Replicated broker defaults to using 3 containers -- each running a replicated view of the broker using a share-nothing architecture. You can use more than 3 if you like, but use an odd number (5, 7, etc.) so that you can still achieve a majority of brokers even during failure scenarios.
For each cluster of replicated brokers, one broker is always elected Master while the others become Slaves. If the Master fails, an election will happen among the remaining slaves and a new Master will be elected. On the client side, the failover will be transparent if using failover: transport, fabric discovery transport, or a local gateway to the cluster.