5.4. Configuring Kafka

Kafka brokers need ZooKeeper to store some parts of their configuration as well as to coordinate the cluster (for example to decide which node is a leader for which partition). Connection details for the ZooKeeper cluster are stored in the configuration file. The field zookeeper.connect contains a comma-separated list of hostnames and ports of members of the zookeeper cluster.

For example:

zookeeper.connect=zoo1.my-domain.com:2181,zoo2.my-domain.com:2181,zoo3.my-domain.com:2181

Kafka will use these addresses to connect to the ZooKeeper cluster. With this configuration, all Kafka znodes will be created directly in the root of ZooKeeper database. Therefore, such a ZooKeeper cluster could be used only for a single Kafka cluster. To configure multiple Kafka clusters to use single ZooKeeper cluster, specify a base (prefix) path at the end of the ZooKeeper connection string in the Kafka configuration file:

zookeeper.connect=zoo1.my-domain.com:2181,zoo2.my-domain.com:2181,zoo3.my-domain.com:2181/my-cluster-1

Listeners are used to connect to Kafka brokers. Each Kafka broker can be configured to use multiple listeners. Each listener requires a different configuration so it can listen on a different port or network interface.

To configure listeners, edit the listeners property in the Kafka configuration properties file. Add listeners to the listeners property as a comma-separated list. Configure each property as follows:

<listener_name>://<hostname>:<port>

If <hostname> is empty, Kafka uses the java.net.InetAddress.getCanonicalHostName() class as the hostname.

listeners=internal-1://:9092,internal-2://:9093,replication://:9094

When a Kafka client wants to connect to a Kafka cluster, it first connects to the bootstrap server, which is one of the cluster nodes. The bootstrap server provides the client with a list of all the brokers in the cluster, and the client connects to each one individually. The list of brokers is based on the configured listeners.

The advertised.listeners property is formatted in the same way as the listeners property.

listeners=internal-1://:9092,internal-2://:9093
advertised.listeners=internal-1://my-broker-1.my-domain.com:1234,internal-2://my-broker-1.my-domain.com:1235

The inter-broker listener can be assigned to a port of your choice. When multiple listeners are configured, you can define the name of the inter-broker listener in the inter.broker.listener.name property of your broker configuration.

Here, the inter-broker listener is named as REPLICATION:

listeners=REPLICATION://0.0.0.0:9091
inter.broker.listener.name=REPLICATION

By default, communication between the controllers and brokers uses a dedicated controller listener. Controllers are responsible for coordinating administrative tasks, such as partition leadership changes, so one or more of these listeners is required.

Specify listeners to use for controllers using the controller.listener.names property. You can specify a dynamic quorum of controllers using the controller.quorum.bootstrap.servers property. The quorum enables a leader-follower structure for administrative tasks, with the leader actively managing operations and followers as hot standbys, ensuring metadata consistency in memory and facilitating failover.

listeners=CONTROLLER://0.0.0.0:9090
controller.listener.names=CONTROLLER
controller.quorum.bootstrap.servers=localhost:9090

The format for the controller quorum is <hostname>:<port>.

Apache Kafka stores all records it receives from producers in logs. The logs contain the actual data, in the form of records, that Kafka needs to deliver. Note that these records differ from application log files, which detail the broker’s activities.

log.dirs=/var/lib/kafka

For performance reasons, you can configure log.dirs to multiple directories and place each of them on a different physical device to improve disk I/O performance. For example:

log.dirs=/var/lib/kafka1,/var/lib/kafka2,/var/lib/kafka3

Broker ID is a unique identifier for each broker in the cluster. You can assign an integer greater than or equal to 0 as broker ID. The broker ID is used to identify the brokers after restarts or crashes and it is therefore important that the id is stable and does not change over time. The broker ID is configured in the broker properties file:

broker.id=1