The guide shows you how to use the Confluent for Kubernetes (CFK) operator to deploy Apache Kafka clusters on Google Kubernetes Engine (GKE).
Kafka is an open source, distributed publish-subscribe messaging system for handling high-volume, high-throughput, and real-time streaming data. You can use Kafka to build streaming data pipelines that move data reliably across different systems and applications for processing and analysis.
This guide is intended for platform administrators, cloud architects, and operations professionals interested in deploying Kafka clusters on GKE.
You can also use the CFK operator to deploy other components of the Confluent Platform, such as the web-based Confluent Control center, Schema Registry, or KsqlDB. However, this guide focuses only on Kafka deployments.
Objectives
- Plan and deploy GKE infrastructure for Apache Kafka
- Deploy and configure the CFK operator
- Configure Apache Kafka using the CFK operator to ensure availability, security, observability, and performance
Benefits
CFK offers the following benefits:
- Automated rolling updates for configuration changes.
- Automated rolling upgrades with no impact to Kafka availability.
- If a failure occurs, CFK restores a Kafka Pod with the same Kafka broker ID, configuration, and persistent storage volumes.
- Automated rack awareness to spread replicas of a partition across different racks (or zones), improving availability of Kafka brokers and limiting the risk of data loss.
- Support for aggregated metrics export to Prometheus.
Deployment architecture
Each data partition in a Kafka cluster has one leader broker and can have one or more follower brokers. The leader broker handles all reads and writes to the partition. Each follower broker passively replicates the leader broker.
In a typical Kafka setup, you also use an open source service called ZooKeeper to coordinate your Kafka clusters. This service helps by electing a leader among the brokers and triggering failover in case of failures.
You can also deploy Kafka configuration without Zookeeper by activating KRaft mode, but this method is not considered production-ready due to lack of support for KafkaTopic resources, and credential authentication.
Availability and disaster recovery
This tutorial uses separate node pools and zones for Kafka and ZooKeeper clusters to ensure high availability and prepare for disaster recovery.
Highly available Kubernetes clusters in Google Cloud rely on regional clusters spanning multiple nodes and availability zones. This configuration improves fault tolerance, scalability, and geographic redundancy. This configuration also lets you perform rolling updates and maintenance while providing SLAs for uptime and availability. For more information, see Regional clusters.
Deployment diagram
The following diagram shows a Kafka cluster running on multiple nodes and zones in a GKE cluster:
In the diagram, the Kafka StatefulSet is deployed across three nodes
in three different zones. You can control this configuration by setting the
required Pod
affinity
and
topology spread
rules on the Kafka
custom resource specification.
If one zone fails, using the recommended configuration, GKE reschedules Pods on new nodes and replicates data from the remaining replicas, for both Kafka and Zookeeper.
The following diagram shows a ZooKeeper StatefulSet
deployed across three
nodes in three different zones:
Costs
In this document, you use the following billable components of Google Cloud:
To generate a cost estimate based on your projected usage,
use the pricing calculator.
When you finish the tasks that are described in this document, you can avoid continued billing by deleting the resources that you created. For more information, see Clean up.
Before you begin
- Sign in to your Google Cloud account. If you're new to Google Cloud, create an account to evaluate how our products perform in real-world scenarios. New customers also get $300 in free credits to run, test, and deploy workloads.
- Install the Google Cloud CLI.
-
To initialize the gcloud CLI, run the following command:
gcloud init
-
Create or select a Google Cloud project.
-
Create a Google Cloud project:
gcloud projects create PROJECT_ID
Replace
PROJECT_ID
with a name for the Google Cloud project you are creating. -
Select the Google Cloud project that you created:
gcloud config set project PROJECT_ID
Replace
PROJECT_ID
with your Google Cloud project name.
-
-
Make sure that billing is enabled for your Google Cloud project.
