The process of replicating data between Kafka clusters is called "mirroring", to differentiate cross-cluster replication from replication among nodes within a single cluster. A common use for mirroring is to maintain a separate copy of a Kafka cluster in another data center.

Kafka's MirrorMaker tool reads data from topics in one or more source Kafka clusters, and writes corresponding topics to a destination Kafka cluster (using the same topic names):

To mirror more than one source cluster, start at least one MirrorMaker instance for each source cluster.

You can also use multiple MirrorMaker processes to mirror topics within the same consumer group. This can increase throughput and enhance fault-tolerance: if one process dies, the others will take over the additional load.

The source and destination clusters are completely independent, so they can have different numbers of partitions and different offsets. The destination (mirror) cluster is not intended to be a mechanism for fault-tolerance, because the consumer position will be different. (The MirrorMaker process will, however, retain and use the message key for partitioning, preserving order on a per-key basis.) For fault tolerance we recommend using standard within-cluster replication.