1. Deployment overview

Strimzi simplifies the process of running Apache Kafka in a Kubernetes cluster.

This guide provides instructions on all the options available for deploying and upgrading Strimzi, describing what is deployed, and the order of deployment required to run Apache Kafka in a Kubernetes cluster.

As well as describing the deployment steps, the guide also provides pre- and post-deployment instructions to prepare for and verify a deployment. The guide also describes additional deployment options for introducing metrics.

Upgrade instructions are provided for Strimzi and Kafka upgrades.

Strimzi is designed to work on all types of Kubernetes cluster regardless of distribution, from public and private clouds to local deployments intended for development.

1.1. Configuring a deployment

The deployment procedures in this guide are designed to help you set up the initial structure of your deployment. After setting up the structure, you can use custom resources to configure the deployment to your precise needs. The deployment procedures use the example installation files provided with Strimzi. The procedures highlight any important configuration considerations, but they do not describe all the configuration options available.

You might want to review the configuration options available for Kafka components before you deploy Strimzi. For more information on the configuration options, see Configuring Strimzi.

1.1.1. Securing Kafka

On deployment, the Cluster Operator automatically sets up TLS certificates for data encryption and authentication within your cluster.

Strimzi provides additional configuration options for encryption, authentication and authorization:

1.1.2. Monitoring a deployment

Strimzi supports additional deployment options to monitor your deployment.

1.1.3. CPU and memory resource limits and requests

By default, the Strimzi Cluster Operator does not specify requests and limits for CPU and memory resources for any operands it deploys.

Having sufficient resources is important for applications like Kafka to be stable and deliver good performance.

The right amount of resources you should use depends on the specific requirements and use-cases.

You should consider configuring the CPU and memory resources. You can set resource requests and limits for each container in the Strimzi custom resources.

1.2. Strimzi custom resources

A deployment of Kafka components to a Kubernetes cluster using Strimzi is highly configurable through the application of custom resources. Custom resources are created as instances of APIs added by Custom resource definitions (CRDs) to extend Kubernetes resources.

CRDs act as configuration instructions to describe the custom resources in a Kubernetes cluster, and are provided with Strimzi for each Kafka component used in a deployment, as well as users and topics. CRDs and custom resources are defined as YAML files. Example YAML files are provided with the Strimzi distribution.

CRDs also allow Strimzi resources to benefit from native Kubernetes features like CLI accessibility and configuration validation.

1.2.1. Strimzi custom resource example

CRDs require a one-time installation in a cluster to define the schemas used to instantiate and manage Strimzi-specific resources.

After a new custom resource type is added to your cluster by installing a CRD, you can create instances of the resource based on its specification.

Depending on the cluster setup, installation typically requires cluster admin privileges.

Note
 Access to manage custom resources is limited to Strimzi administrators. For more information, see Designating Strimzi administrators.

A CRD defines a new kind of resource, such as kind:Kafka, within a Kubernetes cluster.

The Kubernetes API server allows custom resources to be created based on the kind and understands from the CRD how to validate and store the custom resource when it is added to the Kubernetes cluster.

Warning
 When CRDs are deleted, custom resources of that type are also deleted. Additionally, the resources created by the custom resource, such as pods and statefulsets are also deleted.

Each Strimzi-specific custom resource conforms to the schema defined by the CRD for the resource’s kind. The custom resources for Strimzi components have common configuration properties, which are defined under spec.

To understand the relationship between a CRD and a custom resource, let’s look at a sample of the CRD for a Kafka topic.

Kafka topic CRD
apiVersion: kafka.strimzi.io/v1beta2
kind: CustomResourceDefinition
metadata: (1)
  name: kafkatopics.kafka.strimzi.io
  labels:
    app: strimzi
spec: (2)
  group: kafka.strimzi.io
  versions:
    v1beta2
  scope: Namespaced
  names:
    # ...
    singular: kafkatopic
    plural: kafkatopics
    shortNames:
    - kt (3)
  additionalPrinterColumns: (4)
      # ...
  subresources:
    status: {} (5)
  validation: (6)
    openAPIV3Schema:
      properties:
        spec:
          type: object
          properties:
            partitions:
              type: integer
              minimum: 1
            replicas:
              type: integer
              minimum: 1
              maximum: 32767
      # ...

  1. The metadata for the topic CRD, its name and a label to identify the CRD.

  2. The specification for this CRD, including the group (domain) name, the plural name and the supported schema version, which are used in the URL to access the API of the topic. The other names are used to identify instance resources in the CLI. For example, kubectl get kafkatopic my-topic or kubectl get kafkatopics.

  3. The shortname can be used in CLI commands. For example, kubectl get kt can be used as an abbreviation instead of kubectl get kafkatopic.

  4. The information presented when using a get command on the custom resource.

  5. The current status of the CRD as described in the schema reference for the resource.

  6. openAPIV3Schema validation provides validation for the creation of topic custom resources. For example, a topic requires at least one partition and one replica.

Note
 You can identify the CRD YAML files supplied with the Strimzi installation files, because the file names contain an index number followed by ‘Crd’.

Here is a corresponding example of a KafkaTopic custom resource.

Kafka topic custom resource
apiVersion: kafka.strimzi.io/v1beta2
kind: KafkaTopic (1)
metadata:
  name: my-topic
  labels:
    strimzi.io/cluster: my-cluster (2)
spec: (3)
  partitions: 1
  replicas: 1
  config:
    retention.ms: 7200000
    segment.bytes: 1073741824
status:
  conditions: (4)
    lastTransitionTime: "2019-08-20T11:37:00.706Z"
    status: "True"
    type: Ready
  observedGeneration: 1
  / ...

  1. The kind and apiVersion identify the CRD of which the custom resource is an instance.

  2. A label, applicable only to KafkaTopic and KafkaUser resources, that defines the name of the Kafka cluster (which is same as the name of the Kafka resource) to which a topic or user belongs.

  3. The spec shows the number of partitions and replicas for the topic as well as the configuration parameters for the topic itself. In this example, the retention period for a message to remain in the topic and the segment file size for the log are specified.

  4. Status conditions for the KafkaTopic resource. The type condition changed to Ready at the lastTransitionTime.

Custom resources can be applied to a cluster through the platform CLI. When the custom resource is created, it uses the same validation as the built-in resources of the Kubernetes API.

After a KafkaTopic custom resource is created, the Topic Operator is notified and corresponding Kafka topics are created in Strimzi.

Additional resources

1.3. Using the Kafka Bridge to connect with a Kafka cluster

You can use the Strimzi Kafka Bridge API to create and manage consumers and send and receive records over HTTP rather than the native Kafka protocol.

When you set up the Kafka Bridge you configure HTTP access to the Kafka cluster. You can then use the Kafka Bridge to produce and consume messages from the cluster, as well as performing other operations through its REST interface.

Additional resources

1.4. Document Conventions

User-replaced values

User-replaced values, also known as replaceables, are shown in italics with angle brackets (< >). Underscores ( _ ) are used for multi-word values. If the value refers to code or commands, monospace is also used.

For example, in the following code, you will want to replace <my_namespace> with the name of your namespace:

sed -i 's/namespace: .*/namespace: <my_namespace>/' install/cluster-operator/*RoleBinding*.yaml

1.5. Additional resources

2. Strimzi installation methods

You can install Strimzi on Kubernetes 1.19 and later in three ways.

Installation method Description

Installation artifacts (YAML files)

Download the release artifacts from the GitHub releases page.

Download the strimzi-<version>.zip or strimzi-<version>.tar.gz archive file. The archive file contains installation artifacts and example configuration files.

Deploy the YAML installation artifacts to your Kubernetes cluster using kubectl. You start by deploying the Cluster Operator from install/cluster-operator to a single namespace, multiple namespaces, or all namespaces.

You can also use the install/ artifacts to deploy the following:

  • Strimi administrator roles (strimzi-admin)

  • A standalone Topic Operator (topic-operator)

  • A standalone User Operator (user-operator)

  • Strimzi Drain Cleaner (drain-cleaner)

OperatorHub.io

Use the Strimzi Kafka operator in the OperatorHub.io to deploy the Cluster Operator. You then deploy Strimzi components using custom resources.

Helm chart

Use a Helm chart to deploy the Cluster Operator. You then deploy Strimzi components using custom resources.

For the greatest flexibility, choose the installation artifacts method. The OperatorHub.io method provides a standard configuration and allows you to take advantage of automatic updates. Helm charts provide a convenient way to manage the installation of applications.

3. What is deployed with Strimzi

Apache Kafka components are provided for deployment to Kubernetes with the Strimzi distribution. The Kafka components are generally run as clusters for availability.

A typical deployment incorporating Kafka components might include:

  • Kafka cluster of broker nodes

  • ZooKeeper cluster of replicated ZooKeeper instances

  • Kafka Connect cluster for external data connections

  • Kafka MirrorMaker cluster to mirror the Kafka cluster in a secondary cluster

  • Kafka Exporter to extract additional Kafka metrics data for monitoring

  • Kafka Bridge to make HTTP-based requests to the Kafka cluster

Not all of these components are mandatory, though you need Kafka and ZooKeeper as a minimum. Some components can be deployed without Kafka, such as MirrorMaker or Kafka Connect.

3.1. Order of deployment

The required order of deployment to a Kubernetes cluster is as follows:

  1. Deploy the Cluster Operator to manage your Kafka cluster

  2. Deploy the Kafka cluster with the ZooKeeper cluster, and include the Topic Operator and User Operator in the deployment

  3. Optionally deploy:

    • The Topic Operator and User Operator standalone if you did not deploy them with the Kafka cluster

    • Kafka Connect

    • Kafka MirrorMaker

    • Kafka Bridge

    • Components for the monitoring of metrics

The Cluster Operator creates Kubernetes resources for the components, such as Deployment, Service, and Pod resources. The names of the Kubernetes resources are appended with the name specified for a component when it’s deployed. For example, a Kafka cluster named my-kafka-cluster has a service named my-kafka-cluster-kafka.

4. Preparing for your Strimzi deployment

This section shows how you prepare for a Strimzi deployment, describing:

Note
 To run the commands in this guide, your cluster user must have the rights to manage role-based access control (RBAC) and CRDs.

4.1. Deployment prerequisites

To deploy Strimzi, you will need the following:

  • A Kubernetes 1.19 and later cluster.

    If you do not have access to a Kubernetes cluster, you can install Strimzi with Minikube.

  • The kubectl command-line tool is installed and configured to connect to the running cluster.

Note
 Strimzi supports some features that are specific to OpenShift, where such integration benefits OpenShift users and there is no equivalent implementation using standard Kubernetes.

oc and kubectl commands

The oc command functions as an alternative to kubectl. In almost all cases the example kubectl commands used in this guide can be done using oc simply by replacing the command name (options and arguments remain the same).

In other words, instead of using:

kubectl apply -f your-file

when using OpenShift you can use:

oc apply -f your-file
Note
 As an exception to this general rule, oc uses oc adm subcommands for cluster management functionality, whereas kubectl does not make this distinction. For example, the oc equivalent of kubectl taint is oc adm taint.

4.2. Downloading Strimzi release artifacts

To use deployment files to install Strimzi, download and extract the files from the GitHub releases page.

Strimzi release artifacts include sample YAML files to help you deploy the components of Strimzi to Kubernetes, perform common operations, and configure your Kafka cluster.

Use kubectl to deploy the Cluster Operator from the install/cluster-operator folder of the downloaded ZIP file. For more information about deploying and configuring the Cluster Operator, see Deploying the Cluster Operator.

In addition, if you want to use standalone installations of the Topic and User Operators with a Kafka cluster that is not managed by the Strimzi Cluster Operator, you can deploy them from the install/topic-operator and install/user-operator folders.

Note
 Additionally, Strimzi container images are available through the Container Registry. However, we recommend that you use the YAML files provided to deploy Strimzi.

4.3. Example configuration and deployment files

Use the example configuration and deployment files provided with Strimzi to deploy Kafka components with different configurations and monitor your deployment. Example configuration files for custom resources contain important properties and values, which you can extend with additional supported configuration properties for your own deployment.

4.3.1. Example files location

The example files are provided with the downloadable release artifacts from the GitHub releases page.

You can also access the example files directly from the examples directory.

You can download and apply the examples using the kubectl command-line tool. The examples can serve as a starting point when building your own Kafka component configuration for deployment.

Note
 If you installed Strimzi using the Operator, you can still download the example files and use them to upload configuration.

4.3.2. Example files provided with Strimzi

The release artifacts include an examples directory that contains the configuration examples.

Examples directory
examples
├── user (1)
├── topic (2)
├── security (3)
│   ├── tls-auth
│   ├── scram-sha-512-auth
│   └── keycloak-authorization
├── mirror-maker (4)
├── metrics (5)
├── kafka (6)
├── cruise-control (7)
├── connect (8)
└── bridge (9)

  1. KafkaUser custom resource configuration, which is managed by the User Operator.

  2. KafkaTopic custom resource configuration, which is managed by Topic Operator.

  3. Authentication and authorization configuration for Kafka components. Includes example configuration for TLS and SCRAM-SHA-512 authentication. The Keycloak example includes Kafka custom resource configuration and a Keycloak realm specification. You can use the example to try Keycloak authorization services. There is also an example with enabled oauth authentication and keycloak authorization metrics.

  4. Kafka custom resource configuration for a deployment of Mirror Maker. Includes example configuration for replication policy and synchronization frequency.

  5. Metrics configuration, including Prometheus installation and Grafana dashboard files.

  6. Kafka custom resource configuration for a deployment of Kafka. Includes example configuration for an ephemeral or persistent single or multi-node deployment.

  7. Kafka custom resource with a deployment configuration for Cruise Control. Includes KafkaRebalance custom resources to generate optimizations proposals from Cruise Control, with example configurations to use the default or user optimization goals.

  8. KafkaConnect and KafkaConnector custom resource configuration for a deployment of Kafka Connect. Includes example configuration for a single or multi-node deployment.

  9. KafkaBridge custom resource configuration for a deployment of Kafka Bridge.

Additional resources

4.4. Pushing container images to your own registry

Container images for Strimzi are available in the Container Registry. The installation YAML files provided by Strimzi will pull the images directly from the Container Registry.

If you do not have access to the Container Registry or want to use your own container repository:

  1. Pull all container images listed here

  2. Push them into your own registry

  3. Update the image names in the YAML files used in deployment

Note
 Each Kafka version supported for the release has a separate image.
Container image Namespace/Repository Description

Kafka

  • quay.io/strimzi/kafka:latest-kafka-3.2.0

  • quay.io/strimzi/kafka:latest-kafka-3.2.1

  • quay.io/strimzi/kafka:latest-kafka-3.2.3

  • quay.io/strimzi/kafka:latest-kafka-3.3.1

  • quay.io/strimzi/kafka:latest-kafka-3.3.2

Strimzi image for running Kafka, including:

  • Kafka Broker

  • Kafka Connect

  • Kafka MirrorMaker

  • ZooKeeper

  • TLS Sidecars

Operator

  • quay.io/strimzi/operator:latest

Strimzi image for running the operators:

  • Cluster Operator

  • Topic Operator

  • User Operator

  • Kafka Initializer

Kafka Bridge

  • quay.io/strimzi/kafka-bridge:0.24.0

Strimzi image for running the Strimzi kafka Bridge

Strimzi Drain Cleaner

  • quay.io/strimzi/drain-cleaner:0.3.1

Strimzi image for running the Strimzi Drain Cleaner

JmxTrans

  • quay.io/strimzi/jmxtrans:latest

Strimzi image for running the Strimzi JmxTrans

4.5. Designating Strimzi administrators

Strimzi provides custom resources for configuration of your deployment. By default, permission to view, create, edit, and delete these resources is limited to Kubernetes cluster administrators. Strimzi provides two cluster roles that you can use to assign these rights to other users:

  • strimzi-view allows users to view and list Strimzi resources.

  • strimzi-admin allows users to also create, edit or delete Strimzi resources.

When you install these roles, they will automatically aggregate (add) these rights to the default Kubernetes cluster roles. strimzi-view aggregates to the view role, and strimzi-admin aggregates to the edit and admin roles. Because of the aggregation, you might not need to assign these roles to users who already have similar rights.

The following procedure shows how to assign a strimzi-admin role that allows non-cluster administrators to manage Strimzi resources.

A system administrator can designate Strimzi administrators after the Cluster Operator is deployed.

Prerequisites
Procedure

  1. Create the strimzi-view and strimzi-admin cluster roles in Kubernetes.

    kubectl create -f install/strimzi-admin

  2. If needed, assign the roles that provide access rights to users that require them.

    kubectl create clusterrolebinding strimzi-admin --clusterrole=strimzi-admin --user=user1 --user=user2

4.6. Installing a local Kubernetes cluster with Minikube

Minikube offers an easy way to get started with Kubernetes. If a Kubernetes cluster is unavailable, you can use Minikube to create a local cluster.

You can download and install Minikube from the Kubernetes website, which also provides documentation. Depending on the number of brokers you want to deploy inside the cluster, and whether you want to run Kafka Connect as well, try running Minikube with at least with 4 GB of RAM instead of the default 2 GB.

Once installed, start Minikube using:

minikube start --memory 4096

To interact with the cluster, install the kubectl utility.

5. Deploying Strimzi using installation artifacts

Having prepared your environment for a deployment of Strimzi, you can deploy Strimzi to a Kubernetes cluster. Use the installation files provided with the release artifacts.

You can deploy Strimzi latest on Kubernetes 1.19 and later.

The steps to deploy Strimzi using the installation files are as follows:

  1. Deploy the Cluster Operator

  2. Use the Cluster Operator to deploy the following:

    1. Kafka cluster

    2. Topic Operator

    3. User Operator

  3. Optionally, deploy the following Kafka components according to your requirements:

Note
 To run the commands in this guide, a Kubernetes user must have the rights to manage role-based access control (RBAC) and CRDs.

5.1. Basic deployment path

You can set up a deployment where Strimzi manages a single Kafka cluster in the same namespace. You might use this configuration for development or testing. Or you can use Strimzi in a production environment to manage a number of Kafka clusters in different namespaces.

The first step for any deployment of Strimzi is to install the Cluster Operator using the install/cluster-operator files.

A single command applies all the installation files in the cluster-operator folder: kubectl apply -f ./install/cluster-operator.