-
Enable the GKE, Backup for GKE, Compute Engine, Identity and Access Management, and Resource Manager APIs:
gcloud services enable compute.googleapis.com
iam.googleapis.com container.googleapis.com gkebackup.googleapis.com cloudresourcemanager.googleapis.com - Install the Google Cloud CLI.
-
To initialize the gcloud CLI, run the following command:
gcloud init
-
Create or select a Google Cloud project.
-
Create a Google Cloud project:
gcloud projects create PROJECT_ID
Replace
PROJECT_ID
with a name for the Google Cloud project you are creating. -
Select the Google Cloud project that you created:
gcloud config set project PROJECT_ID
Replace
PROJECT_ID
with your Google Cloud project name.
-
-
Make sure that billing is enabled for your Google Cloud project.
-
Enable the GKE, Backup for GKE, Compute Engine, Identity and Access Management, and Resource Manager APIs:
gcloud services enable compute.googleapis.com
iam.googleapis.com container.googleapis.com gkebackup.googleapis.com cloudresourcemanager.googleapis.com -
Grant roles to your user account. Run the following command once for each of the following IAM roles:
role/storage.objectViewer, role/logging.logWriter, roles/container.clusterAdmin, role/container.serviceAgent, roles/iam.serviceAccountAdmin, roles/serviceusage.serviceUsageAdmin, roles/iam.serviceAccountAdmin
gcloud projects add-iam-policy-binding PROJECT_ID --member="user:USER_IDENTIFIER" --role=ROLE
- Replace
PROJECT_ID
with your project ID. -
Replace
USER_IDENTIFIER
with the identifier for your user account. For example,user:[email protected]
. - Replace
ROLE
with each individual role.
- Replace
Prepare the environment
In this tutorial, you use
Cloud Shell
to manage resources hosted
on Google Cloud. Cloud Shell is preinstalled with the software
you need for this tutorial, including
kubectl
,
the
gcloud CLI,
Helm,
and
Terraform.
To set up your environment with Cloud Shell, follow these steps:
Launch a Cloud Shell session from the Google Cloud console, by clicking Activate Cloud Shell in the Google Cloud console. This launches a session in the bottom pane of the Google Cloud console.
Set environment variables:
export PROJECT_ID=PROJECT_ID export KUBERNETES_CLUSTER_PREFIX=kafka export REGION=us-central1
Replace
PROJECT_ID
: your Google Cloud with your project ID.Clone the GitHub repository:
git clone https://github.com/GoogleCloudPlatform/kubernetes-engine-samples
Change to the working directory:
cd kubernetes-engine-samples/streaming
Create your cluster infrastructure
In this section, you run a Terraform script to create a private, highly-available, regional GKE cluster. The following steps allow public access to the control plane. To restrict access, create a private cluster.
You can install the operator using a Standard or Autopilot cluster.
Standard
The following diagram shows a private regional Standard GKE cluster deployed across three different zones:
To deploy this infrastructure, run the following commands from the Cloud Shell:
export GOOGLE_OAUTH_ACCESS_TOKEN=$(gcloud auth print-access-token)
terraform -chdir=kafka/terraform/gke-standard init
terraform -chdir=kafka/terraform/gke-standard apply -var project_id=${PROJECT_ID} \
-var region=${REGION} \
-var cluster_prefix=${KUBERNETES_CLUSTER_PREFIX}
When prompted, type yes
. It might take several minutes for this command
to complete and for the cluster to show a ready status.
Terraform creates the following resources:
- A VPC network and private subnet for the Kubernetes nodes.
- A router to access the internet through NAT.
- A private GKE cluster in the
us-central1
region. - 2 node pools with autoscaling enabled (1-2 nodes per zone, 1 node per zone minimum)
- A
ServiceAccount
with logging and monitoring permissions. - Backup for GKE for disaster recovery.
- Google Cloud Managed Service for Prometheus for cluster monitoring.
The output is similar to the following:
...
Apply complete! Resources: 14 added, 0 changed, 0 destroyed.
