How to Set Up and Run Kafka on Kubernetes

Apache Kafka is a leading open-source distributed streaming platform first developed at LinkedIn. It consists of several APIs such as the Producer, the Consumer, the Connector and the Streams. Together, those systems act as high-throughput, low-latency platforms for handling real-time data. This is why Kafka is preferred by many top-tier tech companies including Uber, Zalando and AirBnB.

Quite often, we would like to deploy a fully-fledged Kafka cluster in Kubernetes, just because we have a collection of microservices and we need a resilient message broker in the center. We also want to spread the Kafka instances across nodes, to minimize the impact of a failure.

In this tutorial we are going to see an example Kafka deployment with Platform9 Managed Kubernetes (PMK), backed up by some DigitalOcean droplets.

Let’s get started.

Pre-requisites

• Working Kubernetes Cluster
• Permissions to create resources on the cluster

Creating Persistent Volumes

Before we install Helm and the Kafka chart, we need to create some persistent volumes for storing Kafka replication message files.

This step is crucial to be able to enable persistence in our cluster because without that, the topics and messages would disappear after we shutdown any of the servers, as they live in memory.

In our example, we are going to use a local file system, Persistent Volume (PV), and we need one persistent volume for each Kafka instance; so if we plan to deploy three instances, we need three PV’s.

Create and apply first the Kafka namespace and the PV specs:

$ cat namespace.yml
---
apiVersion: v1
kind: Namespace
metadata:
  name: kafka

$ kubectl apply -f namespace.yml
namespace/kafka created


$ cat pv.yml
---
apiVersion: v1
kind: PersistentVolume
metadata:
  name: kafka-pv-volume
  labels:
    type: local
spec:
  storageClassName: manual
  capacity:
    storage: 10Gi
  accessModes:
    - ReadWriteOnce
  hostPath:
    path: "/mnt/data"
---
apiVersion: v1
kind: PersistentVolume
metadata:
  name: kafka-pv-volume-2
  labels:
    type: local
spec:
  storageClassName: manual
  capacity:
    storage: 10Gi
  accessModes:
    - ReadWriteOnce
  hostPath:
    path: "/mnt/data"
---
apiVersion: v1
kind: PersistentVolume
metadata:
  name: kafka-pv-volume-3
  labels:
    type: local
spec:
  storageClassName: manual
  capacity:
    storage: 10Gi
  accessModes:
    - ReadWriteOnce
  hostPath:
    path: "/mnt/data"

$ kubectl apply -f pv.yml

If you are using the Kubernetes UI, you should be able to see the PV volumes on standby:

Installing Helm

We begin by installing Helm on our computer and installing it in Kubernetes, as it’s not bundled by default.

First we download the install script:

$ curl https://raw.githubusercontent.com/kubernetes/helm/master/scripts/get > install-helm.sh

Make the script executable with chmod:

$ chmod u+x install-helm.sh

Create the tiller service account:

$ kubectl -n kube-system create serviceaccount tiller

Next, bind the tiller serviceaccount to the cluster-admin role:

$ kubectl create clusterrolebinding tiller --clusterrole cluster-admin --serviceaccount=kube-system:tiller

Now we can run helm init:

$ helm init --service-account tiller

Now we are ready to install the Kafka chart.

Deploying the Helm Chart

In the past, trying to deploy Kafka on Kubernetes was a good exercise. You had to deploy a working Zookeeper Cluster, role bindings, persistent volume claims and apply correct configuration.

Fortunately for us, with the use of the Kafka Incubator Chart, the whole process is mostly automated (with a few quirks here and there).

We add the Helm chart:

 $ helm repo add incubator https://charts.helm.sh/incubator

Export the chart values in a file:

$ curl https://raw.githubusercontent.com/helm/charts/master/incubator/kafka/values.yaml > values.yaml

Carefully inspect the configuration values, particularly around the parts about persistence and about the number of Kafka stateful sets to deploy.

