15 Minutes to get a Kafka Cluster running on Kubernetes – and start producing and consuming from a Node application

For  workshop I will present on microservices and communication patterns I need attendees to have their own local Kafka Cluster. I have found a way to have them up and running in virtually no time at all. Thanks to the combination of:

• Kubernetes
• Minikube
• The Yolean/kubernetes-kafka GitHub Repo with Kubernetes yaml files that creates all we need (including Kafka Manager)

Prerequisites:

• Minikube and Kubectl are installed
• The Minikube cluster is running (minikube start)

In my case the versions are:

Minikube: v0.22.3, Kubectl Client 1.9 and (Kubernetes) Server 1.7:

The steps I went through:

Git Clone the GitHub Repository: https://github.com/Yolean/kubernetes-kafka

From the root directory of the cloned repository, run the following kubectl commands:

(note: I did not know until today that kubectl apply –f can be used with a directory reference and will then apply all yaml files in that directory. That is incredibly useful!)

kubectl apply -f ./configure/minikube-storageclass-broker.yml
kubectl apply -f ./configure/minikube-storageclass-zookeeper.yml

(note: I had to comment out the reclaimPolicy attribute in both files – probably because I am running a fairly old version of Kubernetes)

kubectl apply -f ./zookeeper

kubectl apply -f ./kafka

(note: I had to change API version in 50pzoo and 51zoo as well as in 50kafka.yaml from apiVersion: apps/v1beta2 to apiVersion: apps/v1beta1 – see https://github.com/kubernetes/kubernetes/issues/55894 for details; again, I should upgrade my Kubernetes version)

To make Kafka accessible from the minikube host (outside the K8S cluster itself)

kubectl apply -f ./outside-services

This exposes Services as type NodePort instead of ClusterIP, making them available for client applications that can access the Kubernetes host.

I also installed (Yahoo) Kafka Manager:

kubectl apply -f ./yahoo-kafka-manager

(I had to change API version in kafka-manager from apiVersion: apps/v1beta2 to apiVersion: apps/v1beta1 )

At this point, the Kafka Cluster is running. I can check the pods and services in the Kubernetes Dashboard as well as through kubectl on the command line. I can get the Port at which I can access the Kafka Brokers:

And I can access the Kafka Manager at the indicated Port.

Initially, no cluster is visible in Kafka Manager. By providing the Zookeeper information highlighted in the figure (zookeeper.kafka:2181) I can make the cluster visible in this user interface tool.

Finally the eating of the pudding: programmatic production and consumption of messages to and from the cluster. Using the world’s simplest Node Kafka clients, it is easy to see the stuff is working. I am impressed.

I have created the Node application and its package.json file. Then added the kafka-node dependency (npm install kafka-node –save). Next I created the producer:

// before running, either globally install kafka-node  (npm install kafka-node)
// or add kafka-node to the dependencies of the local application

var kafka = require('kafka-node')
var Producer = kafka.Producer
KeyedMessage = kafka.KeyedMessage;

var client;
KeyedMessage = kafka.KeyedMessage;

var APP_VERSION = "0.8.5"
var APP_NAME = "KafkaProducer"

var topicName = "a516817-kentekens";
var KAFKA_BROKER_IP = '192.168.99.100:32400';

// from the Oracle Event Hub - Platform Cluster Connect Descriptor
var kafkaConnectDescriptor = KAFKA_BROKER_IP;

console.log("Running Module " + APP_NAME + " version " + APP_VERSION);

function initializeKafkaProducer(attempt) {
  try {
    console.log(`Try to initialize Kafka Client at ${kafkaConnectDescriptor} and Producer, attempt ${attempt}`);
    const client = new kafka.KafkaClient({ kafkaHost: kafkaConnectDescriptor });
    console.log("created client");
    producer = new Producer(client);
    console.log("submitted async producer creation request");
    producer.on('ready', function () {
      console.log("Producer is ready in " + APP_NAME);
    });
    producer.on('error', function (err) {
      console.log("failed to create the client or the producer " + JSON.stringify(err));
    })
  }
  catch (e) {
    console.log("Exception in initializeKafkaProducer" + JSON.stringify(e));
    console.log("Try again in 5 seconds");
    setTimeout(initializeKafkaProducer, 5000, ++attempt);
  }
}//initializeKafkaProducer
initializeKafkaProducer(1);

var eventPublisher = module.exports;

eventPublisher.publishEvent = function (eventKey, event) {
  km = new KeyedMessage(eventKey, JSON.stringify(event));
  payloads = [
    { topic: topicName, messages: [km], partition: 0 }
  ];
  producer.send(payloads, function (err, data) {
    if (err) {
      console.error("Failed to publish event with key " + eventKey + " to topic " + topicName + " :" + JSON.stringify(err));
    }
    console.log("Published event with key " + eventKey + " to topic " + topicName + " :" + JSON.stringify(data));
  });

}

//example calls: (after waiting for three seconds to give the producer time to initialize)
setTimeout(function () {
  eventPublisher.publishEvent("mykey", { "kenteken": "56-TAG-2", "country": "nl" })
}
  , 3000)

and ran the producer:

The create the consumer:

var kafka = require('kafka-node');

var client;

var APP_VERSION = "0.8.5"
var APP_NAME = "KafkaConsumer"

var eventListenerAPI = module.exports;

var kafka = require('kafka-node')
var Consumer = kafka.Consumer

// from the Oracle Event Hub - Platform Cluster Connect Descriptor

var topicName = "a516817-kentekens";

console.log("Running Module " + APP_NAME + " version " + APP_VERSION);
console.log("Event Hub Topic " + topicName);

var KAFKA_BROKER_IP = '192.168.99.100:32400';

var consumerOptions = {
    kafkaHost: KAFKA_BROKER_IP,
    groupId: 'local-consume-events-from-event-hub-for-kenteken-applicatie',
    sessionTimeout: 15000,
    protocol: ['roundrobin'],
    fromOffset: 'earliest' // equivalent of auto.offset.reset valid values are 'none', 'latest', 'earliest'
};

var topics = [topicName];
var consumerGroup = new kafka.ConsumerGroup(Object.assign({ id: 'consumerLocal' }, consumerOptions), topics);
consumerGroup.on('error', onError);
consumerGroup.on('message', onMessage);

consumerGroup.on('connect', function () {
    console.log('connected to ' + topicName + " at " + consumerOptions.host);
})

function onMessage(message) {
    console.log('%s read msg Topic="%s" Partition=%s Offset=%d'
    , this.client.clientId, message.topic, message.partition, message.offset);
}

function onError(error) {
    console.error(error);
    console.error(error.stack);
}

process.once('SIGINT', function () {
    async.each([consumerGroup], function (consumer, callback) {
        consumer.close(true, callback);
    });
});

and ran the consumer – which duly consumed the event published by the publisher. It is wonderful.

Use

kubectl delete ns kafka

or:

kubectl delete pods,services,pvc,pv,sts,cm –all –namespace=”kafka”

to remove the namespace kafka and its contents.

Resources

The main resources is the GitHub Repo: https://github.com/Yolean/kubernetes-kafka . Absolutely great stuff.

Also useful: npm package kafka-node – https://www.npmjs.com/package/kafka-node

Documentation on Kubernetes: https://kubernetes.io/docs/user-journeys/users/application-developer/foundational/#section-2 – with references to Kubectl and Minikube – and the Katakoda playground: https://www.katacoda.com/courses/kubernetes/playground