Outputs:
kubectl_connection_command = "gcloud container clusters get-credentials kafka-cluster --region us-central1"
Autopilot
The following diagram shows a private regional Autopilot GKE cluster:
To deploy the infrastructure, run the following commands from the Cloud Shell:
export GOOGLE_OAUTH_ACCESS_TOKEN=$(gcloud auth print-access-token)
terraform -chdir=kafka/terraform/gke-autopilot init
terraform -chdir=kafka/terraform/gke-autopilot apply -var project_id=${PROJECT_ID} \
-var region=${REGION} \
-var cluster_prefix=${KUBERNETES_CLUSTER_PREFIX}
When prompted, type yes
. It might take several minutes for this command
to complete and for the cluster to show a ready status.
Terraform creates the following resources:
- VPC network and private subnet for the Kubernetes nodes.
- A router to access the internet through NAT.
- A private GKE cluster in the
us-central1
region. - A
ServiceAccount
with logging and monitoring permissions - Google Cloud Managed Service for Prometheus for cluster monitoring.
The output is similar to the following:
...
Apply complete! Resources: 12 added, 0 changed, 0 destroyed.
Outputs:
kubectl_connection_command = "gcloud container clusters get-credentials kafka-cluster --region us-central1"
Connect to the cluster
Configure kubectl
to communicate with the cluster:
gcloud container clusters get-credentials ${KUBERNETES_CLUSTER_PREFIX}-cluster --region ${REGION}
Deploy the CFK operator to your cluster
In this section, you deploy the Confluent for Kubernetes (CFK) operator using a Helm chart and then deploy a Kafka cluster.
Add the Confluent Helm Chart repository:
helm repo add confluentinc https://packages.confluent.io/helm
Add a namespace for the CFK operator and the Kafka cluster:
kubectl create ns kafka
Deploy the CFK cluster operator using Helm:
helm install confluent-operator confluentinc/confluent-for-kubernetes -n kafka
To enable CFK to manage resources across all namespaces, add the parameter
--set-namespaced=false
to the Helm command.Verify that the Confluent operator has been deployed successfully using Helm:
helm ls -n kafka
The output is similar to the following:
NAME NAMESPACE REVISION UPDATED STATUS CHART APP VERSION confluent-operator kafka 1 2023-07-07 10:57:45.409158 +0200 CEST deployed confluent-for-kubernetes-0.771.13 2.6.0
Deploy Kafka
In this section, you deploy Kafka in a basic configuration and then try various advanced configuration scenarios to address availability, security, and observability requirements.
Basic configuration
The basic configuration for the Kafka instance includes the following components:
- Three replicas of Kafka brokers, with a minimum of two available replicas required for cluster consistency.
- Three replicas of ZooKeeper nodes, forming a cluster.
- Two Kafka listeners: one without authentication, and one utilizing TLS authentication with a certificate generated by CFK.
- Java MaxHeapSize and MinHeapSize set to 4 GB for Kafka.
- CPU resource allocation of 1 CPU request and 2 CPU limits, and 5 GB memory requests and limits for Kafka (4 GB for the main service and 0.5 GB for the metrics exporter) and 3 GB for Zookeeper (2 GB for the main service and 0.5 GB for the metrics exporter).
- 100 GB of storage allocated to each Pod using the
premium-rwo
storageClass, 100 for Kafka Data and 90/10 for Zookeeper Data/Log. - Tolerations, nodeAffinities, and podAntiAffinities configured for each workload, ensuring proper distribution across nodes, utilizing their respective node pools and different zones.
- Communication inside the cluster secured by self-signed certificates using a Certificate Authority that you provide.
This configuration represents the minimal setup required to create a production-ready Kafka cluster. The following sections demonstrate custom configurations to address aspects such as cluster security, Access Control Lists (ACLs), topic management, certificate management and more.
