Currently java clustering requires to statically set the list of cluster nodes within the connection URL. The only way of achieving a kind of load balancing is by manually randomizing the list. Several improvements could be made:
          o The java library should randomly pickup one node name. This would allow to use a single deployment file for all the clients.
          o The java library should update the list of nodes when a new node is added in the cluster. That is to say that the java client should handle the clustering management messages and somehow persist the updated list of broker nodes.
          o A less important but very nice to have feature would be for the java client library to gather information on the node loads at connection time. The client library would then use a pluggable heuristic to decide on which node of the cluster it should potentially reconnect.