Controller Failure and Leader Election

The active Kafka controller fails while an administrator is changing a topic configuration. Other controllers still have copies of the metadata, but one of them must become the new leader before metadata management can proceed normally.

Choosing that leader requires more than selecting an available server. The replacement must preserve committed history, obtain enough votes, and establish leadership that the other participants can recognize.

This chapter follows that recovery process and explains what it means for pending requests and application traffic. The examples use Kafka 4.3, a stable voting set of three dedicated controllers, and healthy brokers unless the example explicitly introduces a failure. The election flow includes Kafka’s pre-vote mechanism.