How to use Apache Kafka on DC/OS


Apache Kafka is a distributed high-throughput publish-subscribe messaging system with strong ordering guarantees. Kafka clusters are highly available, fault tolerant, and very durable. DC/OS Kafka gives you direct access to the Kafka API so that existing producers and consumers can interoperate. You can configure and install DC/OS Kafka in moments. Multiple Kafka clusters can be installed on DC/OS and managed independently, so you can offer Kafka as a managed service to your organization.

Kafka uses Apache ZooKeeper for coordination. Kafka serves real-time data ingestion systems with high-throughput and low-latency. Kafka is written in Scala.

Time Estimate:

Approximately 10 minutes.

Target Audience:

  • Anyone interested in Kafka


  • Broker: A Kafka message broker that routes messages to one or more topics.
  • Topic: A Kafka topic is message filtering mechanism in the pub/sub systems. Subscribers register to receive/consume messages from topics.
  • Producer: An application that producers messages to a Kafka topic.
  • Consumer: An application that consumes messages from a Kafka topic.


In this tutorial you will learn how to:

  • Install the Kafka service
  • Use the enhanced DC/OS CLI to create Kafka topics
  • Use Kafka on DC/OS to produce and consume messages

Table of Contents


  • A running DC/OS cluster with 3 private agents, each with 2 CPUs and 2 GB of RAM available.
  • DC/OS CLI installed.

Install Kafka

Typical installation

Install a Kafka cluster with 3 brokers using the DC/OS CLI:

dcos package install kafka

While the DC/OS command line interface (CLI) is immediately available, it takes a few minutes for the Kafka service to start.

Minimal installation

To start a minimal cluster with a single broker, create a JSON options file named kafka-minimal.json:

    "brokers": {
        "count": 1,
        "mem": 512,
        "disk": 1000

Install the Kafka cluster:

dcos package install kafka --options=kafka-minimal.json

Topic management

Add a topic:

dcos kafka topic create topic1 --partitions 1 --replication 1

Produce and consume messages

List Kafka client endpoints

dcos kafka connection
    "address": [
    "dns": [
    "zookeeper": "master.mesos:2181/kafka"

The above shows an example of what a Kafka client endpoint will look like. Note the address and ports will be different from cluster to cluster, since these services are dynamically provisioned. Record the "address" value from your cluster for use in the next step.

Produce a message

dcos node ssh --master-proxy --leader

core@ip-10-0-6-153 ~ docker run -it mesosphere/kafka-client

root@7d0aed75e582:/bin# echo "Hello, World." | ./ --broker-list KAFKA_ADDRESS:PORT --topic topic1

Replace the above KAFKA_ADDRESS:PORT with the Kafka client endpoint address from your cluster.

Consume a message

root@7d0aed75e582:/bin# ./ --zookeeper master.mesos:2181/kafka --topic topic1 --from-beginning
Hello, World.

Hit CTRL-C to stop the Kafka consumer process.



Return to the DC/OS CLI environment (exit the Docker container with CTRL-D, exit the SSH session on the master node with another CTRL-D).

dcos package uninstall --app-id=kafka kafka

Then, use the framework cleaner script to remove your Kafka instance from ZooKeeper and to destroy all data associated with it. The script requires several arguments, the values for which are derived from your service name:

framework-role is kafka-role framework-principal is kafka-principal zk_path is kafka

Further resources