The command sets up everything you need to be able to create and manage a Kafka deployment, including the following:

  • Cluster Operator (Deployment, ConfigMap)

  • Strimzi CRDs (CustomResourceDefinition)

  • RBAC resources (ClusterRole, ClusterRoleBinding, RoleBinding)

  • Service account (ServiceAccount)

The basic deployment path is as follows:

  1. Download the release artifacts

  2. Create a Kubernetes namespace in which to deploy the Cluster Operator

  3. Deploy the Cluster Operator

    1. Update the install/cluster-operator files to use the namespace created for the Cluster Operator

    2. Install the Cluster Operator to watch one, multiple, or all namespaces

  4. Create a Kafka cluster

After which, you can deploy other Kafka components and set up monitoring of your deployment.

5.2. Deploying the Cluster Operator

The Cluster Operator is responsible for deploying and managing Kafka clusters within a Kubernetes cluster.

When the Cluster Operator is running, it starts to watch for updates of Kafka resources.

By default, a single replica of the Cluster Operator is deployed. You can add replicas with leader election so that additional Cluster Operators are on standby in case of disruption. For more information, see Running multiple Cluster Operator replicas with leader election.

5.2.1. Specifying the namespaces the Cluster Operator watches

The Cluster Operator watches for updates in the namespaces where the Kafka resources are deployed. When you deploy the Cluster Operator, you specify which namespaces to watch. You can specify the following namespaces:

Note
 The Cluster Operator can watch one, multiple, or all namespaces in a Kubernetes cluster. The Topic Operator and User Operator watch for KafkaTopic and KafkaUser resources in a single namespace. For more information, see Watching namespaces with Strimzi operators.

The Cluster Operator watches for changes to the following resources:

  • Kafka for the Kafka cluster.

  • KafkaConnect for the Kafka Connect cluster.

  • KafkaConnector for creating and managing connectors in a Kafka Connect cluster.

  • KafkaMirrorMaker for the Kafka MirrorMaker instance.

  • KafkaMirrorMaker2 for the Kafka MirrorMaker 2.0 instance.

  • KafkaBridge for the Kafka Bridge instance.

  • KafkaRebalance for the Cruise Control optimization requests.

When one of these resources is created in the Kubernetes cluster, the operator gets the cluster description from the resource and starts creating a new cluster for the resource by creating the necessary Kubernetes resources, such as StatefulSets, Services and ConfigMaps.

Each time a Kafka resource is updated, the operator performs corresponding updates on the Kubernetes resources that make up the cluster for the resource.

Resources are either patched or deleted, and then recreated in order to make the cluster for the resource reflect the desired state of the cluster. This operation might cause a rolling update that might lead to service disruption.

When a resource is deleted, the operator undeploys the cluster and deletes all related Kubernetes resources.

5.2.2. Deploying the Cluster Operator to watch a single namespace

This procedure shows how to deploy the Cluster Operator to watch Strimzi resources in a single namespace in your Kubernetes cluster.

Prerequisites

  • You need an account with permission to create and manage CustomResourceDefinition and RBAC (ClusterRole, and RoleBinding) resources.

Procedure

  1. Edit the Strimzi installation files to use the namespace the Cluster Operator is going to be installed into.

    For example, in this procedure the Cluster Operator is installed into the namespace my-cluster-operator-namespace.

    On Linux, use:

    sed -i 's/namespace: .*/namespace: my-cluster-operator-namespace/' install/cluster-operator/*RoleBinding*.yaml

    On MacOS, use:

    sed -i '' 's/namespace: .*/namespace: my-cluster-operator-namespace/' install/cluster-operator/*RoleBinding*.yaml

  2. Deploy the Cluster Operator:

    kubectl create -f install/cluster-operator -n my-cluster-operator-namespace

  3. Check the status of the deployment:

    kubectl get deployments -n my-cluster-operator-namespace
    Output shows the deployment name and readiness
    NAME                      READY  UP-TO-DATE  AVAILABLE
strimzi-cluster-operator  1/1    1           1

    READY shows the number of replicas that are ready/expected. The deployment is successful when the AVAILABLE output shows 1.

5.2.3. Deploying the Cluster Operator to watch multiple namespaces

This procedure shows how to deploy the Cluster Operator to watch Strimzi resources across multiple namespaces in your Kubernetes cluster.

Prerequisites

  • You need an account with permission to create and manage CustomResourceDefinition and RBAC (ClusterRole, and RoleBinding) resources.

Procedure

  1. Edit the Strimzi installation files to use the namespace the Cluster Operator is going to be installed into.

    For example, in this procedure the Cluster Operator is installed into the namespace my-cluster-operator-namespace.

    On Linux, use:

    sed -i 's/namespace: .*/namespace: my-cluster-operator-namespace/' install/cluster-operator/*RoleBinding*.yaml

    On MacOS, use:

    sed -i '' 's/namespace: .*/namespace: my-cluster-operator-namespace/' install/cluster-operator/*RoleBinding*.yaml

  2. Edit the install/cluster-operator/060-Deployment-strimzi-cluster-operator.yaml file to add a list of all the namespaces the Cluster Operator will watch to the STRIMZI_NAMESPACE environment variable.

    For example, in this procedure the Cluster Operator will watch the namespaces watched-namespace-1, watched-namespace-2, watched-namespace-3.

    apiVersion: apps/v1
kind: Deployment
spec:
  # ...
  template:
    spec:
      serviceAccountName: strimzi-cluster-operator
      containers:
      - name: strimzi-cluster-operator
        image: quay.io/strimzi/operator:latest
        imagePullPolicy: IfNotPresent
        env:
        - name: STRIMZI_NAMESPACE
          value: watched-namespace-1,watched-namespace-2,watched-namespace-3

  3. For each namespace listed, install the RoleBindings.

    In this example, we replace watched-namespace in these commands with the namespaces listed in the previous step, repeating them for watched-namespace-1, watched-namespace-2, watched-namespace-3:

    kubectl create -f install/cluster-operator/020-RoleBinding-strimzi-cluster-operator.yaml -n <watched_namespace>
kubectl create -f install/cluster-operator/023-RoleBinding-strimzi-cluster-operator.yaml -n <watched_namespace>
kubectl create -f install/cluster-operator/031-RoleBinding-strimzi-cluster-operator-entity-operator-delegation.yaml -n <watched_namespace>

  4. Deploy the Cluster Operator:

    kubectl create -f install/cluster-operator -n my-cluster-operator-namespace

  5. Check the status of the deployment:

    kubectl get deployments -n my-cluster-operator-namespace
    Output shows the deployment name and readiness
    NAME                      READY  UP-TO-DATE  AVAILABLE
strimzi-cluster-operator  1/1    1           1

    READY shows the number of replicas that are ready/expected. The deployment is successful when the AVAILABLE output shows 1.

5.2.4. Deploying the Cluster Operator to watch all namespaces

This procedure shows how to deploy the Cluster Operator to watch Strimzi resources across all namespaces in your Kubernetes cluster.

When running in this mode, the Cluster Operator automatically manages clusters in any new namespaces that are created.

Prerequisites

  • You need an account with permission to create and manage CustomResourceDefinition and RBAC (ClusterRole, and RoleBinding) resources.

Procedure

  1. Edit the Strimzi installation files to use the namespace the Cluster Operator is going to be installed into.

    For example, in this procedure the Cluster Operator is installed into the namespace my-cluster-operator-namespace.

    On Linux, use:

    sed -i 's/namespace: .*/namespace: my-cluster-operator-namespace/' install/cluster-operator/*RoleBinding*.yaml

    On MacOS, use:

    sed -i '' 's/namespace: .*/namespace: my-cluster-operator-namespace/' install/cluster-operator/*RoleBinding*.yaml

  2. Edit the install/cluster-operator/060-Deployment-strimzi-cluster-operator.yaml file to set the value of the STRIMZI_NAMESPACE environment variable to *.

    apiVersion: apps/v1
kind: Deployment
spec:
  # ...
  template:
    spec:
      # ...
      serviceAccountName: strimzi-cluster-operator
      containers:
      - name: strimzi-cluster-operator
        image: quay.io/strimzi/operator:latest
        imagePullPolicy: IfNotPresent
        env:
        - name: STRIMZI_NAMESPACE
          value: "*"
        # ...

  3. Create ClusterRoleBindings that grant cluster-wide access for all namespaces to the Cluster Operator.

    kubectl create clusterrolebinding strimzi-cluster-operator-namespaced --clusterrole=strimzi-cluster-operator-namespaced --serviceaccount my-cluster-operator-namespace:strimzi-cluster-operator
kubectl create clusterrolebinding strimzi-cluster-operator-watched --clusterrole=strimzi-cluster-operator-watched --serviceaccount my-cluster-operator-namespace:strimzi-cluster-operator
kubectl create clusterrolebinding strimzi-cluster-operator-entity-operator-delegation --clusterrole=strimzi-entity-operator --serviceaccount my-cluster-operator-namespace:strimzi-cluster-operator

  4. Deploy the Cluster Operator to your Kubernetes cluster.

    kubectl create -f install/cluster-operator -n my-cluster-operator-namespace

  5. Check the status of the deployment:

    kubectl get deployments -n my-cluster-operator-namespace
    Output shows the deployment name and readiness
    NAME                      READY  UP-TO-DATE  AVAILABLE
strimzi-cluster-operator  1/1    1           1

    READY shows the number of replicas that are ready/expected. The deployment is successful when the AVAILABLE output shows 1.

5.3. Deploying Kafka

To be able to manage a Kafka cluster with the Cluster Operator, you must deploy it as a Kafka resource. Strimzi provides example deployment files to do this. You can use these files to deploy the Topic Operator and User Operator at the same time.

After you have deployed the Cluster Operator, use a Kafka resource to deploy the following components:

When installing Kafka, Strimzi also installs a ZooKeeper cluster and adds the necessary configuration to connect Kafka with ZooKeeper.

If you haven’t deployed a Kafka cluster as a Kafka resource, you can’t use the Cluster Operator to manage it. This applies, for example, to a Kafka cluster running outside of Kubernetes. However, you can use the Topic Operator and User Operator with a Kafka cluster that is not managed by Strimzi, by deploying them as standalone components. You can also deploy and use other Kafka components with a Kafka cluster not managed by Strimzi.

5.3.1. Deploying the Kafka cluster

This procedure shows how to deploy a Kafka cluster to your Kubernetes cluster using the Cluster Operator.

The deployment uses a YAML file to provide the specification to create a Kafka resource.

Strimzi provides the following example files you can use to create a Kafka cluster:

kafka-persistent.yaml

Deploys a persistent cluster with three ZooKeeper and three Kafka nodes.

kafka-jbod.yaml

Deploys a persistent cluster with three ZooKeeper and three Kafka nodes (each using multiple persistent volumes).

kafka-persistent-single.yaml

Deploys a persistent cluster with a single ZooKeeper node and a single Kafka node.

kafka-ephemeral.yaml

Deploys an ephemeral cluster with three ZooKeeper and three Kafka nodes.

kafka-ephemeral-single.yaml

Deploys an ephemeral cluster with three ZooKeeper nodes and a single Kafka node.

In this procedure, we use the examples for an ephemeral and persistent Kafka cluster deployment.

Ephemeral cluster

In general, an ephemeral (or temporary) Kafka cluster is suitable for development and testing purposes, not for production. This deployment uses emptyDir volumes for storing broker information (for ZooKeeper) and topics or partitions (for Kafka). Using an emptyDir volume means that its content is strictly related to the pod life cycle and is deleted when the pod goes down.

Persistent cluster

A persistent Kafka cluster uses persistent volumes to store ZooKeeper and Kafka data. A PersistentVolume is acquired using a PersistentVolumeClaim to make it independent of the actual type of the PersistentVolume. The PersistentVolumeClaim can use a StorageClass to trigger automatic volume provisioning. When no StorageClass is specified, Kubernetes will try to use the default StorageClass.

The following examples show some common types of persistent volumes:

  • If your Kubernetes cluster runs on Amazon AWS, Kubernetes can provision Amazon EBS volumes

  • If your Kubernetes cluster runs on Microsoft Azure, Kubernetes can provision Azure Disk Storage volumes

  • If your Kubernetes cluster runs on Google Cloud, Kubernetes can provision Persistent Disk volumes

  • If your Kubernetes cluster runs on bare metal, Kubernetes can provision local persistent volumes

The example YAML files specify the latest supported Kafka version, and configuration for its supported log message format version and inter-broker protocol version. The inter.broker.protocol.version property for the Kafka config must be the version supported by the specified Kafka version (spec.kafka.version). The property represents the version of Kafka protocol used in a Kafka cluster.

From Kafka 3.0.0, when the inter.broker.protocol.version is set to 3.0 or higher, the log.message.format.version option is ignored and doesn’t need to be set.

An update to the inter.broker.protocol.version is required when upgrading Kafka.

The example clusters are named my-cluster by default. The cluster name is defined by the name of the resource and cannot be changed after the cluster has been deployed. To change the cluster name before you deploy the cluster, edit the Kafka.metadata.name property of the Kafka resource in the relevant YAML file.

Default cluster name and specified Kafka versions
apiVersion: kafka.strimzi.io/v1beta2
kind: Kafka
metadata:
  name: my-cluster
spec:
  kafka:
    version: 3.3.2
    #...
    config:
      #...
      log.message.format.version: "3.3"
      inter.broker.protocol.version: "3.3"
  # ...
Prerequisites
Procedure

  1. Create and deploy an ephemeral or persistent cluster.

    • To create and deploy an ephemeral cluster:

      kubectl apply -f examples/kafka/kafka-ephemeral.yaml

    • To create and deploy a persistent cluster:

      kubectl apply -f examples/kafka/kafka-persistent.yaml

  2. Check the status of the deployment:

    kubectl get pods -n <my_cluster_operator_namespace>
    Output shows the pod names and readiness
    NAME                        READY   STATUS    RESTARTS
my-cluster-entity-operator  3/3     Running   0
my-cluster-kafka-0          1/1     Running   0
my-cluster-kafka-1          1/1     Running   0
my-cluster-kafka-2          1/1     Running   0
my-cluster-zookeeper-0      1/1     Running   0
my-cluster-zookeeper-1      1/1     Running   0
my-cluster-zookeeper-2      1/1     Running   0

    my-cluster is the name of the Kafka cluster.

    With the default deployment, you install an Entity Operator cluster, 3 Kafka pods, and 3 ZooKeeper pods.

    READY shows the number of replicas that are ready/expected. The deployment is successful when the STATUS shows as Running.

Additional resources

Kafka cluster configuration

5.3.2. Deploying the Topic Operator using the Cluster Operator

This procedure describes how to deploy the Topic Operator using the Cluster Operator.

You configure the entityOperator property of the Kafka resource to include the topicOperator. By default, the Topic Operator watches for KafkaTopic resources in the namespace of the Kafka cluster deployed by the Cluster Operator. You can also specify a namespace using watchedNamespace in the Topic Operator spec. A single Topic Operator can watch a single namespace. One namespace should be watched by only one Topic Operator.

If you use Strimzi to deploy multiple Kafka clusters into the same namespace, enable the Topic Operator for only one Kafka cluster or use the watchedNamespace property to configure the Topic Operators to watch other namespaces.

If you want to use the Topic Operator with a Kafka cluster that is not managed by Strimzi, you must deploy the Topic Operator as a standalone component.

For more information about configuring the entityOperator and topicOperator properties, see Configuring the Entity Operator.

Prerequisites
Procedure

  1. Edit the entityOperator properties of the Kafka resource to include topicOperator:

    apiVersion: kafka.strimzi.io/v1beta2
kind: Kafka
metadata:
  name: my-cluster
spec:
  #...
  entityOperator:
    topicOperator: {}
    userOperator: {}

  2. Configure the Topic Operator spec using the properties described in EntityTopicOperatorSpec schema reference.

    Use an empty object ({}) if you want all properties to use their default values.

  3. Create or update the resource:

    kubectl apply -f <kafka_configuration_file>

  4. Check the status of the deployment:

    kubectl get pods -n <my_cluster_operator_namespace>
    Output shows the pod name and readiness
    NAME                        READY   STATUS    RESTARTS
my-cluster-entity-operator  3/3     Running   0
# ...

    my-cluster is the name of the Kafka cluster.

    READY shows the number of replicas that are ready/expected. The deployment is successful when the STATUS shows as Running.

5.3.3. Deploying the User Operator using the Cluster Operator

This procedure describes how to deploy the User Operator using the Cluster Operator.

You configure the entityOperator property of the Kafka resource to include the userOperator. By default, the User Operator watches for KafkaUser resources in the namespace of the Kafka cluster deployment. You can also specify a namespace using watchedNamespace in the User Operator spec. A single User Operator can watch a single namespace. One namespace should be watched by only one User Operator.

If you want to use the User Operator with a Kafka cluster that is not managed by Strimzi, you must deploy the User Operator as a standalone component.

For more information about configuring the entityOperator and userOperator properties, see Configuring the Entity Operator.

Prerequisites
Procedure

  1. Edit the entityOperator properties of the Kafka resource to include userOperator:

    apiVersion: kafka.strimzi.io/v1beta2
kind: Kafka
metadata:
  name: my-cluster
spec:
  #...
  entityOperator:
    topicOperator: {}
    userOperator: {}

  2. Configure the User Operator spec using the properties described in EntityUserOperatorSpec schema reference.

    Use an empty object ({}) if you want all properties to use their default values.

  3. Create or update the resource:

    kubectl apply -f <kafka_configuration_file>

  4. Check the status of the deployment:

    kubectl get pods -n <my_cluster_operator_namespace>
    Output shows the pod name and readiness
    NAME                        READY   STATUS    RESTARTS
my-cluster-entity-operator  3/3     Running   0
# ...

    my-cluster is the name of the Kafka cluster.

    READY shows the number of replicas that are ready/expected. The deployment is successful when the STATUS shows as Running.

5.4. Deploying Kafka Connect

Kafka Connect is a tool for streaming data between Apache Kafka and other systems. For example, Kafka Connect might integrate Kafka with external databases or storage and messaging systems.

In Strimzi, Kafka Connect is deployed in distributed mode. Kafka Connect can also work in standalone mode, but this is not supported by Strimzi.

Using the concept of connectors, Kafka Connect provides a framework for moving large amounts of data into and out of your Kafka cluster while maintaining scalability and reliability.

The Cluster Operator manages Kafka Connect clusters deployed using the KafkaConnect resource and connectors created using the KafkaConnector resource.

In order to use Kafka Connect, you need to do the following.

