Message queues, ranked for teams that just need a queue

SQS, RabbitMQ, NATS, Kafka and Redis Streams ranked for ordinary background work. Licensing drama included at no extra charge.

The bench2 min readDatabases

Ranked dot plot of five queue options scored on operations burden and cost

You want SQS. Not because it's elegant (it isn't), but because it's the only queue on this list that has never once become your team's problem at 3 a.m. If you're not on AWS, NATS. The rest of the list is for people whose requirements earned the complexity.

Criteria, stated before scoring: operational burden 40%, cost at modest scale (1M messages/day) 30%, delivery semantics fit for background jobs 20%, tooling 10%. Prices as of March 2026.

The table

RankQueueCost at 1M msgs/dayOps burden
1SQS~$12/monone
2NATSa small VMlow
3RabbitMQa medium VMmoderate, well-documented
4Redis Streams (Valkey)often already runninglow, with caveats
5Kafka (MSK / Redpanda)starts ~$200/mo managedyou know if you need it

Notes per rung

SQS at a million messages a day costs about what lunch does, off the public per-request pricing. Visibility timeouts are the whole learning curve. It is boring in the way a queue should be boring.

NATS is the best thing that ever happened to the self-hosted tier: a single static binary, JetStream when you need persistence, and clustering that doesn't require a consultancy. It wins the "not on AWS" bracket outright.

RabbitMQ remains the reference implementation of "actual queue semantics": routing, dead-lettering, per-message acks. It also remains something you operate, with Erlang cluster behaviors to learn the week you least want to.

Redis Streams gets an asterisk the size of the license file. Redis changed licenses in 2024, the community forked to Valkey, and the tooling world spent two years sorting itself out. The Streams API is fine for light work; run it on Valkey and don't build your delivery guarantees on XAUTOCLAIM folklore.

The skip

Kafka, for background jobs. Kafka is a distributed log for event streaming at volumes with commas in them; as a work queue for resizing images it's a $200-per-month way to make five engineers learn about consumer group rebalancing. When you genuinely need it you will not need this site to tell you. Until then: skip it, and put the difference toward literally anything else.