Create a basic Kafka cluster
Generate a CA pair:
openssl genrsa -out ca-key.pem 2048 openssl req -new -key ca-key.pem -x509 \ -days 1000 \ -out ca.pem \ -subj "/C=US/ST=CA/L=Confluent/O=Confluent/OU=Operator/CN=MyCA"
Confluent for Kubernetes provides auto-generated certificates for Confluent Platform components to use for TLS network encryption. You must generate and provide a Certificate Authority (CA).
Create a Kubernetes Secret for the certificate authority:
kubectl create secret tls ca-pair-sslcerts --cert=ca.pem --key=ca-key.pem -n kafka
The name of the Secret is predefined
Create a new Kafka cluster using the basic configuration:
kubectl apply -n kafka -f kafka-confluent/manifests/01-basic-cluster/my-cluster.yaml
This command creates a Kafka custom resource and Zookeeper custom resource of the CFK operator that include CPU and memory requests and limits, block storage requests, and taints and affinities to distribute the provisioned Pods across Kubernetes nodes.
Wait a few minutes while Kubernetes starts the required workloads:
kubectl wait pods -l app=my-cluster --for condition=Ready --timeout=300s -n kafka
Verify that the Kafka workloads were created:
kubectl get pod,svc,statefulset,deploy,pdb -n kafka
The output is similar to the following:
NAME READY STATUS RESTARTS AGE pod/confluent-operator-864c74d4b4-fvpxs 1/1 Running 0 49m pod/my-cluster-0 1/1 Running 0 17m pod/my-cluster-1 1/1 Running 0 17m pod/my-cluster-2 1/1 Running 0 17m pod/zookeeper-0 1/1 Running 0 18m pod/zookeeper-1 1/1 Running 0 18m pod/zookeeper-2 1/1 Running 0 18m NAME TYPE CLUSTER-IP EXTERNAL-IP PORT(S) AGE service/confluent-operator ClusterIP 10.52.13.164 <none> 7778/TCP 49m service/my-cluster ClusterIP None <none> 9092/TCP,8090/TCP,9071/TCP,7203/TCP,7777/TCP,7778/TCP,9072/TCP 17m service/my-cluster-0-internal ClusterIP 10.52.2.242 <none> 9092/TCP,8090/TCP,9071/TCP,7203/TCP,7777/TCP,7778/TCP,9072/TCP 17m service/my-cluster-1-internal ClusterIP 10.52.7.98 <none> 9092/TCP,8090/TCP,9071/TCP,7203/TCP,7777/TCP,7778/TCP,9072/TCP 17m service/my-cluster-2-internal ClusterIP 10.52.4.226 <none> 9092/TCP,8090/TCP,9071/TCP,7203/TCP,7777/TCP,7778/TCP,9072/TCP 17m service/zookeeper ClusterIP None <none> 2181/TCP,7203/TCP,7777/TCP,3888/TCP,2888/TCP,7778/TCP 18m service/zookeeper-0-internal ClusterIP 10.52.8.52 <none> 2181/TCP,7203/TCP,7777/TCP,3888/TCP,2888/TCP,7778/TCP 18m service/zookeeper-1-internal ClusterIP 10.52.12.44 <none> 2181/TCP,7203/TCP,7777/TCP,3888/TCP,2888/TCP,7778/TCP 18m service/zookeeper-2-internal ClusterIP 10.52.12.134 <none> 2181/TCP,7203/TCP,7777/TCP,3888/TCP,2888/TCP,7778/TCP 18m NAME READY AGE statefulset.apps/my-cluster 3/3 17m statefulset.apps/zookeeper 3/3 18m NAME READY UP-TO-DATE AVAILABLE AGE deployment.apps/confluent-operator 1/1 1 1 49m NAME MIN AVAILABLE MAX UNAVAILABLE ALLOWED DISRUPTIONS AGE poddisruptionbudget.policy/my-cluster N/A 1 1 17m poddisruptionbudget.policy/zookeeper N/A 1 1 18m
The operator creates the following resources:
- Two StatefulSets for Kafka and ZooKeeper.