Note
 The term connector is used interchangeably to mean a connector instance running within a Kafka Connect cluster, or a connector class. In this guide, the term connector is used when the meaning is clear from the context.

5.4.1. Deploying Kafka Connect to your Kubernetes cluster

This procedure shows how to deploy a Kafka Connect cluster to your Kubernetes cluster using the Cluster Operator.

A Kafka Connect cluster is implemented as a Deployment with a configurable number of nodes (also called workers) that distribute the workload of connectors as tasks so that the message flow is highly scalable and reliable.

The deployment uses a YAML file to provide the specification to create a KafkaConnect resource.

Strimzi provides example configuration files. In this procedure, we use the following example file:

  • examples/connect/kafka-connect.yaml

Prerequisites
Procedure

  1. Deploy Kafka Connect to your Kubernetes cluster. Use the examples/connect/kafka-connect.yaml file to deploy Kafka Connect.

    kubectl apply -f examples/connect/kafka-connect.yaml

  2. Check the status of the deployment:

    kubectl get deployments -n <my_cluster_operator_namespace>
    Output shows the deployment name and readiness
    NAME                        READY  UP-TO-DATE  AVAILABLE
my-connect-cluster-connect  1/1    1           1

    my-connect-cluster is the name of the Kafka Connect cluster.

    READY shows the number of replicas that are ready/expected. The deployment is successful when the AVAILABLE output shows 1.

Additional resources

Kafka Connect cluster configuration

5.4.2. Configuring Kafka Connect for multiple instances

If you are running multiple instances of Kafka Connect, you have to change the default configuration of the following config properties:

apiVersion: kafka.strimzi.io/v1beta2
kind: KafkaConnect
metadata:
  name: my-connect
spec:
  # ...
  config:
    group.id: connect-cluster (1)
    offset.storage.topic: connect-cluster-offsets (2)
    config.storage.topic: connect-cluster-configs (3)
    status.storage.topic: connect-cluster-status  (4)
    # ...
# ...

  1. The Kafka Connect cluster ID within Kafka.

  2. Kafka topic that stores connector offsets.

  3. Kafka topic that stores connector and task status configurations.

  4. Kafka topic that stores connector and task status updates.

Note
 Values for the three topics must be the same for all Kafka Connect instances with the same group.id.

Unless you change the default settings, each Kafka Connect instance connecting to the same Kafka cluster is deployed with the same values. What happens, in effect, is all instances are coupled to run in a cluster and use the same topics.

If multiple Kafka Connect clusters try to use the same topics, Kafka Connect will not work as expected and generate errors.

If you wish to run multiple Kafka Connect instances, change the values of these properties for each instance.

5.4.3. Adding connectors

Kafka Connect uses connectors to integrate with other systems to stream data. A connector is an instance of a Kafka Connector class, which can be one of the following type:

Source connector

A source connector is a runtime entity that fetches data from an external system and feeds it to Kafka as messages.

Sink connector

A sink connector is a runtime entity that fetches messages from Kafka topics and feeds them to an external system.

Kafka Connect uses a plugin architecture to provide the implementation artifacts for connectors. Plugins allow connections to other systems and provide additional configuration to manipulate data. Plugins include connectors and other components, such as data converters and transforms. A connector operates with a specific type of external system. Each connector defines a schema for its configuration. You supply the configuration to Kafka Connect to create a connector instance within Kafka Connect. Connector instances then define a set of tasks for moving data between systems.

Add connector plugins to Kafka Connect in one of the following ways:

After plugins have been added to the container image, you can start, stop, and manage connector instances in the following ways:

You can also create new connector instances using these options.

Building a new container image with connector plugins automatically

Configure Kafka Connect so that Strimzi automatically builds a new container image with additional connectors. You define the connector plugins using the .spec.build.plugins property of the KafkaConnect custom resource. Strimzi will automatically download and add the connector plugins into a new container image. The container is pushed into the container repository specified in .spec.build.output and automatically used in the Kafka Connect deployment.

Prerequisites

You need to provide your own container registry where images can be pushed to, stored, and pulled from. Strimzi supports private container registries as well as public registries such as Quay or Docker Hub.

Procedure

  1. Configure the KafkaConnect custom resource by specifying the container registry in .spec.build.output, and additional connectors in .spec.build.plugins:

    apiVersion: kafka.strimzi.io/v1beta2
kind: KafkaConnect
metadata:
  name: my-connect-cluster
spec: # (1)
  #...
  build:
    output: # (2)
      type: docker
      image: my-registry.io/my-org/my-connect-cluster:latest
      pushSecret: my-registry-credentials
    plugins: # (3)
      - name: debezium-postgres-connector
        artifacts:
          - type: tgz
            url: https://repo1.maven.org/maven2/io/debezium/debezium-connector-postgres/2.1.1.Final/debezium-connector-postgres-2.1.1.Final-plugin.tar.gz
            sha512sum: 962a12151bdf9a5a30627eebac739955a4fd95a08d373b86bdcea2b4d0c27dd6e1edd5cb548045e115e33a9e69b1b2a352bee24df035a0447cb820077af00c03
      - name: camel-telegram
        artifacts:
          - type: tgz
            url: https://repo.maven.apache.org/maven2/org/apache/camel/kafkaconnector/camel-telegram-kafka-connector/0.9.0/camel-telegram-kafka-connector-0.9.0-package.tar.gz
            sha512sum: a9b1ac63e3284bea7836d7d24d84208c49cdf5600070e6bd1535de654f6920b74ad950d51733e8020bf4187870699819f54ef5859c7846ee4081507f48873479
  #...

    1. The specification for the Kafka Connect cluster.

    2. (Required) Configuration of the container registry where new images are pushed.

    3. (Required) List of connector plugins and their artifacts to add to the new container image. Each plugin must be configured with at least one artifact.

  2. Create or update the resource:

    $ kubectl apply -f <kafka_connect_configuration_file>

  3. Wait for the new container image to build, and for the Kafka Connect cluster to be deployed.

  4. Use the Kafka Connect REST API or KafkaConnector custom resources to use the connector plugins you added.

Additional resources
Building a new container image with connector plugins from the Kafka Connect base image

Create a custom Docker image with connector plugins from the Kafka Connect base image Add the custom image to the /opt/kafka/plugins directory.

You can use the Kafka container image on Container Registry as a base image for creating your own custom image with additional connector plugins.

At startup, the Strimzi version of Kafka Connect loads any third-party connector plugins contained in the /opt/kafka/plugins directory.

Prerequisites
Procedure

  1. Create a new Dockerfile using quay.io/strimzi/kafka:latest-kafka-3.3.2 as the base image:

    FROM quay.io/strimzi/kafka:latest-kafka-3.3.2
USER root:root
COPY ./my-plugins/ /opt/kafka/plugins/
USER 1001
    Example plugins file
    $ tree ./my-plugins/
./my-plugins/
├── debezium-connector-mongodb
│   ├── bson-<version>.jar
│   ├── CHANGELOG.md
│   ├── CONTRIBUTE.md
│   ├── COPYRIGHT.txt
│   ├── debezium-connector-mongodb-<version>.jar
│   ├── debezium-core-<version>.jar
│   ├── LICENSE.txt
│   ├── mongodb-driver-core-<version>.jar
│   ├── README.md
│   └── # ...
├── debezium-connector-mysql
│   ├── CHANGELOG.md
│   ├── CONTRIBUTE.md
│   ├── COPYRIGHT.txt
│   ├── debezium-connector-mysql-<version>.jar
│   ├── debezium-core-<version>.jar
│   ├── LICENSE.txt
│   ├── mysql-binlog-connector-java-<version>.jar
│   ├── mysql-connector-java-<version>.jar
│   ├── README.md
│   └── # ...
└── debezium-connector-postgres
    ├── CHANGELOG.md
    ├── CONTRIBUTE.md
    ├── COPYRIGHT.txt
    ├── debezium-connector-postgres-<version>.jar
    ├── debezium-core-<version>.jar
    ├── LICENSE.txt
    ├── postgresql-<version>.jar
    ├── protobuf-java-<version>.jar
    ├── README.md
    └── # ...

    The COPY command points to the plugin files to copy to the container image.

    This example adds plugins for Debezium connectors (MongoDB, MySQL, and PostgreSQL), though not all files are listed for brevity. Debezium running in Kafka Connect looks the same as any other Kafka Connect task.

  2. Build the container image.

  3. Push your custom image to your container registry.

  4. Point to the new container image.

    You can point to the image in one of the following ways:

    • Edit the KafkaConnect.spec.image property of the KafkaConnect custom resource.

      If set, this property overrides the STRIMZI_KAFKA_CONNECT_IMAGES environment variable in the Cluster Operator.

      apiVersion: kafka.strimzi.io/v1beta2
kind: KafkaConnect
metadata:
  name: my-connect-cluster
spec: (1)
  #...
  image: my-new-container-image (2)
  config: (3)
    #...

      1. The specification for the Kafka Connect cluster.

      2. The docker image for the pods.

      3. Configuration of the Kafka Connect workers (not connectors).

    • Edit the STRIMZI_KAFKA_CONNECT_IMAGES environment variable in the install/cluster-operator/060-Deployment-strimzi-cluster-operator.yaml file to point to the new container image, and then reinstall the Cluster Operator.

Additional resources
Deploying KafkaConnector resources

Deploy KafkaConnector resources to manage connectors. The KafkaConnector custom resource offers a Kubernetes-native approach to management of connectors by the Cluster Operator. You don’t need to send HTTP requests to manage connectors, as with the Kafka Connect REST API. You manage a running connector instance by updating its corresponding KafkaConnector resource, and then applying the updates. The Cluster Operator updates the configurations of the running connector instances. You remove a connector by deleting its corresponding KafkaConnector.

KafkaConnector resources must be deployed to the same namespace as the Kafka Connect cluster they link to.

In the configuration shown in this procedure, the autoRestart property is set to true. This enables automatic restarts of failed connectors and tasks. Up to seven restart attempts are made, after which restarts must be made manually. You annotate the KafkaConnector resource to restart a connector or restart a connector task manually.

Example connectors

You can use your own connectors or try the examples provided by Strimzi. Up until Apache Kafka 3.1.0, example file connector plugins were included with Apache Kafka. Starting from the 3.1.1 and 3.2.0 releases of Apache Kafka, the examples need to be added to the plugin path as any other connector.

Strimzi provides an example KafkaConnector configuration file (examples/connect/source-connector.yaml) for the example file connector plugins, which creates the following connector instances as KafkaConnector resources:

  • A FileStreamSourceConnector instance that reads each line from the Kafka license file (the source) and writes the data as messages to a single Kafka topic.

  • A FileStreamSinkConnector instance that reads messages from the Kafka topic and writes the messages to a temporary file (the sink).

We use the example file to create connectors in this procedure.

Note
 The example connectors are not intended for use in a production environment.
Prerequisites

  • A Kafka Connect deployment

  • The Cluster Operator is running

Procedure

  1. Add the FileStreamSourceConnector and FileStreamSinkConnector plugins to Kafka Connect in one of the following ways:

  2. Set the strimzi.io/use-connector-resources annotation to true in the Kafka Connect configuration.

    apiVersion: kafka.strimzi.io/v1beta2
kind: KafkaConnect
metadata:
  name: my-connect-cluster
  annotations:
    strimzi.io/use-connector-resources: "true"
spec:
    # ...

    With the KafkaConnector resources enabled, the Cluster Operator watches for them.

  3. Edit the examples/connect/source-connector.yaml file:

    apiVersion: kafka.strimzi.io/v1beta2
kind: KafkaConnector
metadata:
  name: my-source-connector # (1)
  labels:
    strimzi.io/cluster: my-connect-cluster # (2)
spec:
  class: org.apache.kafka.connect.file.FileStreamSourceConnector # (3)
  tasksMax: 2 # (4)
  autoRestart: # (5)
    enabled: true
  config: # (6)
    file: "/opt/kafka/LICENSE" # (7)
    topic: my-topic (8)
    # ...

    1. Name of the KafkaConnector resource, which is used as the name of the connector. Use any name that is valid for a Kubernetes resource.

    2. Name of the Kafka Connect cluster to create the connector instance in. Connectors must be deployed to the same namespace as the Kafka Connect cluster they link to.

    3. Full name or alias of the connector class. This should be present in the image being used by the Kafka Connect cluster.

    4. Maximum number of Kafka Connect tasks that the connector can create.

    5. Enables automatic restarts of failed connectors and tasks.

    6. Connector configuration as key-value pairs.

    7. This example source connector configuration reads data from the /opt/kafka/LICENSE file.

    8. Kafka topic to publish the source data to.

  4. Create the source KafkaConnector in your Kubernetes cluster:

    kubectl apply -f examples/connect/source-connector.yaml

  5. Create an examples/connect/sink-connector.yaml file:

    touch examples/connect/sink-connector.yaml

  6. Paste the following YAML into the sink-connector.yaml file:

    apiVersion: kafka.strimzi.io/v1beta2
kind: KafkaConnector
metadata:
  name: my-sink-connector
  labels:
    strimzi.io/cluster: my-connect
spec:
  class: org.apache.kafka.connect.file.FileStreamSinkConnector (1)
  tasksMax: 2
  config: (2)
    file: "/tmp/my-file" (3)
    topics: my-topic (4)

    1. Full name or alias of the connector class. This should be present in the image being used by the Kafka Connect cluster.

    2. Connector configuration as key-value pairs.

    3. Temporary file to publish the source data to.

    4. Kafka topic to read the source data from.

  7. Create the sink KafkaConnector in your Kubernetes cluster:

    kubectl apply -f examples/connect/sink-connector.yaml

  8. Check that the connector resources were created:

    kubectl get kctr --selector strimzi.io/cluster=<my_connect_cluster> -o name

my-source-connector
my-sink-connector

    Replace <my_connect_cluster> with the name of your Kafka Connect cluster.

  9. In the container, execute kafka-console-consumer.sh to read the messages that were written to the topic by the source connector:

    kubectl exec <my_kafka_cluster>-kafka-0 -i -t -- bin/kafka-console-consumer.sh --bootstrap-server <my_kafka_cluster>-kafka-bootstrap.NAMESPACE.svc:9092 --topic my-topic --from-beginning

    Replace <my_kafka_cluster> with the name of your Kafka cluster.

Source and sink connector configuration options

The connector configuration is defined in the spec.config property of the KafkaConnector resource.

The FileStreamSourceConnector and FileStreamSinkConnector classes support the same configuration options as the Kafka Connect REST API. Other connectors support different configuration options.

Table 1. Configuration options for the FileStreamSource connector class
Name Type Default value Description

file

String

Null

Source file to write messages to. If not specified, the standard input is used.

topic

List

Null

The Kafka topic to publish data to.
Table 2. Configuration options for FileStreamSinkConnector class
Name Type Default value Description

file

String

Null

Destination file to write messages to. If not specified, the standard output is used.

topics

List

Null

One or more Kafka topics to read data from.

topics.regex

String

Null

A regular expression matching one or more Kafka topics to read data from.
Manually restarting connectors

If you are using KafkaConnector resources to manage connectors, use the restart annotation to manually trigger a restart of a connector.

Prerequisites

  • The Cluster Operator is running.

Procedure

  1. Find the name of the KafkaConnector custom resource that controls the Kafka connector you want to restart:

    kubectl get KafkaConnector

  2. Restart the connector by annotating the KafkaConnector resource in Kubernetes.

    kubectl annotate KafkaConnector <kafka_connector_name> strimzi.io/restart=true

    The restart annotation is set to true.

  3. Wait for the next reconciliation to occur (every two minutes by default).

    The Kafka connector is restarted, as long as the annotation was detected by the reconciliation process. When Kafka Connect accepts the restart request, the annotation is removed from the KafkaConnector custom resource.

Manually restarting Kafka connector tasks

If you are using KafkaConnector resources to manage connectors, use the restart-task annotation to manually trigger a restart of a connector task.

Prerequisites

  • The Cluster Operator is running.

Procedure

  1. Find the name of the KafkaConnector custom resource that controls the Kafka connector task you want to restart:

    kubectl get KafkaConnector

  2. Find the ID of the task to be restarted from the KafkaConnector custom resource. Task IDs are non-negative integers, starting from 0:

    kubectl describe KafkaConnector <kafka_connector_name>

  3. Use the ID to restart the connector task by annotating the KafkaConnector resource in Kubernetes:

    kubectl annotate KafkaConnector <kafka_connector_name> strimzi.io/restart-task=0

    In this example, task 0 is restarted.

  4. Wait for the next reconciliation to occur (every two minutes by default).

    The Kafka connector task is restarted, as long as the annotation was detected by the reconciliation process. When Kafka Connect accepts the restart request, the annotation is removed from the KafkaConnector custom resource.

Exposing the Kafka Connect API

Use the Kafka Connect REST API as an alternative to using KafkaConnector resources to manage connectors. The Kafka Connect REST API is available as a service running on <connect_cluster_name>-connect-api:8083, where <connect_cluster_name> is the name of your Kafka Connect cluster. The service is created when you create a Kafka Connect instance.

The operations supported by the Kafka Connect REST API are described in the Apache Kafka Connect API documentation.

Note
 The strimzi.io/use-connector-resources annotation enables KafkaConnectors. If you applied the annotation to your KafkaConnect resource configuration, you need to remove it to use the Kafka Connect API. Otherwise, manual changes made directly using the Kafka Connect REST API are reverted by the Cluster Operator.

You can add the connector configuration as a JSON object.

Example curl request to add connector configuration
curl -X POST \
  http://my-connect-cluster-connect-api:8083/connectors \
  -H 'Content-Type: application/json' \
  -d '{ "name": "my-source-connector",
    "config":
    {
      "connector.class":"org.apache.kafka.connect.file.FileStreamSourceConnector",
      "file": "/opt/kafka/LICENSE",
      "topic":"my-topic",
      "tasksMax": "4",
      "type": "source"
    }
}'

The API is only accessible within the Kubernetes cluster. If you want to make the Kafka Connect API accessible to applications running outside of the Kubernetes cluster, you can expose it manually by creating one of the following features:

  • LoadBalancer or NodePort type services

  • Ingress resources

  • OpenShift routes

Note
 The connection is insecure, so allow external access advisedly.

If you decide to create services, use the labels from the selector of the <connect_cluster_name>-connect-api service to configure the pods to which the service will route the traffic:

Selector configuration for the service
# ...
selector:
  strimzi.io/cluster: my-connect-cluster (1)
  strimzi.io/kind: KafkaConnect
  strimzi.io/name: my-connect-cluster-connect (2)
#...

