RabbitMQ has had streams since 3.9. How does a stream differ from a queue, and when would you pick one?

A queue deletes a message once a consumer acknowledges it, and each message goes to one consumer. A stream keeps messages: it is a replicated, append-only log, and reading does not remove anything.

How a stream works:

• Each consumer chooses where to start: first, last, next, a specific offset or a timestamp. An offset is a position in the log.
• Data is removed only by retention limits on size or age.

Pick a stream when:

• Many consumers need the same large volume of messages, without one queue per consumer.
• Consumers need to replay history.
• You need very high throughput.

Pick a queue when you want to share work between consumers, with per-message acks, dead-lettering and requeueing.

Streams make RabbitMQ behave a little like Kafka. But if your whole design is built around partitioned logs and long retention, Kafka is usually the better fit.