Something to keep in mind— different clustering options provide different capabilities. Network of Brokers provides scaling, reduced latency for the “next” message to be processed (fault tolerance), while Master-Slave / LevelDB replication provide high availability of messages that have yet to be consumed.
The right clustering / HA strategy can change the expected behavior of the overall system— specifically, are you expecting to maintain message ordering?
Testing master-slave HA is straight forward with KahaDB.. just point two instances to the same KahaDB data directory and you’ll see the second instance logging that it can’t obtain the lock. Drop the first instance, and in a few seconds the slave will take over the data store.
LevelDB w/ replication generally starts with three nodes to get up and running.