  1. Name of the Kafka Connect custom resource in your Kubernetes cluster.

  2. Name of the Kafka Connect deployment created by the Cluster Operator.

You must also create a NetworkPolicy that allows HTTP requests from external clients.

Example NetworkPolicy to allow requests to the Kafka Connect API
apiVersion: networking.k8s.io/v1
kind: NetworkPolicy
metadata:
  name: my-custom-connect-network-policy
spec:
  ingress:
  - from:
    - podSelector: (1)
        matchLabels:
          app: my-connector-manager
    ports:
    - port: 8083
      protocol: TCP
  podSelector:
    matchLabels:
      strimzi.io/cluster: my-connect-cluster
      strimzi.io/kind: KafkaConnect
      strimzi.io/name: my-connect-cluster-connect
  policyTypes:
  - Ingress

  1. The label of the pod that is allowed to connect to the API.

To add the connector configuration outside the cluster, use the URL of the resource that exposes the API in the curl command.

Switching from using the Kafka Connect API to using KafkaConnector custom resources

You can switch from using the Kafka Connect API to using KafkaConnector custom resources to manage your connectors. To make the switch, do the following in the order shown:

  1. Deploy KafkaConnector resources with the configuration to create your connector instances.

  2. Enable KafkaConnector resources in your Kafka Connect configuration by setting the strimzi.io/use-connector-resources annotation to true.

Warning
 If you enable KafkaConnector resources before creating them, you delete all connectors.

To switch from using KafkaConnector resources to using the Kafka Connect API, first remove the annotation that enables the KafkaConnector resources from your Kafka Connect configuration. Otherwise, manual changes made directly using the Kafka Connect REST API are reverted by the Cluster Operator.

When making the switch, check the status of the KafkaConnect resource. The value of metadata.generation (the current version of the deployment) must match status.observedGeneration (the latest reconciliation of the resource). When the Kafka Connect cluster is Ready, you can delete the KafkaConnector resources.

5.5. Deploying Kafka MirrorMaker

The Cluster Operator deploys one or more Kafka MirrorMaker replicas to replicate data between Kafka clusters. This process is called mirroring to avoid confusion with the Kafka partitions replication concept. MirrorMaker consumes messages from the source cluster and republishes those messages to the target cluster.

5.5.1. Deploying Kafka MirrorMaker to your Kubernetes cluster

This procedure shows how to deploy a Kafka MirrorMaker cluster to your Kubernetes cluster using the Cluster Operator.

The deployment uses a YAML file to provide the specification to create a KafkaMirrorMaker or KafkaMirrorMaker2 resource depending on the version of MirrorMaker deployed.

Important
 Kafka MirrorMaker 1 (referred to as just MirrorMaker in the documentation) has been deprecated in Apache Kafka 3.0.0 and will be removed in Apache Kafka 4.0.0. As a result, the KafkaMirrorMaker custom resource which is used to deploy Kafka MirrorMaker 1 has been deprecated in Strimzi as well. The KafkaMirrorMaker resource will be removed from Strimzi when we adopt Apache Kafka 4.0.0. As a replacement, use the KafkaMirrorMaker2 custom resource with the IdentityReplicationPolicy.

Strimzi provides example configuration files. In this procedure, we use the following example files:

  • examples/mirror-maker/kafka-mirror-maker.yaml

  • examples/mirror-maker/kafka-mirror-maker-2.yaml

Prerequisites
Procedure

  1. Deploy Kafka MirrorMaker to your Kubernetes cluster:

    For MirrorMaker:

    kubectl apply -f examples/mirror-maker/kafka-mirror-maker.yaml

    For MirrorMaker 2.0:

    kubectl apply -f examples/mirror-maker/kafka-mirror-maker-2.yaml

  2. Check the status of the deployment:

    kubectl get deployments -n <my_cluster_operator_namespace>
    Output shows the deployment name and readiness
    NAME                          READY  UP-TO-DATE  AVAILABLE
my-mirror-maker-mirror-maker  1/1    1           1
my-mm2-cluster-mirrormaker2   1/1    1           1

    my-mirror-maker is the name of the Kafka MirrorMaker cluster. my-mm2-cluster is the name of the Kafka MirrorMaker 2.0 cluster.

    READY shows the number of replicas that are ready/expected. The deployment is successful when the AVAILABLE output shows 1.

Additional resources

5.6. Deploying Kafka Bridge

The Cluster Operator deploys one or more Kafka bridge replicas to send data between Kafka clusters and clients via HTTP API.

5.6.1. Deploying Kafka Bridge to your Kubernetes cluster

This procedure shows how to deploy a Kafka Bridge cluster to your Kubernetes cluster using the Cluster Operator.

The deployment uses a YAML file to provide the specification to create a KafkaBridge resource.

Strimzi provides example configuration files. In this procedure, we use the following example file:

  • examples/bridge/kafka-bridge.yaml

Prerequisites
Procedure

  1. Deploy Kafka Bridge to your Kubernetes cluster:

    kubectl apply -f examples/bridge/kafka-bridge.yaml

  2. Check the status of the deployment:

    kubectl get deployments -n <my_cluster_operator_namespace>
    Output shows the deployment name and readiness
    NAME              READY  UP-TO-DATE  AVAILABLE
my-bridge-bridge  1/1    1           1

    my-bridge is the name of the Kafka Bridge cluster.

    READY shows the number of replicas that are ready/expected. The deployment is successful when the AVAILABLE output shows 1.

Additional resources

5.6.2. Exposing the Kafka Bridge service to your local machine

Use port forwarding to expose the Strimzi Kafka Bridge service to your local machine on http://localhost:8080.

Note
 Port forwarding is only suitable for development and testing purposes.
Procedure

  1. List the names of the pods in your Kubernetes cluster:

    kubectl get pods -o name

pod/kafka-consumer
# ...
pod/my-bridge-bridge-7cbd55496b-nclrt

  2. Connect to the Kafka Bridge pod on port 8080:

    kubectl port-forward pod/my-bridge-bridge-7cbd55496b-nclrt 8080:8080 &
    Note
    		 If port 8080 on your local machine is already in use, use an alternative HTTP port, such as 8008.

API requests are now forwarded from port 8080 on your local machine to port 8080 in the Kafka Bridge pod.

5.6.3. Accessing the Kafka Bridge outside of Kubernetes

After deployment, the Strimzi Kafka Bridge can only be accessed by applications running in the same Kubernetes cluster. These applications use the <kafka_bridge_name>-bridge-service service to access the API.

If you want to make the Kafka Bridge accessible to applications running outside of the Kubernetes cluster, you can expose it manually by creating one of the following features:

  • LoadBalancer or NodePort type services

  • Ingress resources

  • OpenShift routes

If you decide to create Services, use the labels from the selector of the <kafka_bridge_name>-bridge-service service to configure the pods to which the service will route the traffic:

  # ...
  selector:
    strimzi.io/cluster: kafka-bridge-name (1)
    strimzi.io/kind: KafkaBridge
  #...

  1. Name of the Kafka Bridge custom resource in your Kubernetes cluster.

5.7. Alternative standalone deployment options for Strimzi operators

You can perform a standalone deployment of the Topic Operator and User Operator. Consider a standalone deployment of these operators if you are using a Kafka cluster that is not managed by the Cluster Operator.

You deploy the operators to Kubernetes. Kafka can be running outside of Kubernetes. For example, you might be using a Kafka as a managed service. You adjust the deployment configuration for the standalone operator to match the address of your Kafka cluster.

5.7.1. Deploying the standalone Topic Operator

This procedure shows how to deploy the Topic Operator as a standalone component for topic management. You can use a standalone Topic Operator with a Kafka cluster that is not managed by the Cluster Operator.

A standalone deployment can operate with any Kafka cluster.

Standalone deployment files are provided with Strimzi. Use the 05-Deployment-strimzi-topic-operator.yaml deployment file to deploy the Topic Operator. Add or set the environment variables needed to make a connection to a Kafka cluster.

The Topic Operator watches for KafkaTopic resources in a single namespace. You specify the namespace to watch, and the connection to the Kafka cluster, in the Topic Operator configuration. A single Topic Operator can watch a single namespace. One namespace should be watched by only one Topic Operator. If you want to use more than one Topic Operator, configure each of them to watch different namespaces. In this way, you can use Topic Operators with multiple Kafka clusters.

Prerequisites

  • You are running a Kafka cluster for the Topic Operator to connect to.

    As long as the standalone Topic Operator is correctly configured for connection, the Kafka cluster can be running on a bare-metal environment, a virtual machine, or as a managed cloud application service.

Procedure

  1. Edit the env properties in the install/topic-operator/05-Deployment-strimzi-topic-operator.yaml standalone deployment file.

    Example standalone Topic Operator deployment configuration
    apiVersion: apps/v1
kind: Deployment
metadata:
  name: strimzi-topic-operator
  labels:
    app: strimzi
spec:
  # ...
  template:
    # ...
    spec:
      # ...
      containers:
        - name: strimzi-topic-operator
          # ...
          env:
            - name: STRIMZI_NAMESPACE (1)
              valueFrom:
                fieldRef:
                  fieldPath: metadata.namespace
            - name: STRIMZI_KAFKA_BOOTSTRAP_SERVERS (2)
              value: my-kafka-bootstrap-address:9092
            - name: STRIMZI_RESOURCE_LABELS (3)
              value: "strimzi.io/cluster=my-cluster"
            - name: STRIMZI_ZOOKEEPER_CONNECT (4)
              value: my-cluster-zookeeper-client:2181
            - name: STRIMZI_ZOOKEEPER_SESSION_TIMEOUT_MS (5)
              value: "18000"
            - name: STRIMZI_FULL_RECONCILIATION_INTERVAL_MS (6)
              value: "120000"
            - name: STRIMZI_TOPIC_METADATA_MAX_ATTEMPTS (7)
              value: "6"
            - name: STRIMZI_LOG_LEVEL (8)
              value: INFO
            - name: STRIMZI_TLS_ENABLED (9)
              value: "false"
            - name: STRIMZI_JAVA_OPTS (10)
              value: "-Xmx=512M -Xms=256M"
            - name: STRIMZI_JAVA_SYSTEM_PROPERTIES (11)
              value: "-Djavax.net.debug=verbose -DpropertyName=value"
            - name: STRIMZI_PUBLIC_CA (12)
              value: "false"
            - name: STRIMZI_TLS_AUTH_ENABLED (13)
              value: "false"
            - name: STRIMZI_SASL_ENABLED (14)
              value: "false"
            - name: STRIMZI_SASL_USERNAME (15)
              value: "admin"
            - name: STRIMZI_SASL_PASSWORD (16)
              value: "password"
            - name: STRIMZI_SASL_MECHANISM (17)
              value: "scram-sha-512"
            - name: STRIMZI_SECURITY_PROTOCOL (18)
              value: "SSL"

    1. The Kubernetes namespace for the Topic Operator to watch for KafkaTopic resources. Specify the namespace of the Kafka cluster.

    2. The host and port pair of the bootstrap broker address to discover and connect to all brokers in the Kafka cluster. Use a comma-separated list to specify two or three broker addresses in case a server is down.

    3. The label to identify the KafkaTopic resources managed by the Topic Operator. This does not have to be the name of the Kafka cluster. It can be the label assigned to the KafkaTopic resource. If you deploy more than one Topic Operator, the labels must be unique for each. That is, the operators cannot manage the same resources.

    4. The host and port pair of the address to connect to the ZooKeeper cluster. This must be the same ZooKeeper cluster that your Kafka cluster is using.

    5. The ZooKeeper session timeout, in milliseconds. The default is 18000 (18 seconds).

    6. The interval between periodic reconciliations, in milliseconds. The default is 120000 (2 minutes).

    7. The number of attempts at getting topic metadata from Kafka. The time between each attempt is defined as an exponential backoff. Consider increasing this value when topic creation takes more time due to the number of partitions or replicas. The default is 6 attempts.

    8. The level for printing logging messages. You can set the level to ERROR, WARNING, INFO, DEBUG, or TRACE.

    9. Enables TLS support for encrypted communication with the Kafka brokers.

    10. (Optional) The Java options used by the JVM running the Topic Operator.

    11. (Optional) The debugging (-D) options set for the Topic Operator.

    12. (Optional) Skips the generation of trust store certificates if TLS is enabled through STRIMZI_TLS_ENABLED. If this environment variable is enabled, the brokers must use a public trusted certificate authority for their TLS certificates. The default is false.

    13. (Optional) Generates key store certificates for mTLS authentication. Setting this to false disables client authentication with mTLS to the Kafka brokers. The default is true.

    14. (Optional) Enables SASL support for client authentication when connecting to Kafka brokers. The default is false.

    15. (Optional) The SASL username for client authentication. Mandatory only if SASL is enabled through STRIMZI_SASL_ENABLED.

    16. (Optional) The SASL password for client authentication. Mandatory only if SASL is enabled through STRIMZI_SASL_ENABLED.

    17. (Optional) The SASL mechanism for client authentication. Mandatory only if SASL is enabled through STRIMZI_SASL_ENABLED. You can set the value to plain, scram-sha-256, or scram-sha-512.

    18. (Optional) The security protocol used for communication with Kafka brokers. The default value is "PLAINTEXT". You can set the value to PLAINTEXT, SSL, SASL_PLAINTEXT, or SASL_SSL.

  2. If you want to connect to Kafka brokers that are using certificates from a public certificate authority, set STRIMZI_PUBLIC_CA to true. Set this property to true, for example, if you are using Amazon AWS MSK service.

  3. If you enabled mTLS with the STRIMZI_TLS_ENABLED environment variable, specify the keystore and truststore used to authenticate connection to the Kafka cluster.

    Example mTLS configuration
    # ....
env:
  - name: STRIMZI_TRUSTSTORE_LOCATION (1)
    value: "/path/to/truststore.p12"
  - name: STRIMZI_TRUSTSTORE_PASSWORD (2)
    value: "TRUSTSTORE-PASSWORD"
  - name: STRIMZI_KEYSTORE_LOCATION (3)
    value: "/path/to/keystore.p12"
  - name: STRIMZI_KEYSTORE_PASSWORD (4)
    value: "KEYSTORE-PASSWORD"
# ...

    1. The truststore contains the public keys of the Certificate Authorities used to sign the Kafka and ZooKeeper server certificates.

    2. The password for accessing the truststore.

    3. The keystore contains the private key for mTLS authentication.

    4. The password for accessing the keystore.

  4. Deploy the Topic Operator.

    kubectl create -f install/topic-operator

  5. Check the status of the deployment:

    kubectl get deployments
    Output shows the deployment name and readiness
    NAME                    READY  UP-TO-DATE  AVAILABLE
strimzi-topic-operator  1/1    1           1

    READY shows the number of replicas that are ready/expected. The deployment is successful when the AVAILABLE output shows 1.

5.7.2. Deploying the standalone User Operator

This procedure shows how to deploy the User Operator as a standalone component for user management. You can use a standalone User Operator with a Kafka cluster that is not managed by the Cluster Operator.

A standalone deployment can operate with any Kafka cluster.

Standalone deployment files are provided with Strimzi. Use the 05-Deployment-strimzi-user-operator.yaml deployment file to deploy the User Operator. Add or set the environment variables needed to make a connection to a Kafka cluster.

The User Operator watches for KafkaUser resources in a single namespace. You specify the namespace to watch, and the connection to the Kafka cluster, in the User Operator configuration. A single User Operator can watch a single namespace. One namespace should be watched by only one User Operator. If you want to use more than one User Operator, configure each of them to watch different namespaces. In this way, you can use the User Operator with multiple Kafka clusters.

Prerequisites

  • You are running a Kafka cluster for the User Operator to connect to.

    As long as the standalone User Operator is correctly configured for connection, the Kafka cluster can be running on a bare-metal environment, a virtual machine, or as a managed cloud application service.

Procedure

  1. Edit the following env properties in the install/user-operator/05-Deployment-strimzi-user-operator.yaml standalone deployment file.

    Example standalone User Operator deployment configuration
    apiVersion: apps/v1
kind: Deployment
metadata:
  name: strimzi-user-operator
  labels:
    app: strimzi
spec:
  # ...
  template:
    # ...
    spec:
      # ...
      containers:
        - name: strimzi-user-operator
          # ...
          env:
            - name: STRIMZI_NAMESPACE (1)
              valueFrom:
                fieldRef:
                  fieldPath: metadata.namespace
            - name: STRIMZI_KAFKA_BOOTSTRAP_SERVERS (2)
              value: my-kafka-bootstrap-address:9092
            - name: STRIMZI_CA_CERT_NAME (3)
              value: my-cluster-clients-ca-cert
            - name: STRIMZI_CA_KEY_NAME (4)
              value: my-cluster-clients-ca
            - name: STRIMZI_LABELS (5)
              value: "strimzi.io/cluster=my-cluster"
            - name: STRIMZI_FULL_RECONCILIATION_INTERVAL_MS (6)
              value: "120000"
            - name: STRIMZI_WORK_QUEUE_SIZE (7)
              value: 10000
            - name: STRIMZI_CONTROLLER_THREAD_POOL_SIZE (8)
              value: 10
            - name: STRIMZI_USER_OPERATIONS_THREAD_POOL_SIZE (9)
              value: 4
            - name: STRIMZI_LOG_LEVEL (10)
              value: INFO
            - name: STRIMZI_GC_LOG_ENABLED (11)
              value: "true"
            - name: STRIMZI_CA_VALIDITY (12)
              value: "365"
            - name: STRIMZI_CA_RENEWAL (13)
              value: "30"
            - name: STRIMZI_JAVA_OPTS (14)
              value: "-Xmx=512M -Xms=256M"
            - name: STRIMZI_JAVA_SYSTEM_PROPERTIES (15)
              value: "-Djavax.net.debug=verbose -DpropertyName=value"
            - name: STRIMZI_SECRET_PREFIX (16)
              value: "kafka-"
            - name: STRIMZI_ACLS_ADMIN_API_SUPPORTED (17)
              value: "true"
            - name: STRIMZI_MAINTENANCE_TIME_WINDOWS (18)
              value: '* * 8-10 * * ?;* * 14-15 * * ?'
            - name: STRIMZI_KAFKA_ADMIN_CLIENT_CONFIGURATION (19)
              value: |
                default.api.timeout.ms=120000
                request.timeout.ms=60000

    1. The Kubernetes namespace for the User Operator to watch for KafkaUser resources. Only one namespace can be specified.

    2. The host and port pair of the bootstrap broker address to discover and connect to all brokers in the Kafka cluster. Use a comma-separated list to specify two or three broker addresses in case a server is down.

