (RetryInterceptor is also available).

But I suppose you will not solve each gap, I found sporadicaly such Exceptions during shutdown a note from a cluster under load.

For me the problem is more strange because we use the TransactionStickyLB with UserTransactions, and here the gap is big because the SLSB from app is down but the UTx is available and the UTx.begin() will work but can not reach the SLSB.

[michael.jacobs]

We are using the clustered-stateless-unified-invoker that contains the SingleRetryInterceptor, but I don't think that is helping us since it is only triggered on ServiceUnavailableException which is not what we are getting.

[wdfink]

The I suppose you have to code yourselve if you would not accept the sporadical failure in case of shutdown one node.

But you have to keep in mind that it will only work if you don't use UserTransactions.

[michael.jacobs]

Ok, thanks for the clarification.