In clustered environment the process of shutting down single AS instance should be transparent to clients. This is not the case since applications are undeployed at first while connectors are still servicing requests. This causes clients not to be redirected to another cluster nodes and to receive resource not available exceptions.

From the client perspective, shutdown process should work as follows:

1) stop connectors so they refuse servicing new requests
2) complete running requests
3) undeploy applications