How is Kafka different from a traditional message broker such as RabbitMQ?

Persistence: Kafka persists everything to disk by design and keeps it for a retention period. RabbitMQ can persist messages, but queues are meant to drain, and long backlogs hurt performance (quorum queues and streams improve this).

Partitions and scale: Kafka scales a topic by adding partitions across brokers, and a consumer group splits partitions among its members. RabbitMQ scales through competing consumers on a queue, and a single queue is traditionally bound to one node (sharding and quorum queues change that picture).

Replay: Kafka consumers can rewind offsets and reprocess. With classic RabbitMQ queues, an acknowledged message is gone. RabbitMQ Streams add a log-style replay capability.

Ordering: Kafka orders within a partition. RabbitMQ preserves order within a queue for a single consumer, but redelivery and multiple consumers break it.

Consumer groups: In Kafka, each group gets its own full copy of the stream. In RabbitMQ you get fan-out by binding multiple queues to an exchange.

Throughput: Kafka typically wins at very high sustained throughput because of sequential I/O and batching. RabbitMQ is often lower latency for small workloads and offers richer routing, per-message TTL, priority queues and dead-letter exchanges.