NATS gains ground as teams rethink RabbitMQ

Cloud-native and AI-driven software systems are turning to NATS as a potential alternative to RabbitMQ, according to developers. The shift is driven by lower latency messaging, easier deployment, and a communication model that's designed for distributed services rather than traditional broker-centric queues. A recent update, NATS Server 2.15, released on September 17, emphasizes reliability and operations.

Its developers have optimized stream and consumer scaling, movement, and peer changes. They've also improved backups and source recovery, and reported notable performance gains for replicated JetStream streams that synchronize every message to disk. NATS claims a 10% increase in throughput under typical replicated-stream settings, with tail latency reduced by over 40% using the stricter "sync: always" setting.

However, these benchmarks may not apply universally to all production workloads. The move towards NATS aligns with an architectural shift where developers use a single system for both ephemeral service communication and durable messaging, rather than managing separate infrastructure for each pattern. This is particularly beneficial for AI applications that require real-time communication between many autonomous components that can appear and disappear dynamically.

While there are signs of migration, quantifying the scale of this shift is challenging. Sophotech, a technology consultancy, reports moving around 50 microservices from RabbitMQ to NATS. The consultancy observed a significant reduction in messaging latency and operational work, as well as a decrease in burst-related queue delays.

Despite these benefits, RabbitMQ remains actively developed, introducing features like strict message priorities, delayed retries, and consumer timeouts in version 4.3. It also offers streams and super streams for durable, high-throughput workloads using the Raft consensus algorithm. The choice between NATS and RabbitMQ isn't a straightforward upgrade.

Enterprises dependent on mature AMQP semantics, complex routing, message priorities, or established operational tooling may not find much benefit in switching. RabbitMQ also supports MQTT and STOMP, which could be crucial in heterogeneous environments. NATS has narrowed an earlier functional gap through JetStream, a persistence layer that supports durable streams, consumers, and replay while preserving NATS's subject-based addressing model.

This allows teams to start with transient messaging and add storage selectively, rather than making durability a requirement for every exchange between services.