    3. The Kubernetes Secret that contains the public key (ca.crt) value of the CA (certificate authority) that signs new user certificates for mTLS authentication.

    4. The Kubernetes Secret that contains the private key (ca.key) value of the CA that signs new user certificates for mTLS authentication.

    5. The label to identify the KafkaUser resources managed by the User Operator. This does not have to be the name of the Kafka cluster. It can be the label assigned to the KafkaUser resource. If you deploy more than one User Operator, the labels must be unique for each. That is, the operators cannot manage the same resources.

    6. The interval between periodic reconciliations, in milliseconds. The default is 120000 (2 minutes).

    7. The size of the controller event queue. The size of the queue should be at least as big as the maximal amount of users you expect the User Operator to operate. The default is 1024.

    8. The size of the worker pool for reconciling the users. Bigger pool might require more resources, but it will also handle more KafkaUser resources The default is 50.

    9. The size of the worker pool for Kafka Admin API and Kubernetes operations. Bigger pool might require more resources, but it will also handle more KafkaUser resources The default is 4.

    10. The level for printing logging messages. You can set the level to ERROR, WARNING, INFO, DEBUG, or TRACE.

    11. Enables garbage collection (GC) logging. The default is true.

    12. The validity period for the CA. The default is 365 days.

    13. The renewal period for the CA. The renewal period is measured backwards from the expiry date of the current certificate. The default is 30 days to initiate certificate renewal before the old certificates expire.

    14. (Optional) The Java options used by the JVM running the User Operator

    15. (Optional) The debugging (-D) options set for the User Operator

    16. (Optional) Prefix for the names of Kubernetes secrets created by the User Operator.

    17. (Optional) Indicates whether the Kafka cluster supports management of authorization ACL rules using the Kafka Admin API. When set to false, the User Operator will reject all resources with simple authorization ACL rules. This helps to avoid unnecessary exceptions in the Kafka cluster logs. The default is true.

    18. (Optional) Semi-colon separated list of Cron Expressions defining the maintenance time windows during which the expiring user certificates will be renewed.

    19. (Optional) Configuration options for configuring the Kafka Admin client used by the User Operator in the properties format.

  2. If you are using mTLS to connect to the Kafka cluster, specify the secrets used to authenticate connection. Otherwise, go to the next step.

    Example mTLS configuration
    # ....
env:
  - name: STRIMZI_CLUSTER_CA_CERT_SECRET_NAME (1)
    value: my-cluster-cluster-ca-cert
  - name: STRIMZI_EO_KEY_SECRET_NAME (2)
    value: my-cluster-entity-operator-certs
# ..."

    1. The Kubernetes Secret that contains the public key (ca.crt) value of the CA that signs Kafka broker certificates.

    2. The Kubernetes Secret that contains the keystore (entity-operator.p12) with the private key and certificate for mTLS authentication against the Kafka cluster. The Secret must also contain the password (entity-operator.password) for accessing the keystore.

  3. Deploy the User Operator.

    kubectl create -f install/user-operator

  4. Check the status of the deployment:

    kubectl get deployments
    Output shows the deployment name and readiness
    NAME                   READY  UP-TO-DATE  AVAILABLE
strimzi-user-operator  1/1    1           1

    READY shows the number of replicas that are ready/expected. The deployment is successful when the AVAILABLE output shows 1.

6. Deploying Strimzi from OperatorHub.io

OperatorHub.io is a catalog of Kubernetes operators sourced from multiple providers. It offers you an alternative way to install a stable version of Strimzi.

The Operator Lifecycle Manager is used for the installation and management of all operators published on OperatorHub.io. Operator Lifecycle Manager is a prerequisite for installing the Strimzi Kafka operator

To install Strimzi, locate Strimzi from OperatorHub.io, and follow the instructions provided to deploy the Cluster Operator. After you have deployed the Cluster Operator, you can deploy Strimzi components using custom resources. For example, you can deploy the Kafka custom resource, and the installed Cluster Operator will create a Kafka cluster.

Upgrades between versions might include manual steps. Always read the release notes before upgrading.

For information on upgrades, see Cluster Operator upgrade options.

Warning
 Make sure you use the appropriate update channel. Installing Strimzi from the default stable channel is generally safe. However, we do not recommend enabling automatic OLM updates on the stable channel. An automatic upgrade will skip any necessary steps prior to upgrade. For example, to upgrade from 0.22 or earlier you must first update custom resources to support the v1beta2 API version. Use automatic upgrades only on version-specific channels.

7. Deploying Strimzi using Helm

Helm charts are used to package, configure, and deploy Kubernetes resources. Strimzi provides a Helm chart to deploy the Cluster Operator.

After you have deployed the Cluster Operator this way, you can deploy Strimzi components using custom resources. For example, you can deploy the Kafka custom resource, and the installed Cluster Operator will create a Kafka cluster.

For information on upgrades, see Cluster Operator upgrade options.

Prerequisites

  • The Helm client must be installed on a local machine.

  • Helm must be installed to the Kubernetes cluster.

Procedure

  1. Add the Strimzi Helm Chart repository:

    helm repo add strimzi https://strimzi.io/charts/

  2. Deploy the Cluster Operator using the Helm command line tool:

    helm install strimzi-operator strimzi/strimzi-kafka-operator

  3. Verify that the Cluster Operator has been deployed successfully using the Helm command line tool:

    helm ls

  4. Deploy Kafka and other Kafka components using custom resources.

8. Setting up client access to a Kafka cluster

After you have deployed Strimzi, you can set up client access to your Kafka cluster. To verify the deployment, you can deploy example producer and consumer clients. Otherwise, create listeners that provide client access within or outside the Kubernetes cluster.

8.1. Deploying example clients

Deploy example producer and consumer clients to send and receive messages. You can use these clients to verify a deployment of Strimzi.

Prerequisites

  • The Kafka cluster is available for the clients.

Procedure

  1. Deploy a Kafka producer.

    kubectl run kafka-producer -ti --image=quay.io/strimzi/kafka:latest-kafka-3.3.2 --rm=true --restart=Never -- bin/kafka-console-producer.sh --bootstrap-server cluster-name-kafka-bootstrap:9092 --topic my-topic

  2. Type a message into the console where the producer is running.

  3. Press Enter to send the message.

  4. Deploy a Kafka consumer.

    kubectl run kafka-consumer -ti --image=quay.io/strimzi/kafka:latest-kafka-3.3.2 --rm=true --restart=Never -- bin/kafka-console-consumer.sh --bootstrap-server cluster-name-kafka-bootstrap:9092 --topic my-topic --from-beginning

  5. Confirm that you see the incoming messages in the consumer console.

8.2. Configuring listeners to connect to Kafka brokers

Use listeners for client connection to Kafka brokers. Strimzi provides a generic GenericKafkaListener schema with properties to configure listeners through the Kafka resource. The GenericKafkaListener provides a flexible approach to listener configuration. You can specify properties to configure internal listeners for connecting within the Kubernetes cluster or external listeners for connecting outside the Kubernetes cluster.

Specify a connection type to expose Kafka in the listener configuration. The type chosen depends on your requirements, and your environment and infrastructure. The following listener types are supported:

Internal listeners

  • internal to connect within the same Kubernetes cluster

  • cluster-ip to expose Kafka using per-broker ClusterIP services

External listeners

  • nodeport to use ports on Kubernetes nodes

  • loadbalancer to use loadbalancer services

  • ingress to use Kubernetes Ingress and the Ingress NGINX Controller for Kubernetes (Kubernetes only)

  • route to use OpenShift Route and the default HAProxy router (OpenShift only)

Important
 Do not use ingress on OpenShift, use the route type instead. The Ingress NGINX Controller is only intended for use on Kubernetes. The route type is only supported on OpenShift.

An internal type listener configuration uses a headless service and the DNS names given to the broker pods. You might want to join your Kubernetes network to an outside network. In which case, you can configure an internal type listener (using the useServiceDnsDomain property) so that the Kubernetes service DNS domain (typically .cluster.local) is not used. You can also configure a cluster-ip type of listener that exposes a Kafka cluster based on per-broker ClusterIP services. This is a useful option when you can’t route through the headless service or you wish to incorporate a custom access mechanism. For example, you might use this listener when building your own type of external listener for a specific Ingress controller or the Kubernetes Gateway API.

External listeners handle access to a Kafka cluster from networks that require different authentication mechanisms. You can configure external listeners for client access outside a Kubernetes environment using a specified connection mechanism, such as a loadbalancer or route. For example, loadbalancers might not be suitable for certain infrastructure, such as bare metal, where node ports provide a better option.

Each listener is defined as an array in the Kafka resource.

Example listener configuration
apiVersion: kafka.strimzi.io/v1beta2
kind: Kafka
metadata:
  name: my-cluster
spec:
  kafka:
    # ...
    listeners:
      - name: plain
        port: 9092
        type: internal
        tls: false
        configuration:
          useServiceDnsDomain: true
      - name: tls
        port: 9093
        type: internal
        tls: true
        authentication:
          type: tls
      - name: external
        port: 9094
        type: route
        tls: true
        configuration:
          brokerCertChainAndKey:
            secretName: my-secret
            certificate: my-certificate.crt
            key: my-key.key
    # ...

You can configure as many listeners as required, as long as their names and ports are unique. You can also configure listeners for secure connection using authentication.

If you want to know more about the pros and cons of each connection type, refer to Accessing Apache Kafka in Strimzi.

Note
 If you scale your Kafka cluster while using external listeners, it might trigger a rolling update of all Kafka brokers. This depends on the configuration.
Additional resources

8.3. Setting up client access to a Kafka cluster using listeners

Using the address of the Kafka cluster, you can provide access to a client in the same Kubernetes cluster; or provide external access to a client on a different Kubernetes namespace or outside Kubernetes entirely. This procedure shows how to configure client access to a Kafka cluster from outside Kubernetes or from another Kubernetes cluster.

A Kafka listener provides access to the Kafka cluster. Client access is secured using the following configuration:

  1. An external listener is configured for the Kafka cluster, with TLS encryption and mTLS authentication, and Kafka simple authorization enabled.

  2. A KafkaUser is created for the client, with mTLS authentication, and Access Control Lists (ACLs) defined for simple authorization.

You can configure your listener to use mutual tls, scram-sha-512, or oauth authentication. mTLS always uses encryption, but encryption is also recommended when using SCRAM-SHA-512 and OAuth 2.0 authentication.

You can configure simple, oauth, opa, or custom authorization for Kafka brokers. When enabled, authorization is applied to all enabled listeners.

When you configure the KafkaUser authentication and authorization mechanisms, ensure they match the equivalent Kafka configuration:

  • KafkaUser.spec.authentication matches Kafka.spec.kafka.listeners[*].authentication

  • KafkaUser.spec.authorization matches Kafka.spec.kafka.authorization

You should have at least one listener supporting the authentication you want to use for the KafkaUser.

Note
 Authentication between Kafka users and Kafka brokers depends on the authentication settings for each. For example, it is not possible to authenticate a user with mTLS if it is not also enabled in the Kafka configuration.

Strimzi operators automate the configuration process and create the certificates required for authentication:

  • The Cluster Operator creates the listeners and sets up the cluster and client certificate authority (CA) certificates to enable authentication with the Kafka cluster.

  • The User Operator creates the user representing the client and the security credentials used for client authentication, based on the chosen authentication type.

You add the certificates to your client configuration.

In this procedure, the CA certificates generated by the Cluster Operator are used, but you can replace them by installing your own certificates. You can also configure your listener to use a Kafka listener certificate managed by an external CA (certificate authority)^.

Certificates are available in PEM (.crt) and PKCS #12 (.p12) formats. This procedure uses PEM certificates. Use PEM certificates with clients that use certificates in X.509 format.

Note
 For internal clients in the same Kubernetes cluster and namespace, you can mount the cluster CA certificate in the pod specification. For more information, see Configuring internal clients to trust the cluster CA.
Prerequisites

  • The Kafka cluster is available for connection by a client running outside the Kubernetes cluster

  • The Cluster Operator and User Operator are running in the cluster

Procedure

  1. Configure the Kafka cluster with a Kafka listener.

    • Define the authentication required to access the Kafka broker through the listener.

    • Enable authorization on the Kafka broker.

      Example listener configuration
      apiVersion: kafka.strimzi.io/v1beta2
kind: Kafka
metadata:
  name: my-cluster
  namespace: myproject
spec:
  kafka:
    # ...
    listeners: # (1)
    - name: external # (2)
      port: 9094 (3)
      type: <listener_type> # (4)
      tls: true # (5)
      authentication:
        type: tls # (6)
      configuration: # (7)
        #...
    authorization: # (8)
      type: simple
      superUsers:
        - super-user-name # (9)
  # ...

      1. Configuration options for enabling external listeners are described in the Generic Kafka listener schema reference.

      2. Name to identify the listener. Must be unique within the Kafka cluster.

      3. Port number used by the listener inside Kafka. The port number has to be unique within a given Kafka cluster. Allowed port numbers are 9092 and higher with the exception of ports 9404 and 9999, which are already used for Prometheus and JMX. Depending on the listener type, the port number might not be the same as the port number that connects Kafka clients.

      4. External listener type specified as route, loadbalancer, nodeport or ingress. An internal listener is specified as internal or cluster-ip.

      5. Required. TLS encryption on the listener. For route and ingress type listeners it must be set to true. For mTLS authentication, also use the authentication property.

      6. Client authentication mechanism on the listener. For server and client authentication using mTLS, you specify tls: true and authentication.type: tls.

      7. (Optional) Depending on the requirements of the listener type, you can specify additional listener configuration.

      8. Authorization specified as simple, which uses the AclAuthorizer Kafka plugin.

      9. (Optional) Super users can access all brokers regardless of any access restrictions defined in ACLs.

      Warning
      		 An OpenShift Route address comprises the name of the Kafka cluster, the name of the listener, and the name of the namespace it is created in. For example, my-cluster-kafka-listener1-bootstrap-myproject (CLUSTER-NAME-kafka-LISTENER-NAME-bootstrap-NAMESPACE). If you are using a route listener type, be careful that the whole length of the address does not exceed a maximum limit of 63 characters.

  2. Create or update the Kafka resource.

    kubectl apply -f <kafka_configuration_file>

    The Kafka cluster is configured with a Kafka broker listener using mTLS authentication.

    A service is created for each Kafka broker pod.

    A service is created to serve as the bootstrap address for connection to the Kafka cluster.

    A service is also created as the external bootstrap address for external connection to the Kafka cluster using nodeport listeners.

    The cluster CA certificate to verify the identity of the kafka brokers is also created in the secret <cluster_name>-cluster-ca-cert.

    Note
    		 If you scale your Kafka cluster while using external listeners, it might trigger a rolling update of all Kafka brokers. This depends on the configuration.

  3. Retrieve the bootstrap address you can use to access the Kafka cluster from the status of the Kafka resource.

    kubectl get kafka <kafka_cluster_name> -o=jsonpath='{.status.listeners[?(@.name=="<listener_name>")].bootstrapServers}{"\n"}'

    For example:

    kubectl get kafka my-cluster -o=jsonpath='{.status.listeners[?(@.name=="external")].bootstrapServers}{"\n"}'

    Use the bootstrap address in your Kafka client to connect to the Kafka cluster.

  4. Create or modify a user representing the client that requires access to the Kafka cluster.

    • Specify the same authentication type as the Kafka listener.

    • Specify the authorization ACLs for simple authorization.

      Example user configuration
      apiVersion: kafka.strimzi.io/v1beta2
kind: KafkaUser
metadata:
  name: my-user
  labels:
    strimzi.io/cluster: my-cluster # (1)
spec:
  authentication:
    type: tls # (2)
  authorization:
    type: simple
    acls: # (3)
      - resource:
          type: topic
          name: my-topic
          patternType: literal
        operations:
          - Describe
          - Read
      - resource:
          type: group
          name: my-group
          patternType: literal
        operations:
          - Read

      1. The label must match the label of the Kafka cluster.

      2. Authentication specified as mutual tls.

      3. Simple authorization requires an accompanying list of ACL rules to apply to the user. The rules define the operations allowed on Kafka resources based on the username (my-user).

  5. Create or modify the KafkaUser resource.

    kubectl apply -f USER-CONFIG-FILE

    The user is created, as well as a secret with the same name as the KafkaUser resource. The secret contains a public and private key for mTLS authentication.

    Example secret
    apiVersion: v1
kind: Secret
metadata:
  name: my-user
  labels:
    strimzi.io/kind: KafkaUser
    strimzi.io/cluster: my-cluster
type: Opaque
data:
  ca.crt: <public_key> # Public key of the clients CA
  user.crt: <user_certificate> # Public key of the user
  user.key: <user_private_key> # Private key of the user
  user.p12: <store> # PKCS #12 store for user certificates and keys
  user.password: <password_for_store> # Protects the PKCS #12 store

  6. Extract the cluster CA certificate from the <cluster_name>-cluster-ca-cert secret of the Kafka cluster.

    kubectl get secret <cluster_name>-cluster-ca-cert -o jsonpath='{.data.ca\.crt}' | base64 -d > ca.crt

  7. Extract the user CA certificate from the <user_name> secret.

    kubectl get secret <user_name> -o jsonpath='{.data.user\.crt}' | base64 -d > user.crt

  8. Extract the private key of the user from the <user_name> secret.

    kubectl get secret <user_name> -o jsonpath='{.data.user\.key}' | base64 -d > user.key

  9. Configure your client with the bootstrap address hostname and port for connecting to the Kafka cluster:

    props.put(ConsumerConfig.BOOTSTRAP_SERVERS_CONFIG, "<hostname>:<port>");

  10. Configure your client with the truststore credentials to verify the identity of the Kafka cluster.

    Specify the public cluster CA certificate.

    Example truststore configuration
    props.put(CommonClientConfigs.SECURITY_PROTOCOL_CONFIG, "SSL");
props.put(SslConfigs.SSL_TRUSTSTORE_TYPE_CONFIG, "PEM");
props.put(SslConfigs.SSL_TRUSTSTORE_CERTIFICATES_CONFIG, "<ca.crt_file_content>");

    SSL is the specified security protocol for mTLS authentication. Specify SASL_SSL for SCRAM-SHA-512 authentication over TLS. PEM is the file format of the truststore.