Then install the chart:

$ helm install --name kafka-demo --namespace kafka incubator/kafka -f values.yml --debug

Check the status of the deployment

$ helm status kafka-demo
LAST DEPLOYED: Sun Apr 19 14:05:15 2020
NAMESPACE: kafka
STATUS: DEPLOYED

RESOURCES:
==> v1/ConfigMap
NAME                  DATA  AGE
kafka-demo-zookeeper  3     5m29s

==> v1/Pod(related)
NAME                    READY  STATUS   RESTARTS  AGE
kafka-demo-zookeeper-0  1/1    Running  0         5m28s
kafka-demo-zookeeper-1  1/1    Running  0         4m50s
kafka-demo-zookeeper-2  1/1    Running  0         4m12s
kafka-demo-zookeeper-0  1/1    Running  0         5m28s
kafka-demo-zookeeper-1  1/1    Running  0         4m50s
kafka-demo-zookeeper-2  1/1    Running  0         4m12s

==> v1/Service
NAME                           TYPE       CLUSTER-IP     EXTERNAL-IP  PORT(S)                     AGE
kafka-demo                     ClusterIP  10.21.255.214         9092/TCP                    5m29s
kafka-demo-headless            ClusterIP  None                  9092/TCP                    5m29s
kafka-demo-zookeeper           ClusterIP  10.21.13.232          2181/TCP                    5m29s
kafka-demo-zookeeper-headless  ClusterIP  None                  2181/TCP,3888/TCP,2888/TCP  5m29s

==> v1/StatefulSet
NAME                  READY  AGE
kafka-demo            3/3    5m28s
kafka-demo-zookeeper  3/3    5m28s

==> v1beta1/PodDisruptionBudget
NAME                  MIN AVAILABLE  MAX UNAVAILABLE  ALLOWED DISRUPTIONS  AGE
kafka-demo-zookeeper  N/A            1                1                    5m29s

During this phase, you may want to navigate to the Kubernetes UI and inspect the dashboard for any issues. Once everything is complete, then the pods and Persistent Volume Claims should be bound and green.

Now we can test the Kafka cluster.

Testing the Kafka Cluster

We are going to deploy a test client that will execute scripts against the Kafka cluster.

Create and apply the following deployment:

$ cat testclient.yml

apiVersion: v1
kind: Pod
metadata:
  name: testclient
  namespace: kafka
spec:
  containers:
  - name: kafka
    image: solsson/kafka:0.11.0.0
    command:
      - sh
      - -c
      - "exec tail -f /dev/null"

$ kubectl apply -f testclient

Then, using the testclient, we create the first topic, which we are going to use to post messages:

$ kubectl -n kafka exec -ti testclient -- ./bin/kafka-topics.sh --zookeeper kafka-demo-zookeeper:2181 --topic messages --create --partitions 1 --replication-factor 1
Created topic "messages".

Here we need to use the correct hostname for zookeeper cluster and the topic configuration.

Next, verify that the topic exists:

$ kubectl -n kafka exec -ti testclient -- ./bin/kafka-topics.sh --zookeeper kafka-demo-zookeeper:2181 --list
Messages

Now we can create one consumer and one producer instance so that we can send and consume messages.

First create one or two listeners, each on its own shell:

$ kubectl -n kafka exec -ti testclient -- ./bin/kafka-console-consumer.sh --bootstrap-server kafka-demo:9092 --topic messages --from-beginning

Then create the producer session and type some messages. You will be able to see them propagate to the consumer sessions:

$ kubectl -n kafka exec -ti testclient -- ./bin/kafka-console-producer.sh --broker-list kafka-demo:9092 --topic messages
>Hi
>How are you?
>Hope you're well
>



Hi
How are you?
Hope you're well

Destroying the Helm Chart

To clean up our resources, we just destroy the Helm Chart and delete the PVs we created earlier:

$ helm delete kafka-demo --purge
$ kubectl delete -f pv.yml -n kafka

Next Steps

Look out for more tutorials showcasing common deployment scenarios within Platform9’s fully-managed Kubernetes platform.