Chapter 4. Migrating to KRaft mode

Procedure

1. Retrieve the cluster ID of your Kafka cluster.

   Use the zookeeper-shell tool:

   /opt/kafka/bin/zookeeper-shell.sh localhost:2181 get /cluster/id

   The command returns the cluster ID.

2. Install a KRaft controller quorum to the cluster.

   1. Configure a controller node on each host using the controller.properties file.

      At a minimum, each controller requires the following configuration:

      • A unique node ID
      • The migration enabled flag set to true
      • ZooKeeper connection details
      • Listener name used by the controller quorum
      • A quorum of controller voters

      • Listener name for inter-broker communication

        Example controller configuration

        process.roles=controller
        node.id=1
        
        zookeeper.metadata.migration.enable=true
        zookeeper.connect=zoo1.my-domain.com:2181,zoo2.my-domain.com:2181,zoo3.my-domain.com:2181
        
        listeners=CONTROLLER://0.0.0.0:9090
        controller.listener.names=CONTROLLER
        listener.security.protocol.map=CONTROLLER:PLAINTEXT
        controller.quorum.voters=1@localhost:9090
        inter.broker.listener.name=PLAINTEXT

        The format for the controller quorum is <node_id>@<hostname>:<port> in a comma-separated list. The inter-broker listener name is required for the KRaft controller to initiate the migration.

   2. Set up log directories for each controller node:

      /opt/kafka/bin/kafka-storage.sh format -t <uuid> -c /opt/kafka/config/kraft/controller.properties

      Returns:

      Formatting /tmp/kraft-controller-logs

      Replace <uuid> with the cluster ID you retrieved. Use the same cluster ID for each controller node in your cluster.

      By default, the log directory (log.dirs) specified in the controller.properties configuration file is set to /tmp/kraft-controller-logs. The /tmp directory is typically cleared on each system reboot, making it suitable for development environments only. Set multiple log directories using a comma-separated list, if needed.

   3. Start each controller.

      /opt/kafka/bin/kafka-server-start.sh -daemon /opt/kafka/config/kraft/controller.properties

   4. Check that Kafka is running:

      jcmd | grep kafka

      Returns:

      process ID kafka.Kafka /opt/kafka/config/kraft/controller.properties

      Check the logs of each controller to ensure that they have successfully joined the KRaft cluster:

      tail -f /opt/kafka/logs/controller.log

3. Enable migration on each broker.

   1. If running, stop the Kafka broker running on the host.

      /opt/kafka/bin/kafka-server-stop.sh
      jcmd | grep kafka

      If using a multi-node cluster, refer to Section 3.6, “Performing a graceful rolling restart of Kafka brokers”.

   2. Enable migration using the server.properties file.

      At a minimum, each broker requires the following additional configuration:

      • Inter-broker protocol version set to version 3.8
      • The migration enabled flag
      • Controller configuration that matches the controller nodes
      • A quorum of controller voters

      Example broker configuration

      broker.id=0
      inter.broker.protocol.version=3.8
      
      zookeeper.metadata.migration.enable=true
      zookeeper.connect=zoo1.my-domain.com:2181,zoo2.my-domain.com:2181,zoo3.my-domain.com:2181
      
      listeners=CONTROLLER://0.0.0.0:9090
      controller.listener.names=CONTROLLER
      listener.security.protocol.map=CONTROLLER:PLAINTEXT
      controller.quorum.voters=1@localhost:9090

      The ZooKeeper connection details should already be present.

   3. Restart the updated broker:

      /opt/kafka/bin/kafka-server-start.sh -daemon /opt/kafka/config/kraft/server.properties

      The migration starts automatically and can take some time depending on the number of topics and partitions in the cluster.

   4. Check that Kafka is running:

      jcmd | grep kafka

      Returns:

      process ID kafka.Kafka /opt/kafka/config/kraft/server.properties

4. Check the log on the active controller to confirm that the migration is complete:

   /opt/kafka/bin/zookeeper-shell.sh localhost:2181 get /controller

   Look for an INFO log entry that says the following: Completed migration of metadata from ZooKeeper to KRaft.

5. Switch each broker to KRaft mode.

   1. Stop the broker, as before.

   2. Update the broker configuration in the server.properties file:

      • Replace the broker.id with a node.id using the same ID
      • Add a broker KRaft role for the broker
      • Remove the inter-broker protocol version (inter.broker.protocol.version)
      • Remove the migration enabled flag (zookeeper.metadata.migration.enable)
      • Remove ZooKeeper configuration
      • Remove the listener for controller and broker communication (control.plane.listener.name)

      Example broker configuration for KRaft

      node.id=0
      process.roles=broker
      
      listeners=CONTROLLER://0.0.0.0:9090
      controller.listener.names=CONTROLLER
      listener.security.protocol.map=CONTROLLER:PLAINTEXT
      controller.quorum.voters=1@localhost:9090

   3. If you are using ACLS in your broker configuration, update the authorizer using the authorizer.class.name property to the KRaft-based standard authorizer.

      ZooKeeper-based brokers use authorizer.class.name=kafka.security.authorizer.AclAuthorizer.

      When migrating to KRaft-based brokers, specify authorizer.class.name=org.apache.kafka.metadata.authorizer.StandardAuthorizer.

   4. Restart the broker, as before.

6. Switch each controller out of migration mode.

   1. Stop the controller in the same way as the broker, as described previously.

   2. Update the controller configuration in the controller.properties file:

      • Remove the ZooKeeper connection details
      • Remove the zookeeper.metadata.migration.enable property
      • Remove inter.broker.listener.name

      Example controller configuration following migration

      process.roles=controller
      node.id=1
      
      listeners=CONTROLLER://0.0.0.0:9090
      controller.listener.names=CONTROLLER
      listener.security.protocol.map=CONTROLLER:PLAINTEXT
      controller.quorum.voters=1@localhost:9090

   3. Restart the controller in the same way as the broker, as described previously.