  11. Configure your client with the keystore credentials to verify the user when connecting to the Kafka cluster.

    Specify the public certificate and private key.

    Example keystore configuration
    props.put(CommonClientConfigs.SECURITY_PROTOCOL_CONFIG, "SSL");
props.put(SslConfigs.SSL_KEYSTORE_TYPE_CONFIG, "PEM");
props.put(SslConfigs.SSL_KEYSTORE_CERTIFICATE_CHAIN_CONFIG, "<user.crt_file_content>");
props.put(SslConfigs.SSL_KEYSTORE_KEY_CONFIG, "<user.key_file_content>");

    Add the keystore certificate and the private key directly to the configuration. Add as a single-line format. Between the BEGIN CERTIFICATE and END CERTIFICATE delimiters, start with a newline character (\n). End each line from the original certificate with \n too.

    Example keystore configuration
    props.put(SslConfigs.SSL_KEYSTORE_CERTIFICATE_CHAIN_CONFIG, "-----BEGIN CERTIFICATE----- \n<user_certificate_content_line_1>\n<user_certificate_content_line_n>\n-----END CERTIFICATE---");
props.put(SslConfigs.SSL_KEYSTORE_KEY_CONFIG, "----BEGIN PRIVATE KEY-----\n<user_key_content_line_1>\n<user_key_content_line_n>\n-----END PRIVATE KEY-----");
    Additional resources

8.4. Accessing Kafka using node ports

This procedure describes how to access a Strimzi Kafka cluster from an external client using node ports.

To connect to a broker, you need a hostname and port number for the Kafka bootstrap address, as well as the certificate used for authentication.

Prerequisites

  • A Kubernetes cluster

  • A running Cluster Operator

Procedure

  1. Configure a Kafka resource with an external listener set to the nodeport type.

    For example:

    apiVersion: kafka.strimzi.io/v1beta2
kind: Kafka
spec:
  kafka:
    # ...
    listeners:
      - name: external
        port: 9094
        type: nodeport
        tls: true
        authentication:
          type: tls
        # ...
    # ...
  zookeeper:
    # ...

  2. Create or update the resource.

    kubectl apply -f <kafka_configuration_file>

    NodePort type services are created for each Kafka broker, as well as an external bootstrap service. The bootstrap service routes external traffic to the Kafka brokers. Node addresses used for connection are propagated to the status of the Kafka custom resource.

    The cluster CA certificate to verify the identity of the kafka brokers is also created in the secret <cluster_name>-cluster-ca-cert.

  3. Retrieve the bootstrap address you can use to access the Kafka cluster from the status of the Kafka resource.

    kubectl get kafka <kafka_cluster_name> -o=jsonpath='{.status.listeners[?(@.name=="<listener_name>")].bootstrapServers}{"\n"}'

    For example:

    kubectl get kafka my-cluster -o=jsonpath='{.status.listeners[?(@.name=="external")].bootstrapServers}{"\n"}'

  4. If TLS encryption is enabled, extract the public certificate of the broker certification authority.

    kubectl get secret KAFKA-CLUSTER-NAME-cluster-ca-cert -o jsonpath='{.data.ca\.crt}' | base64 -d > ca.crt

    Use the extracted certificate in your Kafka client to configure TLS connection. If you enabled any authentication, you will also need to configure it in your client.

8.5. Accessing Kafka using loadbalancers

This procedure describes how to access a Strimzi Kafka cluster from an external client using loadbalancers.

To connect to a broker, you need the address of the bootstrap loadbalancer, as well as the certificate used for TLS encryption.

Prerequisites

  • A Kubernetes cluster

  • A running Cluster Operator

Procedure

  1. Configure a Kafka resource with an external listener set to the loadbalancer type.

    For example:

    apiVersion: kafka.strimzi.io/v1beta2
kind: Kafka
spec:
  kafka:
    # ...
    listeners:
      - name: external
        port: 9094
        type: loadbalancer
        tls: true
        # ...
    # ...
  zookeeper:
    # ...

  2. Create or update the resource.

    kubectl apply -f <kafka_configuration_file>

    loadbalancer type services and loadbalancers are created for each Kafka broker, as well as an external bootstrap service. The bootstrap service routes external traffic to all Kafka brokers. DNS names and IP addresses used for connection are propagated to the status of each service.

    The cluster CA certificate to verify the identity of the kafka brokers is also created in the secret <cluster_name>-cluster-ca-cert.

  3. Retrieve the address of the bootstrap service you can use to access the Kafka cluster from the status of the Kafka resource.

    kubectl get kafka <kafka_cluster_name> -o=jsonpath='{.status.listeners[?(@.name=="<listener_name>")].bootstrapServers}{"\n"}'

    For example:

    kubectl get kafka my-cluster -o=jsonpath='{.status.listeners[?(@.name=="external")].bootstrapServers}{"\n"}'

  4. If TLS encryption is enabled, extract the public certificate of the broker certification authority.

    kubectl get secret KAFKA-CLUSTER-NAME-cluster-ca-cert -o jsonpath='{.data.ca\.crt}' | base64 -d > ca.crt

    Use the extracted certificate in your Kafka client to configure the TLS connection. If you enabled any authentication, you will also need to configure it in your client.

8.6. Accessing Kafka using an Ingress NGINX Controller for Kubernetes

Use an Ingress NGINX Controller for Kubernetes to access a Strimzi Kafka cluster from clients outside the Kubernetes cluster.

To be able to use an Ingress NGINX Controller for Kubernetes, add configuration for an ingress type listener in the Kafka custom resource. When applied, the configuration creates a dedicated ingress and service for an external bootstrap and each broker in the cluster. Clients connect to the bootstrap ingress, which routes them through the bootstrap service to connect to a broker. Per-broker connections are then established using DNS names, which route traffic from the client to the broker through the broker-specific ingresses and services.

To connect to a broker, you specify a hostname for the ingress bootstrap address, as well as the TLS certificate. Authentication is optional.

For access using an ingress, the port used in the Kafka client is typically 443.

TLS passthrough

Make sure that you enable TLS passthrough in your Ingress NGINX Controller for Kubernetes deployment. Kafka uses a binary protocol over TCP, but the Ingress NGINX Controller for Kubernetes is designed to work with a HTTP protocol. To be able to route TCP traffic through ingresses, Strimzi uses TLS passthrough with Server Name Indication (SNI).

SNI helps with identifying and passing connection to Kafka brokers. In passthrough mode, TLS encryption is always used. Because the connection passes to the brokers, the listeners use the TLS certificates signed by the internal cluster CA and not the ingress certificates. To configure listeners to use your own listener certificates, use the brokerCertChainAndKey property.

For more information about enabling TLS passthrough, see the TLS passthrough documentation.

Prerequisites

  • An Ingress NGINX Controller for Kubernetes is running with TLS passthrough enabled

  • A running Cluster Operator

In this procedure, the Kafka cluster name is my-cluster.

Procedure

  1. Configure a Kafka resource with an external listener set to the ingress type.

    Specify an ingress hostname for the bootstrap service and each of the Kafka brokers in the Kafka cluster. Add any hostname to the bootstrap and broker-<index> prefixes that identify the bootstrap and brokers.

    For example:

    apiVersion: kafka.strimzi.io/v1beta2
kind: Kafka
metadata:
  labels:
    app: my-cluster
  name: my-cluster
  namespace: myproject
spec:
  kafka:
    # ...
    listeners:
      - name: external
        port: 9094
        type: ingress
        tls: true # (1)
        authentication:
          type: tls
        configuration:
          bootstrap:
            host: bootstrap.myingress.com
          brokers:
          - broker: 0
            host: broker-0.myingress.com
          - broker: 1
            host: broker-1.myingress.com
          - broker: 2
            host: broker-2.myingress.com
          class: nginx  # (2)
    # ...
  zookeeper:
    # ...

    1. For ingress type listeners, TLS encryption must be enabled (true).

    2. (Optional) Class that specifies the ingress controller to use. You might need to add a class if you have not set up a default and a class name is missing in the ingresses created.

  2. Create or update the resource.

    kubectl apply -f <kafka_configuration_file>

    A cluster CA certificate to verify the identity of the kafka brokers is created in the secret my-cluster-cluster-ca-cert.

    ClusterIP type services are created for each Kafka broker, as well as an external bootstrap service.

    An ingress is also created for each service, with a DNS address to expose them using the Ingress NGINX Controller for Kubernetes.

    Ingresses created for the bootstrap and brokers
    NAME                        CLASS  HOSTS                    ADDRESS               PORTS
my-cluster-kafka-0          nginx  broker-0.myingress.com   external.ingress.com  80,443
my-cluster-kafka-1          nginx  broker-1.myingress.com   external.ingress.com  80,443
my-cluster-kafka-2          nginx  broker-2.myingress.com   external.ingress.com  80,443
my-cluster-kafka-bootstrap  nginx  bootstrap.myingress.com  external.ingress.com  80,443

    The DNS addresses used for client connection are propagated to the status of each ingress.

    Status for the bootstrap ingress
    status:
  loadBalancer:
    ingress:
      - hostname: external.ingress.com
 # ...

  3. Use a target broker to check the client-server TLS connection on port 443 using the OpenSSL s_client.

    openssl s_client -connect broker-0.myingress.com:443 -servername broker-0.myingress.com -showcerts

    The server name is the SNI for passing the connection to the broker.

    If the connection is successful, the certificates for the broker are returned.

    Certificates for the broker
    Certificate chain
 0 s:O = io.strimzi, CN = my-cluster-kafka
   i:O = io.strimzi, CN = cluster-ca v0

  4. Extract the cluster CA certificate.

    kubectl get secret my-cluster-cluster-ca-cert -o jsonpath='{.data.ca\.crt}' | base64 -d > ca.crt

  5. Configure your client to connect to the brokers.

    1. Specify the bootstrap host (from the listener configuration) and port 443 in your Kafka client as the bootstrap address to connect to the Kafka cluster. For example, bootstrap.myingress.com:443.

    2. Add the extracted certificate to the truststore of your Kafka client to configure a TLS connection.

      If you enabled any authentication, you will also need to configure it in your client.

Note
 If you are using your own listener certificates, check whether you need to add the CA certificate to the client’s truststore configuration. If it is a public (external) CA, you usually won’t need to add it.

8.7. Accessing Kafka using OpenShift routes

Use OpenShift routes to access a Strimzi Kafka cluster from clients outside the OpenShift cluster.

To be able to use routes, add configuration for a route type listener in the Kafka custom resource. When applied, the configuration creates a dedicated route and service for an external bootstrap and each broker in the cluster. Clients connect to the bootstrap route, which routes them through the bootstrap service to connect to a broker. Per-broker connections are then established using DNS names, which route traffic from the client to the broker through the broker-specific routes and services.

To connect to a broker, you specify a hostname for the route bootstrap address, as well as the certificate used for authentication.

For access using routes, the port is always 443.

Warning
 An OpenShift route address comprises the name of the Kafka cluster, the name of the listener, and the name of the project it is created in. For example, my-cluster-kafka-listener1-bootstrap-myproject (<cluster_name>-kafka-<listener_name>-bootstrap-<namespace>). Be careful that the whole length of the address does not exceed a maximum limit of 63 characters.
TLS passthrough

TLS passthrough is enabled for routes created by Strimzi. Kafka uses a binary protocol over TCP, but routes are designed to work with a HTTP protocol. To be able to route TCP traffic through routes, Strimzi uses TLS passthrough with Server Name Indication (SNI).

SNI helps with identifying and passing connection to Kafka brokers. In passthrough mode, TLS encryption is always used. Because the connection passes to the brokers, the listeners use TLS certificates signed by the internal cluster CA and not the ingress certificates. To configure listeners to use your own listener certificates, use the brokerCertChainAndKey property.

Prerequisites

  • A running Cluster Operator

In this procedure, the Kafka cluster name is my-cluster.

Procedure

  1. Configure a Kafka resource with an external listener set to the route type.

    For example:

    apiVersion: kafka.strimzi.io/v1beta2
kind: Kafka
metadata:
  labels:
    app: my-cluster
  name: my-cluster
  namespace: myproject
spec:
  kafka:
    # ...
    listeners:
      - name: listener1
        port: 9094
        type: route
        tls: true # (1)
        # ...
    # ...
  zookeeper:
    # ...

    1. For route type listeners, TLS encryption must be enabled (true).

  2. Create or update the resource.

    kubectl apply -f <kafka_configuration_file>

    A cluster CA certificate to verify the identity of the kafka brokers is created in the secret my-cluster-cluster-ca-cert.

    ClusterIP type services are created for each Kafka broker, as well as an external bootstrap service.

    A route is also created for each service, with a DNS address (host/port) to expose them using the default OpenShift HAProxy router.

    The routes are preconfigured with TLS passthrough.

    Routes created for the bootstraps and brokers
    NAME                                  HOST/PORT                                                   SERVICES                              PORT  TERMINATION
my-cluster-kafka-listener1-0          my-cluster-kafka-listener1-0-my-project.router.com          my-cluster-kafka-listener1-0          9094  passthrough
my-cluster-kafka-listener1-1          my-cluster-kafka-listener1-1-my-project.router.com          my-cluster-kafka-listener1-1          9094  passthrough
my-cluster-kafka-listener1-2          my-cluster-kafka-listener1-2-my-project.router.com          my-cluster-kafka-listener1-2          9094  passthrough
my-cluster-kafka-listener1-bootstrap  my-cluster-kafka-listener1-bootstrap-my-project.router.com  my-cluster-kafka-listener1-bootstrap  9094  passthrough

    The DNS addresses used for client connection are propagated to the status of each route.

    Example status for the bootstrap route
    status:
  ingress:
    - host: >-
        my-cluster-kafka-listener1-bootstrap-my-project.router.com
 # ...

  3. Use a target broker to check the client-server TLS connection on port 443 using the OpenSSL s_client.

    openssl s_client -connect my-cluster-kafka-listener1-0-my-project.router.com:443 -servername my-cluster-kafka-listener1-0-my-project.router.com -showcerts

    The server name is the SNI for passing the connection to the broker.

    If the connection is successful, the certificates for the broker are returned.

    Certificates for the broker
    Certificate chain
 0 s:O = io.strimzi, CN = my-cluster-kafka
   i:O = io.strimzi, CN = cluster-ca v0

  4. Retrieve the address of the bootstrap service from the status of the Kafka resource.

    kubectl get kafka my-cluster -o=jsonpath='{.status.listeners[?(@.name=="listener1")].bootstrapServers}{"\n"}'

my-cluster-kafka-listener1-bootstrap-my-project.router.com:443

    The address comprises the cluster name, the listener name, the project name and the domain of the router (router.com in this example).

  5. Extract the cluster CA certificate.

    kubectl get secret my-cluster-cluster-ca-cert -o jsonpath='{.data.ca\.crt}' | base64 -d > ca.crt

  6. Configure your client to connect to the brokers.

    1. Specify the address for the bootstrap service and port 443 in your Kafka client as the bootstrap address to connect to the Kafka cluster.

    2. Add the extracted certificate to the truststore of your Kafka client to configure a TLS connection.

      If you enabled any authentication, you will also need to configure it in your client.

Note
 If you are using your own listener certificates, check whether you need to add the CA certificate to the client’s truststore configuration. If it is a public (external) CA, you usually won’t need to add it.

9. Managing secure access to Kafka

Secure your Kafka cluster by managing the access a client has to Kafka brokers. Specify configuration options to secure Kafka brokers and clients

A secure connection between Kafka brokers and clients can encompass the following:

  • Encryption for data exchange

  • Authentication to prove identity

  • Authorization to allow or decline actions executed by users

The authentication and authorization mechanisms specified for a client must match those specified for the Kafka brokers. Strimzi operators automate the configuration process and create the certificates required for authentication.

9.1. Security options for Kafka

Use the Kafka resource to configure the mechanisms used for Kafka authentication and authorization.

9.1.1. Listener authentication

Configure client authentication for Kafka brokers when creating listeners. Specify the listener authentication type using the Kafka.spec.kafka.listeners.authentication property in the Kafka resource.

For clients inside the Kubernetes cluster, you can create plain (without encryption) or tls internal listeners. The internal listener type use a headless service and the DNS names given to the broker pods. As an alternative to the headless service, you can also create a cluster-ip type of internal listener to expose Kafka using per-broker ClusterIP services. For clients outside the Kubernetes cluster, you create external listeners and specify a connection mechanism, which can be nodeport, loadbalancer, ingress, or route (on OpenShift).

For more information on the configuration options for connecting an external client, see Setting up client access to a Kafka cluster.

Supported authentication options:

  1. mTLS authentication (only on the listeners with TLS enabled encryption)

  2. SCRAM-SHA-512 authentication

  3. OAuth 2.0 token-based authentication

  4. Custom authentication

The authentication option you choose depends on how you wish to authenticate client access to Kafka brokers.

Note
 Try exploring the standard authentication options before using custom authentication. Custom authentication allows for any type of kafka-supported authentication. It can provide more flexibility, but also adds complexity.
options for listener authentication configuration
Figure 1. Kafka listener authentication options

The listener authentication property is used to specify an authentication mechanism specific to that listener.

If no authentication property is specified then the listener does not authenticate clients which connect through that listener. The listener will accept all connections without authentication.

Authentication must be configured when using the User Operator to manage KafkaUsers.

The following example shows:

  • A plain listener configured for SCRAM-SHA-512 authentication

  • A tls listener with mTLS authentication

  • An external listener with mTLS authentication

Each listener is configured with a unique name and port within a Kafka cluster.

Note
 Listeners cannot be configured to use the ports reserved for inter-broker communication (9091 or 9090) and metrics (9404).
Example listener authentication configuration
apiVersion: kafka.strimzi.io/v1beta2
kind: Kafka
metadata:
  name: my-cluster
  namespace: myproject
spec:
  kafka:
    # ...
    listeners:
      - name: plain
        port: 9092
        type: internal
        tls: true
        authentication:
          type: scram-sha-512
      - name: tls
        port: 9093
        type: internal
        tls: true
        authentication:
          type: tls
      - name: external
        port: 9094
        type: loadbalancer
        tls: true
        authentication:
          type: tls
# ...
mTLS authentication

mTLS authentication is always used for the communication between Kafka brokers and ZooKeeper pods.