- Three Pods for Kafka broker replicas.
- Three Pods for ZooKeeper replicas.
- Two
PodDisruptionBudget
resources, ensuring a maximum one unavailable replica for cluster consistency. - The Service
my-cluster
which serves as the bootstrap server for Kafka clients connecting from within the Kubernetes cluster. All internal Kafka listeners are available in this Service. - The Service
zookeeper
which allows Kafka brokers to connect to ZooKeeper nodes as clients.
Authentication and user management
This section shows you how to enable the authentication and authorization to secure Kafka Listeners and share credentials with clients.
Confluent for Kubernetes supports various authentication methods for Kafka, such as:
- SASL/PLAIN authentication: Clients use a username and password for authentication. The username and password are stored server-side in a Kubernetes secret.
- SASL/PLAIN with LDAP authentication: Clients use a username and password for authentication. The credentials are stored in an LDAP server.
- mTLS authentication: Clients use TLS certificates for authentication.
Limitations
- CFK does not provide Custom Resources for user management. However, you can store credentials in Secrets and refer to Secrets to in listener specs.
- Although there's no Custom Resource to manage ACLs directly, the official Confluent for Kubernetes provides guidance on configuring ACLs using the Kafka CLI.
Create a user
This section shows you how to deploy a CFK operator that demonstrates user management capabilities, including:
- A Kafka cluster with password-based authentication (SASL/PLAIN) enabled on one of the listeners
- A
KafkaTopic
with 3 replicas - User credentials with read and write permissions
Create a Secret with user credentials:
export USERNAME=my-user export PASSWORD=$(openssl rand -base64 12) kubectl create secret generic my-user-credentials -n kafka \ --from-literal=plain-users.json="{\"$USERNAME\":\"$PASSWORD\"}"
Credentials should be stored in the following format:
{ "username1": "password1", "username2": "password2", ... "usernameN": "passwordN" }
Configure Kafka cluster to use a listener with password-based authentication SCRAM-SHA-512 authentication on port 9094:
kubectl apply -n kafka -f kafka-confluent/manifests/02-auth/my-cluster.yaml
Set up a topic and a client Pod to interact with your Kafka cluster and execute Kafka commands:
kubectl apply -n kafka -f kafka-confluent/manifests/02-auth/my-topic.yaml kubectl apply -n kafka -f kafka-confluent/manifests/02-auth/kafkacat.yaml
GKE mounts the Secret
my-user-credentials
to the client Pod as a Volume.When the client Pod is ready, connect to it and start producing and consuming messages using the provided credentials:
kubectl wait pod kafkacat --for=condition=Ready --timeout=300s -n kafka kubectl exec -it kafkacat -n kafka -- /bin/sh
Produce a message using the
my-user
credentials and then consume the message to verify its receipt.export USERNAME=$(cat /my-user/plain-users.json|cut -d'"' -f 2) export PASSWORD=$(cat /my-user/plain-users.json|cut -d'"' -f 4) echo "Message from my-user" |kcat \ -b my-cluster.kafka.svc.cluster.local:9094 \ -X security.protocol=SASL_SSL \ -X sasl.mechanisms=PLAIN \ -X sasl.username=$USERNAME \ -X sasl.password=$PASSWORD \ -t my-topic -P kcat -b my-cluster.kafka.svc.cluster.local:9094 \ -X security.protocol=SASL_SSL \ -X sasl.mechanisms=PLAIN \ -X sasl.username=$USERNAME \ -X sasl.password=$PASSWORD \ -t my-topic -C
The output is similar to the following:
Message from my-user % Reached end of topic my-topic [1] at offset 1 % Reached end of topic my-topic [2] at offset 0 % Reached end of topic my-topic [0] at offset 0
Type
CTRL+C
to stop the consumer process. If you get aConnect refused
error, wait a few minutes and then try again.Exit the Pod shell
exit
Backups and disaster recovery
Using the Confluent operator, you can implement efficient backup strategies by following certain patterns.
