The second cluster configuration is single-node multiple-broker (SNMB). This cluster is used when there is just one node but inner redundancy is needed.
When a topic is created in Kafka, the system determines how each replica of a partition is mapped to each broker. In general, Kafka tries to spread the replicas across all available brokers.
The messages are first sent to the first replica of a partition (to the current broker leader of that partition) before they are replicated to the remaining brokers.
The producers may choose from different strategies for sending messages (synchronous or asynchronous mode). Producers discover the available brokers in a cluster and the partitions on each (all this by registering watchers in ZooKeeper).
In practice, some of the high volume topics are configured with more than one partition per broker. Remember that having more partitions increases the I/O parallelism for writes and this increases the...
