Kafka vs RabbitMQ vs SQS vs Pulsar

Kafka is a durable, replayable log for high-throughput event streaming; RabbitMQ is a routing-rich message broker for task delivery with low latency and no replay.

Pulsar is a log and a queue in one system, with stateless brokers over BookKeeper storage, native multi-tenancy, and built-in geo-replication.

Kafka records that something happened, SQS says a job needs doing, and RabbitMQ decides where to route this message.

Pulsar decouples stateless brokers from BookKeeper storage for fast scaling and multi-tenancy, while Kafka's broker-owns-storage model wins on maturity, ecosystem, and raw throughput.

Pick a high-cardinality, evenly distributed key so load spreads across all partitions; a skewed key creates a hot partition that bottlenecks one consumer and one broker.