This session explains how we evolved a lightweight external routing control plane for Apache RocketMQ to support large-cluster governance without changing brokers, nameservers, or client access patterns.

As multiple business lines gradually shared one common cluster, the real challenge shifted from throughput to governance: faster failure convergence, controllable topic migration, and flexible lane-based traffic isolation. In this talk, I will show why these problems are better solved in the routing layer, why existing NameServer and Broker mechanisms are not sufficient, and how broker write isolation, topic migration, and lane-based traffic isolation can be unified in one control plane to support large-cluster split with a smaller blast radius and more controllable governance.

Speakers:

Xiao Yang: Middleware Engineer at Tongcheng Travel, Apache RocketMQ Committer

Responsible for the architectural evolution, reliability engineering, and platform development of messaging middleware.