12 Essential RocketMQ Best Practices for Reliable Messaging

RocketMQ is a high‑performance, high‑reliability, and highly‑scalable message queue system used for large‑scale distributed messaging. To run it stably in production, follow a set of usage conventions.

12 Code Usage Guidelines

Topic rule: one application per topic, e.g., topic_recharge_order. Reducing the number of topics lowers broker memory overhead, simplifies statistics, and follows the principle of coarse‑grained topics and fine‑grained tags.

Tag rule: use tags for sub‑types, e.g., tag_order_create, tag_order_pay. Tags are lightweight secondary labels under a topic, enable consumer filtering, improve extensibility and maintainability.

Producer group rule: use pg_业务名称 such as pg_recharge_order. Producer groups help control, trace, and load‑balance messages, and support transactional messages and failure callbacks.

Consumer group rule: use cg_topic名称_tag名称 e.g., cg_topic_recharge_order_tag_order_create. Consumer groups manage offsets and idempotence, and naming aids monitoring and alarm management.

Message unique identifier: set the key field to a business primary key like order ID. This aids fast message tracing, de‑duplication, and troubleshooting of lost messages.

Log send result: record SendResult, which contains message ID, status, queue ID, etc., to verify delivery and support observability.

Synchronous send failure handling: persist failed messages to a database and retry later to ensure eventual delivery.

Log‑type applications: use oneway sending to achieve ultra‑low latency and reduce broker load.

Delivery guarantee: RocketMQ provides at‑least‑once delivery; applications must implement idempotent processing to achieve exactly‑once semantics.

Increase consumption capacity: raise consumeThreadMin / consumeThreadMax or add consumer instances; batch consumption further reduces network and processing overhead.

Ordered consumption error handling: return SUSPEND_CURRENT_QUEUE_A_MOMENT instead of throwing exceptions to preserve order.

Concurrent consumption error handling: return RECONSUME_LATER to trigger delayed retry rather than throwing exceptions.

Cluster Deployment Plan

Deploy 2 master + 2 slave broker nodes to ensure high availability and high concurrency.

Deploy 3 nameserver nodes to avoid a single point of failure.

Use high‑performance SSDs for each broker and ensure sufficient disk space; avoid disk usage exceeding 90%.

Regularly back up commitlog files; default async flush, with cold and hot backup strategies for data safety.

Following these conventions builds a highly available, high‑performance, and observable RocketMQ system.