RabbitMQ

RabbitMQ is an open-source message broker: a server that accepts messages from producers, stores them in queues, and delivers them to consumers, which acknowledge each message once it has been handled. It is the transport layer of a background-job setup, not the job library itself. Celery in Python, for example, can use RabbitMQ as its broker, and services in different languages can talk to each other through it directly.

Messages are published to exchanges, which route them to queues by name, by pattern, or to every bound queue at once. That routing makes RabbitMQ suited to work queues, fan-out of events to several services, and request/reply between services. Unacknowledged messages are redelivered, failed ones can be sent to a dead-letter queue, and consumers can be throttled so a slow worker never receives more than it can handle.

There are three queue types. Quorum queues replicate messages across cluster nodes with the Raft consensus algorithm and are the default for durable work; recent releases added strict priority levels and delayed retries to them. Classic queues remain for simple single-node use, and streams keep an append-only log that consumers can replay, covering some event-log use cases without a separate streaming platform.

The broker speaks AMQP 0-9-1, AMQP 1.0, MQTT, and STOMP, and has client libraries for nearly every language. It runs as a single Docker container for development and as a cluster in production, and is available as a managed service from CloudAMQP, Amazon MQ, and other providers. The core is released under the Mozilla Public License 2.0; Broadcom sells Tanzu RabbitMQ, a commercial edition with extra features and enterprise support.