You can use Backup for GKE to backup:
- Kubernetes resource manifests.
- Confluent API custom resources and their definitions extracted from the Kubernetes API server of the cluster undergoing backup.
- Volumes that correspond to PersistentVolumeClaim resources found in the manifests.
For more information about how to backup and restore Kafka clusters using Backup for GKE, see Prepare for disaster recovery.
You can also perform a manual backup of your Kafka cluster. You should backup:
- The Kafka configuration, which includes all custom resources of the
Confluent API such as
KafkaTopics
orConnect
- The data, which is stored in the PersistentVolumes of the Kafka brokers
Storing Kubernetes resource manifests, including Confluent configurations, in Git repositories can eliminate the need for a separate backup for Kafka configuration as the resources can be reapplied to a new Kubernetes cluster when necessary.
To safeguard Kafka data recovery in scenarios where a Kafka server instance, or
Kubernetes cluster where Kafka is deployed, is lost, we recommend that you
configure the Kubernetes storage class used for provisioning volumes for Kafka
brokers with the reclaimPolicy
option set to Retain
. We also recommended
that you take
snapshots
of Kafka broker volumes.
The following manifest describes a StorageClass that uses the reclaimPolicy
option Retain
:
apiVersion: storage.k8s.io/v1
kind: StorageClass
metadata:
name: premium-rwo-retain
...
reclaimPolicy: Retain
volumeBindingMode: WaitForFirstConsumer
The following example shows the StorageClass added to the spec
of a Kafka
cluster custom resource:
...
spec:
...
dataVolumeCapacity: 100Gi
storageClass:
name: premium-rwo-retain
With this configuration, PersistentVolumes provisioned using the storage class are not deleted even when the corresponding PersistentVolumeClaim is deleted.
To recover the Kafka instance on a new Kubernetes cluster using the existing configuration and broker instance data:
- Apply the existing Confluent custom resources (
Kafka
,KafkaTopic
,Zookeeper
, etc.) to a new Kubernetes cluster - Update the PersistentVolumeClaims with the name of the new Kafka broker
instances to the old PersistentVolumes using the
spec.volumeName
property on the PersistentVolumeClaim.
Clean up
To avoid incurring charges to your Google Cloud account for the resources used in this tutorial, either delete the project that contains the resources, or keep the project and delete the individual resources.
Delete the project
Delete a Google Cloud project:
gcloud projects delete PROJECT_ID
Delete the individual resources
If you used an existing project and you don't want to delete it, delete the individual resources.
Set environment variables:
export PROJECT_ID=PROJECT_ID export KUBERNETES_CLUSTER_PREFIX=kafka export REGION=us-central1
Run the
terraform destroy
command:export GOOGLE_OAUTH_ACCESS_TOKEN=$(gcloud auth print-access-token) terraform -chdir=kafka/terraform/FOLDER destroy -var project_id=${PROJECT_ID} \ -var region=${REGION} \ -var cluster_prefix=${KUBERNETES_CLUSTER_PREFIX}
Replace
FOLDER
with eithergke-autopilot
orgke-standard
.When prompted, type
yes
.Find all unattached disks:
export disk_list=$(gcloud compute disks list --filter="-users:* AND labels.name=${KUBERNETES_CLUSTER_PREFIX}-cluster" --format "value[separator=|](name,zone)")
Delete the disks:
for i in $disk_list; do disk_name=$(echo $i| cut -d'|' -f1) disk_zone=$(echo $i| cut -d'|' -f2|sed 's|.*/||') echo "Deleting $disk_name" gcloud compute disks delete $disk_name --zone $disk_zone --quiet done
What's next
- Explore reference architectures, diagrams, and best practices about Google Cloud. Take a look at our Cloud Architecture Center.