Strimzi can configure Kafka to use TLS (Transport Layer Security) to provide encrypted communication between Kafka brokers and clients either with or without mutual authentication. For mutual, or two-way, authentication, both the server and the client present certificates. When you configure mTLS authentication, the broker authenticates the client (client authentication) and the client authenticates the broker (server authentication).

mTLS listener configuration in the Kafka resource requires the following:

  • tls: true to specify TLS encryption and server authentication

  • authentication.type: tls to specify the client authentication

When a Kafka cluster is created by the Cluster Operator, it creates a new secret with the name <cluster_name>-cluster-ca-cert. The secret contains a CA certificate. The CA certificate is in PEM and PKCS #12 format. To verify a Kafka cluster, add the CA certificate to the truststore in your client configuration. To verify a client, add a user certificate and key to the keystore in your client configuration. For more information on configuring a client for mTLS, see User authentication.

Note
 TLS authentication is more commonly one-way, with one party authenticating the identity of another. For example, when HTTPS is used between a web browser and a web server, the browser obtains proof of the identity of the web server.
SCRAM-SHA-512 authentication

SCRAM (Salted Challenge Response Authentication Mechanism) is an authentication protocol that can establish mutual authentication using passwords. Strimzi can configure Kafka to use SASL (Simple Authentication and Security Layer) SCRAM-SHA-512 to provide authentication on both unencrypted and encrypted client connections.

When SCRAM-SHA-512 authentication is used with a TLS connection, the TLS protocol provides the encryption, but is not used for authentication.

The following properties of SCRAM make it safe to use SCRAM-SHA-512 even on unencrypted connections:

  • The passwords are not sent in the clear over the communication channel. Instead the client and the server are each challenged by the other to offer proof that they know the password of the authenticating user.

  • The server and client each generate a new challenge for each authentication exchange. This means that the exchange is resilient against replay attacks.

When KafkaUser.spec.authentication.type is configured with scram-sha-512 the User Operator will generate a random 12-character password consisting of upper and lowercase ASCII letters and numbers.

Network policies

By default, Strimzi automatically creates a NetworkPolicy resource for every listener that is enabled on a Kafka broker. This NetworkPolicy allows applications to connect to listeners in all namespaces. Use network policies as part of the listener configuration.

If you want to restrict access to a listener at the network level to only selected applications or namespaces, use the networkPolicyPeers property. Each listener can have a different networkPolicyPeers configuration. For more information on network policy peers, refer to the NetworkPolicyPeer API reference.

If you want to use custom network policies, you can set the STRIMZI_NETWORK_POLICY_GENERATION environment variable to false in the Cluster Operator configuration. For more information, see Cluster Operator configuration.

Note
 Your configuration of Kubernetes must support ingress NetworkPolicies in order to use network policies in Strimzi.
Providing listener certificates

You can provide your own server certificates, called Kafka listener certificates, for TLS listeners or external listeners which have TLS encryption enabled. For more information, see Providing your own Kafka listener certificates for TLS encryption.

Additional resources

9.1.2. Kafka authorization

Configure authorization for Kafka brokers using the Kafka.spec.kafka.authorization property in the Kafka resource. If the authorization property is missing, no authorization is enabled and clients have no restrictions. When enabled, authorization is applied to all enabled listeners. The authorization method is defined in the type field.

Supported authorization options:

options for kafka authorization configuration
Figure 2. Kafka cluster authorization options
Super users

Super users can access all resources in your Kafka cluster regardless of any access restrictions, and are supported by all authorization mechanisms.

To designate super users for a Kafka cluster, add a list of user principals to the superUsers property. If a user uses mTLS authentication, the username is the common name from the TLS certificate subject prefixed with CN=. If you are not using the User Operator and using your own certificates for mTLS, the username is the full certificate subject. A full certificate subject can have the following fields: CN=user,OU=my_ou,O=my_org,L=my_location,ST=my_state,C=my_country_code. Omit any fields that are not present.

An example configuration with super users
apiVersion: kafka.strimzi.io/v1beta2
kind: Kafka
metadata:
  name: my-cluster
  namespace: myproject
spec:
  kafka:
    # ...
    authorization:
      type: simple
      superUsers:
        - CN=client_1
        - user_2
        - CN=client_3
        - CN=client_4,OU=my_ou,O=my_org,L=my_location,ST=my_state,C=US
        - CN=client_5,OU=my_ou,O=my_org,C=GB
        - CN=client_6,O=my_org
    # ...

9.2. Security options for Kafka clients

Use the KafkaUser resource to configure the authentication mechanism, authorization mechanism, and access rights for Kafka clients. In terms of configuring security, clients are represented as users.

You can authenticate and authorize user access to Kafka brokers. Authentication permits access, and authorization constrains the access to permissible actions.

You can also create super users that have unconstrained access to Kafka brokers.

The authentication and authorization mechanisms must match the specification for the listener used to access the Kafka brokers.

Configuring users for secure access to Kafka brokers

For more information on configuring a KafkaUser resource to access Kafka brokers securely, see the following sections:

9.2.1. Identifying a Kafka cluster for user handling

A KafkaUser resource includes a label that defines the appropriate name of the Kafka cluster (derived from the name of the Kafka resource) to which it belongs.

apiVersion: kafka.strimzi.io/v1beta2
kind: KafkaUser
metadata:
  name: my-user
  labels:
    strimzi.io/cluster: my-cluster

The label is used by the User Operator to identify the KafkaUser resource and create a new user, and also in subsequent handling of the user.

If the label does not match the Kafka cluster, the User Operator cannot identify the KafkaUser and the user is not created.

If the status of the KafkaUser resource remains empty, check your label.

9.2.2. User authentication

Use the KafkaUser custom resource to configure authentication credentials for users (clients) that require access to a Kafka cluster. Configure the credentials using the authentication property in KafkaUser.spec. By specifying a type, you control what credentials are generated.

Supported authentication types:

  • tls for mTLS authentication

  • tls-external for mTLS authentication using external certificates

  • scram-sha-512 for SCRAM-SHA-512 authentication

If tls or scram-sha-512 is specified, the User Operator creates authentication credentials when it creates the user. If tls-external is specified, the user still uses mTLS, but no authentication credentials are created. Use this option when you’re providing your own certificates. When no authentication type is specified, the User Operator does not create the user or its credentials.

You can use tls-external to authenticate with mTLS using a certificate issued outside the User Operator. The User Operator does not generate a TLS certificate or a secret. You can still manage ACL rules and quotas through the User Operator in the same way as when you’re using the tls mechanism. This means that you use the CN=USER-NAME format when specifying ACL rules and quotas. USER-NAME is the common name given in a TLS certificate.

mTLS authentication

To use mTLS authentication, you set the type field in the KafkaUser resource to tls.

Example user with mTLS authentication enabled
apiVersion: kafka.strimzi.io/v1beta2
kind: KafkaUser
metadata:
  name: my-user
  labels:
    strimzi.io/cluster: my-cluster
spec:
  authentication:
    type: tls
  # ...

The authentication type must match the equivalent configuration for the Kafka listener used to access the Kafka cluster.

When the user is created by the User Operator, it creates a new secret with the same name as the KafkaUser resource. The secret contains a private and public key for mTLS. The public key is contained in a user certificate, which is signed by a clients CA (certificate authority) when it is created. All keys are in X.509 format.

Note
 If you are using the clients CA generated by the Cluster Operator, the user certificates generated by the User Operator are also renewed when the clients CA is renewed by the Cluster Operator.

The user secret provides keys and certificates in PEM and PKCS #12 formats.

Example secret with user credentials
apiVersion: v1
kind: Secret
metadata:
  name: my-user
  labels:
    strimzi.io/kind: KafkaUser
    strimzi.io/cluster: my-cluster
type: Opaque
data:
  ca.crt: <public_key> # Public key of the clients CA
  user.crt: <user_certificate> # Public key of the user
  user.key: <user_private_key> # Private key of the user
  user.p12: <store> # PKCS #12 store for user certificates and keys
  user.password: <password_for_store> # Protects the PKCS #12 store

When you configure a client, you specify the following:

  • Truststore properties for the public cluster CA certificate to verify the identity of the Kafka cluster

  • Keystore properties for the user authentication credentials to verify the client

The configuration depends on the file format (PEM or PKCS #12). This example uses PKCS #12 stores, and the passwords required to access the credentials in the stores.

Example client configuration using mTLS in PKCS #12 format
bootstrap.servers=<kafka_cluster_name>-kafka-bootstrap:9093 # (1)
security.protocol=SSL # (2)
ssl.truststore.location=/tmp/ca.p12 # (3)
ssl.truststore.password=<truststore_password> # (4)
ssl.keystore.location=/tmp/user.p12 # (5)
ssl.keystore.password=<keystore_password> # (6)

  1. The bootstrap server address to connect to the Kafka cluster.

  2. The security protocol option when using TLS for encryption.

  3. The truststore location contains the public key certificate (ca.p12) for the Kafka cluster. A cluster CA certificate and password is generated by the Cluster Operator in the <cluster_name>-cluster-ca-cert secret when the Kafka cluster is created.

  4. The password (ca.password) for accessing the truststore.

  5. The keystore location contains the public key certificate (user.p12) for the Kafka user.

  6. The password (user.password) for accessing the keystore.

mTLS authentication using a certificate issued outside the User Operator

To use mTLS authentication using a certificate issued outside the User Operator, you set the type field in the KafkaUser resource to tls-external. A secret and credentials are not created for the user.

Example user with mTLS authentication that uses a certificate issued outside the User Operator
apiVersion: kafka.strimzi.io/v1beta2
kind: KafkaUser
metadata:
  name: my-user
  labels:
    strimzi.io/cluster: my-cluster
spec:
  authentication:
    type: tls-external
  # ...
SCRAM-SHA-512 authentication

To use the SCRAM-SHA-512 authentication mechanism, you set the type field in the KafkaUser resource to scram-sha-512.

Example user with SCRAM-SHA-512 authentication enabled
apiVersion: kafka.strimzi.io/v1beta2
kind: KafkaUser
metadata:
  name: my-user
  labels:
    strimzi.io/cluster: my-cluster
spec:
  authentication:
    type: scram-sha-512
  # ...

When the user is created by the User Operator, it creates a new secret with the same name as the KafkaUser resource. The secret contains the generated password in the password key, which is encoded with base64. In order to use the password, it must be decoded.

Example secret with user credentials
apiVersion: v1
kind: Secret
metadata:
  name: my-user
  labels:
    strimzi.io/kind: KafkaUser
    strimzi.io/cluster: my-cluster
type: Opaque
data:
  password: Z2VuZXJhdGVkcGFzc3dvcmQ= (1)
  sasl.jaas.config: b3JnLmFwYWNoZS5rYWZrYS5jb21tb24uc2VjdXJpdHkuc2NyYW0uU2NyYW1Mb2dpbk1vZHVsZSByZXF1aXJlZCB1c2VybmFtZT0ibXktdXNlciIgcGFzc3dvcmQ9ImdlbmVyYXRlZHBhc3N3b3JkIjsK (2)

  1. The generated password, base64 encoded.

  2. The JAAS configuration string for SASL SCRAM-SHA-512 authentication, base64 encoded.

Decoding the generated password:

echo "Z2VuZXJhdGVkcGFzc3dvcmQ=" | base64 --decode
Custom password configuration

When a user is created, Strimzi generates a random password. You can use your own password instead of the one generated by Strimzi. To do so, create a secret with the password and reference it in the KafkaUser resource.

Example user with a password set for SCRAM-SHA-512 authentication
apiVersion: kafka.strimzi.io/v1beta2
kind: KafkaUser
metadata:
  name: my-user
  labels:
    strimzi.io/cluster: my-cluster
spec:
  authentication:
    type: scram-sha-512
    password:
      valueFrom:
        secretKeyRef:
          name: my-secret (1)
          key: my-password (2)
  # ...

  1. The name of the secret containing the predefined password.

  2. The key for the password stored inside the secret.

9.2.3. User authorization

Use the KafkaUser custom resource to configure authorization rules for users (clients) that require access to a Kafka cluster. Configure the rules using the authorization property in KafkaUser.spec. By specifying a type, you control what rules are used.

To use simple authorization, you set the type property to simple in KafkaUser.spec.authorization. The simple authorization uses the Kafka Admin API to manage the ACL rules inside your Kafka cluster. Whether ACL management in the User Operator is enabled or not depends on your authorization configuration in the Kafka cluster.

  • For simple authorization, ACL management is always enabled.

  • For OPA authorization, ACL management is always disabled. Authorization rules are configured in the OPA server.

  • For Keycloak authorization, you can manage the ACL rules directly in Keycloak. You can also delegate authorization to the simple authorizer as a fallback option in the configuration. When delegation to the simple authorizer is enabled, the User Operator will enable management of ACL rules as well.

  • For custom authorization using a custom authorization plugin, use the supportsAdminApi property in the .spec.kafka.authorization configuration of the Kafka custom resource to enable or disable the support.

Authorization is cluster-wide. The authorization type must match the equivalent configuration in the Kafka custom resource.

If ACL management is not enabled, Strimzi rejects a resource if it contains any ACL rules.

If you’re using a standalone deployment of the User Operator, ACL management is enabled by default. You can disable it using the STRIMZI_ACLS_ADMIN_API_SUPPORTED environment variable.

If no authorization is specified, the User Operator does not provision any access rights for the user. Whether such a KafkaUser can still access resources depends on the authorizer being used. For example, for the AclAuthorizer this is determined by its allow.everyone.if.no.acl.found configuration.

ACL rules

AclAuthorizer uses ACL rules to manage access to Kafka brokers.

ACL rules grant access rights to the user, which you specify in the acls property.

For more information about the AclRule object, see the AclRule schema reference.

Super user access to Kafka brokers

If a user is added to a list of super users in a Kafka broker configuration, the user is allowed unlimited access to the cluster regardless of any authorization constraints defined in ACLs in KafkaUser.

For more information on configuring super user access to brokers, see Kafka authorization.

User quotas

You can configure the spec for the KafkaUser resource to enforce quotas so that a user does not exceed a configured level of access to Kafka brokers. You can set size-based network usage and time-based CPU utilization thresholds. You can also add a partition mutation quota to control the rate at which requests to change partitions are accepted for user requests.

An example KafkaUser with user quotas
apiVersion: kafka.strimzi.io/v1beta2
kind: KafkaUser
metadata:
  name: my-user
  labels:
    strimzi.io/cluster: my-cluster
spec:
  # ...
  quotas:
    producerByteRate: 1048576 (1)
    consumerByteRate: 2097152 (2)
    requestPercentage: 55 (3)
    controllerMutationRate: 10 (4)

  1. Byte-per-second quota on the amount of data the user can push to a Kafka broker

  2. Byte-per-second quota on the amount of data the user can fetch from a Kafka broker

  3. CPU utilization limit as a percentage of time for a client group

  4. Number of concurrent partition creation and deletion operations (mutations) allowed per second

For more information on these properties, see the KafkaUserQuotas schema reference.

9.3. Securing access to Kafka brokers

To establish secure access to Kafka brokers, you configure and apply:

  • A Kafka resource to:

    • Create listeners with a specified authentication type

    • Configure authorization for the whole Kafka cluster

  • A KafkaUser resource to access the Kafka brokers securely through the listeners

Configure the Kafka resource to set up:

  • Listener authentication

  • Network policies that restrict access to Kafka listeners

  • Kafka authorization

  • Super users for unconstrained access to brokers

Authentication is configured independently for each listener. Authorization is always configured for the whole Kafka cluster.

The Cluster Operator creates the listeners and sets up the cluster and client certificate authority (CA) certificates to enable authentication within the Kafka cluster.

You can replace the certificates generated by the Cluster Operator by installing your own certificates.

You can also provide your own server certificates and private keys for any listener with TLS encryption enabled. These user-provided certificates are called Kafka listener certificates. Providing Kafka listener certificates allows you to leverage existing security infrastructure, such as your organization’s private CA or a public CA. Kafka clients will need to trust the CA which was used to sign the listener certificate. You must manually renew Kafka listener certificates when needed. Certificates are available in PKCS #12 format (.p12) and PEM (.crt) formats.

Use KafkaUser to enable the authentication and authorization mechanisms that a specific client uses to access Kafka.

Configure the KafkaUser resource to set up:

  • Authentication to match the enabled listener authentication

  • Authorization to match the enabled Kafka authorization

  • Quotas to control the use of resources by clients

The User Operator creates the user representing the client and the security credentials used for client authentication, based on the chosen authentication type.

Refer to the schema reference for more information on access configuration properties:

9.3.1. Securing Kafka brokers

This procedure shows the steps involved in securing Kafka brokers when running Strimzi.

The security implemented for Kafka brokers must be compatible with the security implemented for the clients requiring access.

  • Kafka.spec.kafka.listeners[*].authentication matches KafkaUser.spec.authentication

  • Kafka.spec.kafka.authorization matches KafkaUser.spec.authorization

The steps show the configuration for simple authorization and a listener using mTLS authentication. For more information on listener configuration, see the GenericKafkaListener schema reference.

Alternatively, you can use SCRAM-SHA or OAuth 2.0 for listener authentication, and OAuth 2.0 or OPA for Kafka authorization.

Procedure

  1. Configure the Kafka resource.

    1. Configure the authorization property for authorization.

    2. Configure the listeners property to create a listener with authentication.

      For example:

      apiVersion: kafka.strimzi.io/v1beta2
kind: Kafka
spec:
  kafka:
    # ...
    authorization: (1)
      type: simple
      superUsers: (2)
        - CN=client_1
        - user_2
        - CN=client_3
    listeners:
      - name: tls
        port: 9093
        type: internal
        tls: true
        authentication:
          type: tls (3)
    # ...
  zookeeper:
    # ...

      1. Authorization enables simple authorization on the Kafka broker using the AclAuthorizer Kafka plugin.

      2. List of user principals with unlimited access to Kafka. CN is the common name from the client certificate when mTLS authentication is used.

      3. Listener authentication mechanisms may be configured for each listener, and specified as mTLS, SCRAM-SHA-512, or token-based OAuth 2.0.

      If you are configuring an external listener, the configuration is dependent on the chosen connection mechanism.

  2. Create or update the Kafka resource.

    kubectl apply -f <kafka_configuration_file>

    The Kafka cluster is configured with a Kafka broker listener using mTLS authentication.

    A service is created for each Kafka broker pod.

    A service is created to serve as the bootstrap address for connection to the Kafka cluster.

    The cluster CA certificate to verify the identity of the kafka brokers is also created in the secret <cluster_name>-cluster-ca-cert.

9.3.2. Securing user access to Kafka

Create or modify a KafkaUser to represent a client that requires secure access to the Kafka cluster.

