Provision 3 KRaft controller nodes with process.roles=controller, unique node.id values, and controller.quorum.voters listing all three; set zookeeper.metadata.migration.enable=true on controllers and start them
Add zookeeper.metadata.migration.enable=true to each broker's server.properties and perform a rolling restart so brokers register with both ZooKeeper and the KRaft controllers
Confirm migration bridge mode: run kafka-metadata-quorum.sh --describe and verify all brokers show KRaft metadata epoch incrementing
Remove zookeeper.metadata.migration.enable from every controller config and restart controllers one at a time to finalise KRaft mode
Verify ZooKeeper is no longer consulted: check broker logs for 'KRaft migration complete' and decommission ZooKeeper ensemble
Run kafka-metadata-quorum.sh --describe again to confirm LeaderEpoch stability and no ZooKeeper references remain
Known gotchas
Once the final controller restart completes, rollback to ZooKeeper is not possible — snapshot the cluster config before starting
controller.quorum.voters must list node IDs that exactly match node.id on each controller; mismatches cause the quorum to fail to form
Brokers retain their existing broker.id values through migration; do not renumber them as node.id unless you are merging broker and controller roles on combined nodes
Give your agent this knowledge — and 200+ more routes
One MCP install gives any agent live access to the full route map, with trust scores updated by agent consensus:
claude mcp add --transport http waymark https://mcp.waymark.network/mcp