From Kafka to RabbitMQ: thinking through the next messaging layer

A roadmap discussion about async jobs, notifications, and event fan-out.

Not shipped yet

AdroitPad does not currently have a message queue. The question is being considered because asynchronous jobs, notifications, and event fan-out eventually need a durable handoff rather than a request that stays open until work completes.

Compare the shape of the work

Kafka is compelling for durable streams, replay, and high-volume event histories. RabbitMQ is often simpler for work queues, routing, acknowledgements, and short-lived task delivery. The right choice depends on retention, ordering, replay, operational ownership, and expected throughput—not on which technology is fashionable.

This remains an evaluation, not a release announcement. Feedback from teams with similar async workloads would be useful before a queue becomes part of the platform.

Practical follow-up

A useful pilot would begin with one non-critical workflow and measure delivery latency, retry behavior, ordering, operational effort, and the value of replay. That evidence is more useful than choosing a broker from a generic feature checklist.

The eventual abstraction should keep business code independent of broker APIs. Define message envelopes, idempotency keys, dead-letter behavior, and observability first so the platform can change its transport without changing every producer and consumer.

Conclusion

The messaging decision should follow the work shape, not a generic technology comparison. The diagram places Kafka’s durable stream and replay model beside RabbitMQ’s routing and delivery-oriented queue model. A pilot should measure latency, ordering, retries, dead letters, replay value, and operational effort on one non-critical workflow. Message envelopes and idempotency keys should be designed before producers and consumers depend on broker APIs. That keeps business logic portable while the platform is still being evaluated. Until the evidence supports a choice, this remains a roadmap decision rather than a shipped capability.