Controllers and Brokers

A Kafka cluster can keep serving order events while its active controller changes. It can also have healthy controllers while one partition is unavailable. These situations make sense once you separate the work of managing the cluster from the work of storing and serving its records.

Controllers maintain cluster metadata and coordinate changes to broker participation and partition ownership. Brokers use that metadata to handle application traffic and maintain partition replicas. They also perform coordination work for consumer groups and transactions.

This chapter explains how these roles cooperate, where different requests go, and how deployment choices affect failures and capacity. The examples assume Kafka 4.x in KRaft mode, with ordinary consumers fetching from partition leaders.