MQ Commands Cheatsheet - Kafka/RabbitMQ/RocketMQ Reference

Most message queue incidents are chronic rather than dramatic: a consumer that silently stops, a backlog that keeps growing, or a connection count that only ever climbs. The triage order is consistent even though the commands differ per broker. Read the lag or queue depth first, then check how many consumers are registered and what state they are in, so you can tell "nobody is consuming" apart from "consuming too slowly". Then inspect the connection list to see whether the leak lives in the application pool or inside the broker. Kafka, RabbitMQ, RocketMQ and Redis Pub/Sub each report this differently, and Redis Pub/Sub is the odd one out: it has no backlog to inspect and drops messages while a consumer is disconnected, so the diagnostic question changes entirely. Commands are grouped by backlog, connections, cluster inspection and load testing, including throughput and record-count flags for perf tests. Treat a growing backlog as a symptom to be explained rather than a queue to be emptied, and take the message offset as the unit of investigation.