When you configure the KafkaUser authentication and authorization mechanisms, ensure they match the equivalent Kafka configuration:

  • KafkaUser.spec.authentication matches Kafka.spec.kafka.listeners[*].authentication

  • KafkaUser.spec.authorization matches Kafka.spec.kafka.authorization

This procedure shows how a user is created with mTLS authentication. You can also create a user with SCRAM-SHA authentication.

The authentication required depends on the type of authentication configured for the Kafka broker listener.

Note
 Authentication between Kafka users and Kafka brokers depends on the authentication settings for each. For example, it is not possible to authenticate a user with mTLS if it is not also enabled in the Kafka configuration.
Prerequisites

The authentication type in KafkaUser should match the authentication configured in Kafka brokers.

Procedure

  1. Configure the KafkaUser resource.

    For example:

    apiVersion: kafka.strimzi.io/v1beta2
kind: KafkaUser
metadata:
  name: my-user
  labels:
    strimzi.io/cluster: my-cluster
spec:
  authentication: (1)
    type: tls
  authorization:
    type: simple (2)
    acls:
      - resource:
          type: topic
          name: my-topic
          patternType: literal
        operations:
          - Describe
          - Read
      - resource:
          type: group
          name: my-group
          patternType: literal
        operations:
          - Read

    1. User authentication mechanism, defined as mutual tls or scram-sha-512.

    2. Simple authorization, which requires an accompanying list of ACL rules.

  2. Create or update the KafkaUser resource.

    kubectl apply -f <user_config_file>

    The user is created, as well as a Secret with the same name as the KafkaUser resource. The Secret contains a private and public key for mTLS authentication.

For information on configuring a Kafka client with properties for secure connection to Kafka brokers, see Setting up client access to a Kafka cluster using listeners.

9.3.3. Restricting access to Kafka listeners using network policies

You can restrict access to a listener to only selected applications by using the networkPolicyPeers property.

Prerequisites

  • A Kubernetes cluster with support for Ingress NetworkPolicies.

  • The Cluster Operator is running.

Procedure

  1. Open the Kafka resource.

  2. In the networkPolicyPeers property, define the application pods or namespaces that will be allowed to access the Kafka cluster.

    For example, to configure a tls listener to allow connections only from application pods with the label app set to kafka-client:

    apiVersion: kafka.strimzi.io/v1beta2
kind: Kafka
spec:
  kafka:
    # ...
    listeners:
      - name: tls
        port: 9093
        type: internal
        tls: true
        authentication:
          type: tls
        networkPolicyPeers:
          - podSelector:
              matchLabels:
                app: kafka-client
    # ...
  zookeeper:
    # ...

  3. Create or update the resource.

    Use kubectl apply:

    kubectl apply -f your-file
Additional resources

ink:./configuring.html# * networkPolicyPeers configuration * NetworkPolicyPeer API reference

9.3.4. Providing your own Kafka listener certificates for TLS encryption

Listeners provide client access to Kafka brokers. Configure listeners in the Kafka resource, including the configuration required for client access using TLS.

By default, the listeners use certificates signed by the internal CA (certificate authority) certificates generated by Strimzi. A CA certificate is generated by the Cluster Operator when it creates a Kafka cluster. When you configure a client for TLS, you add the CA certificate to its truststore configuration to verify the Kafka cluster. You can also install and use your own CA certificates. Or you can configure a listener using brokerCertChainAndKey properties and use a custom server certificate.

The brokerCertChainAndKey properties allow you to access Kafka brokers using your own custom certificates at the listener-level. You create a secret with your own private key and server certificate, then specify the key and certificate in the listener’s brokerCertChainAndKey configuration. You can use a certificate signed by a public (external) CA or a private CA. If signed by a public CA, you usually won’t need to add it to a client’s truststore configuration. Custom certificates are not managed by Strimzi, so you need to renew them manually.

Note
 Listener certificates are used for TLS encryption and server authentication only. They are not used for TLS client authentication. If you want to use your own certificate for TLS client authentication as well, you must install and use your own clients CA.
Prerequisites

  • The Cluster Operator is running.

  • Each listener requires the following:

    • A compatible server certificate signed by an external CA. (Provide an X.509 certificate in PEM format.)

      You can use one listener certificate for multiple listeners.

    • Subject Alternative Names (SANs) are specified in the certificate for each listener. For more information, see Alternative subjects in server certificates for Kafka listeners.

If you are not using a self-signed certificate, you can provide a certificate that includes the whole CA chain in the certificate.

You can only use the brokerCertChainAndKey properties if TLS encryption (tls: true) is configured for the listener.

Procedure

  1. Create a Secret containing your private key and server certificate:

    kubectl create secret generic my-secret --from-file=my-listener-key.key --from-file=my-listener-certificate.crt

  2. Edit the Kafka resource for your cluster.

    Configure the listener to use your Secret, certificate file, and private key file in the configuration.brokerCertChainAndKey property.

    Example configuration for a loadbalancer external listener with TLS encryption enabled
    # ...
listeners:
  - name: plain
    port: 9092
    type: internal
    tls: false
  - name: external
    port: 9094
    type: loadbalancer
    tls: true
    configuration:
      brokerCertChainAndKey:
        secretName: my-secret
        certificate: my-listener-certificate.crt
        key: my-listener-key.key
# ...
    Example configuration for a TLS listener
    # ...
listeners:
  - name: plain
    port: 9092
    type: internal
    tls: false
  - name: tls
    port: 9093
    type: internal
    tls: true
    configuration:
      brokerCertChainAndKey:
        secretName: my-secret
        certificate: my-listener-certificate.crt
        key: my-listener-key.key
# ...

  3. Apply the new configuration to create or update the resource:

    kubectl apply -f kafka.yaml

    The Cluster Operator starts a rolling update of the Kafka cluster, which updates the configuration of the listeners.

    Note
    		 A rolling update is also started if you update a Kafka listener certificate in a Secret that is already used by a listener.

9.3.5. Alternative subjects in server certificates for Kafka listeners

In order to use TLS hostname verification with your own Kafka listener certificates, you must use the correct Subject Alternative Names (SANs) for each listener. The certificate SANs must specify hostnames for the following:

  • All of the Kafka brokers in your cluster

  • The Kafka cluster bootstrap service

You can use wildcard certificates if they are supported by your CA.

TLS listener SAN examples

Use the following examples to help you specify hostnames of the SANs in your certificates for TLS listeners.

Wildcards example
//Kafka brokers
*.<cluster-name>-kafka-brokers
*.<cluster-name>-kafka-brokers.<namespace>.svc

// Bootstrap service
<cluster-name>-kafka-bootstrap
<cluster-name>-kafka-bootstrap.<namespace>.svc
Non-wildcards example
// Kafka brokers
<cluster-name>-kafka-0.<cluster-name>-kafka-brokers
<cluster-name>-kafka-0.<cluster-name>-kafka-brokers.<namespace>.svc
<cluster-name>-kafka-1.<cluster-name>-kafka-brokers
<cluster-name>-kafka-1.<cluster-name>-kafka-brokers.<namespace>.svc
# ...

// Bootstrap service
<cluster-name>-kafka-bootstrap
<cluster-name>-kafka-bootstrap.<namespace>.svc
External listener SAN examples

For external listeners which have TLS encryption enabled, the hostnames you need to specify in certificates depends on the external listener type.

Table 3. SANs for each type of external listener
External listener type In the SANs, specify…​

Route

Addresses of all Kafka broker Routes and the address of the bootstrap Route.

You can use a matching wildcard name.

loadbalancer

Addresses of all Kafka broker loadbalancers and the bootstrap loadbalancer address.

You can use a matching wildcard name.

NodePort

Addresses of all Kubernetes worker nodes that the Kafka broker pods might be scheduled to.

You can use a matching wildcard name.
Additional resources

9.4. Using OAuth 2.0 token-based authentication

Strimzi supports the use of OAuth 2.0 authentication using the OAUTHBEARER and PLAIN mechanisms.

OAuth 2.0 enables standardized token-based authentication and authorization between applications, using a central authorization server to issue tokens that grant limited access to resources.

Kafka brokers and clients both need to be configured to use OAuth 2.0. You can configure OAuth 2.0 authentication, then OAuth 2.0 authorization.

Note

OAuth 2.0 authentication can be used in conjunction with Kafka authorization.

Using OAuth 2.0 authentication, application clients can access resources on application servers (called resource servers) without exposing account credentials.

The application client passes an access token as a means of authenticating, which application servers can also use to determine the level of access to grant. The authorization server handles the granting of access and inquiries about access.

In the context of Strimzi:

  • Kafka brokers act as OAuth 2.0 resource servers

  • Kafka clients act as OAuth 2.0 application clients

Kafka clients authenticate to Kafka brokers. The brokers and clients communicate with the OAuth 2.0 authorization server, as necessary, to obtain or validate access tokens.

For a deployment of Strimzi, OAuth 2.0 integration provides:

  • Server-side OAuth 2.0 support for Kafka brokers

  • Client-side OAuth 2.0 support for Kafka MirrorMaker, Kafka Connect, and the Kafka Bridge

9.4.1. OAuth 2.0 authentication mechanisms

Strimzi supports the OAUTHBEARER and PLAIN mechanisms for OAuth 2.0 authentication. Both mechanisms allow Kafka clients to establish authenticated sessions with Kafka brokers. The authentication flow between clients, the authorization server, and Kafka brokers is different for each mechanism.

We recommend that you configure clients to use OAUTHBEARER whenever possible. OAUTHBEARER provides a higher level of security than PLAIN because client credentials are never shared with Kafka brokers. Consider using PLAIN only with Kafka clients that do not support OAUTHBEARER.

You configure Kafka broker listeners to use OAuth 2.0 authentication for connecting clients. If necessary, you can use the OAUTHBEARER and PLAIN mechanisms on the same oauth listener. The properties to support each mechanism must be explicitly specified in the oauth listener configuration.

OAUTHBEARER overview

OAUTHBEARER is automatically enabled in the oauth listener configuration for the Kafka broker. You can set the enableOauthBearer property to true, though this is not required.

  # ...
  authentication:
    type: oauth
    # ...
    enableOauthBearer: true

Many Kafka client tools use libraries that provide basic support for OAUTHBEARER at the protocol level. To support application development, Strimzi provides an OAuth callback handler for the upstream Kafka Client Java libraries (but not for other libraries). Therefore, you do not need to write your own callback handlers. An application client can use the callback handler to provide the access token. Clients written in other languages, such as Go, must use custom code to connect to the authorization server and obtain the access token.

With OAUTHBEARER, the client initiates a session with the Kafka broker for credentials exchange, where credentials take the form of a bearer token provided by the callback handler. Using the callbacks, you can configure token provision in one of three ways:

  • Client ID and Secret (by using the OAuth 2.0 client credentials mechanism)

  • A long-lived access token, obtained manually at configuration time

  • A long-lived refresh token, obtained manually at configuration time

Note

OAUTHBEARER authentication can only be used by Kafka clients that support the OAUTHBEARER mechanism at the protocol level.
PLAIN overview

To use PLAIN, you must enable it in the oauth listener configuration for the Kafka broker.

In the following example, PLAIN is enabled in addition to OAUTHBEARER, which is enabled by default. If you want to use PLAIN only, you can disable OAUTHBEARER by setting enableOauthBearer to false.

  # ...
  authentication:
    type: oauth
    # ...
    enablePlain: true
    tokenEndpointUri: https://OAUTH-SERVER-ADDRESS/auth/realms/external/protocol/openid-connect/token

PLAIN is a simple authentication mechanism used by all Kafka client tools. To enable PLAIN to be used with OAuth 2.0 authentication, Strimzi provides OAuth 2.0 over PLAIN server-side callbacks.

With the Strimzi implementation of PLAIN, the client credentials are not stored in ZooKeeper. Instead, client credentials are handled centrally behind a compliant authorization server, similar to when OAUTHBEARER authentication is used.

When used with the OAuth 2.0 over PLAIN callbacks, Kafka clients authenticate with Kafka brokers using either of the following methods:

  • Client ID and secret (by using the OAuth 2.0 client credentials mechanism)

  • A long-lived access token, obtained manually at configuration time

For both methods, the client must provide the PLAIN username and password properties to pass credentials to the Kafka broker. The client uses these properties to pass a client ID and secret or username and access token.

Client IDs and secrets are used to obtain access tokens.

Access tokens are passed as password property values. You pass the access token with or without an $accessToken: prefix.

  • If you configure a token endpoint (tokenEndpointUri) in the listener configuration, you need the prefix.

  • If you don’t configure a token endpoint (tokenEndpointUri) in the listener configuration, you don’t need the prefix. The Kafka broker interprets the password as a raw access token.

If the password is set as the access token, the username must be set to the same principal name that the Kafka broker obtains from the access token. You can specify username extraction options in your listener using the userNameClaim, fallbackUserNameClaim, fallbackUsernamePrefix, and userInfoEndpointUri properties. The username extraction process also depends on your authorization server; in particular, how it maps client IDs to account names.

Note

OAuth over PLAIN does not support password grant mechanism. You can only 'proxy' through SASL PLAIN mechanism the client credentials (clientId + secret) or the access token as described above.
Additional resources

9.4.2. OAuth 2.0 Kafka broker configuration

Kafka broker configuration for OAuth 2.0 involves:

  • Creating the OAuth 2.0 client in the authorization server

  • Configuring OAuth 2.0 authentication in the Kafka custom resource

Note
 In relation to the authorization server, Kafka brokers and Kafka clients are both regarded as OAuth 2.0 clients.
OAuth 2.0 client configuration on an authorization server

To configure a Kafka broker to validate the token received during session initiation, the recommended approach is to create an OAuth 2.0 client definition in an authorization server, configured as confidential, with the following client credentials enabled:

  • Client ID of kafka (for example)

  • Client ID and Secret as the authentication mechanism

Note
 You only need to use a client ID and secret when using a non-public introspection endpoint of the authorization server. The credentials are not typically required when using public authorization server endpoints, as with fast local JWT token validation.
OAuth 2.0 authentication configuration in the Kafka cluster

To use OAuth 2.0 authentication in the Kafka cluster, you specify, for example, a tls listener configuration for your Kafka cluster custom resource with the authentication method oauth:

Assigining the authentication method type for OAuth 2.0
apiVersion: kafka.strimzi.io/v1beta2
kind: Kafka
spec:
  kafka:
    # ...
    listeners:
      - name: tls
        port: 9093
        type: internal
        tls: true
        authentication:
          type: oauth
      #...

You can configure OAuth 2.0 authentication in your listeners. We recommend using OAuth 2.0 authentication together with TLS encryption (tls: true). Without encryption, the connection is vulnerable to network eavesdropping and unauthorized access through token theft.

You configure an external listener with type: oauth for a secure transport layer to communicate with the client.

Using OAuth 2.0 with an external listener
# ...
listeners:
  - name: external
    port: 9094
    type: loadbalancer
    tls: true
    authentication:
      type: oauth
    #...

The tls property is false by default, so it must be enabled.

When you have defined the type of authentication as OAuth 2.0, you add configuration based on the type of validation, either as fast local JWT validation or token validation using an introspection endpoint.

The procedure to configure OAuth 2.0 for listeners, with descriptions and examples, is described in Configuring OAuth 2.0 support for Kafka brokers.

Fast local JWT token validation configuration

Fast local JWT token validation checks a JWT token signature locally.

The local check ensures that a token:

  • Conforms to type by containing a (typ) claim value of Bearer for an access token

  • Is valid (not expired)

  • Has an issuer that matches a validIssuerURI

You specify a validIssuerURI attribute when you configure the listener, so that any tokens not issued by the authorization server are rejected.

The authorization server does not need to be contacted during fast local JWT token validation. You activate fast local JWT token validation by specifying a jwksEndpointUri attribute, the endpoint exposed by the OAuth 2.0 authorization server. The endpoint contains the public keys used to validate signed JWT tokens, which are sent as credentials by Kafka clients.

Note
 All communication with the authorization server should be performed using TLS encryption.

You can configure a certificate truststore as a Kubernetes Secret in your Strimzi project namespace, and use a tlsTrustedCertificates attribute to point to the Kubernetes Secret containing the truststore file.

You might want to configure a userNameClaim to properly extract a username from the JWT token. If you want to use Kafka ACL authorization, you need to identify the user by their username during authentication. (The sub claim in JWT tokens is typically a unique ID, not a username.)

Example configuration for fast local JWT token validation
apiVersion: kafka.strimzi.io/v1beta2
kind: Kafka
spec:
  kafka:
    #...
    listeners:
      - name: tls
        port: 9093
        type: internal
        tls: true
        authentication:
          type: oauth
          validIssuerUri: <https://<auth-server-address>/auth/realms/tls>
          jwksEndpointUri: <https://<auth-server-address>/auth/realms/tls/protocol/openid-connect/certs>
          userNameClaim: preferred_username
          maxSecondsWithoutReauthentication: 3600
          tlsTrustedCertificates:
          - secretName: oauth-server-cert
            certificate: ca.crt
    #...
OAuth 2.0 introspection endpoint configuration

Token validation using an OAuth 2.0 introspection endpoint treats a received access token as opaque. The Kafka broker sends an access token to the introspection endpoint, which responds with the token information necessary for validation. Importantly, it returns up-to-date information if the specific access token is valid, and also information about when the token expires.

To configure OAuth 2.0 introspection-based validation, you specify an introspectionEndpointUri attribute rather than the jwksEndpointUri attribute specified for fast local JWT token validation. Depending on the authorization server, you typically have to specify a clientId and clientSecret, because the introspection endpoint is usually protected.

Example configuration for an introspection endpoint
apiVersion: kafka.strimzi.io/v1beta2
kind: Kafka
spec:
  kafka:
    listeners:
      - name: tls
        port: 9093
        type: internal
        tls: true
        authentication:
          type: oauth
          clientId: kafka-broker
          clientSecret:
            secretName: my-cluster-oauth
            key: clientSecret
          validIssuerUri: <https://<auth-server-address>/auth/realms/tls>
          introspectionEndpointUri: <https://<auth-server-address>/auth/realms/tls/protocol/openid-connect/token/introspect>
          userNameClaim: preferred_username
          maxSecondsWithoutReauthentication: 3600
          tlsTrustedCertificates:
          - secretName: oauth-server-cert
            certificate: ca.crt

9.4.3. Session re-authentication for Kafka brokers

You can configure oauth listeners to use Kafka session re-authentication for OAuth 2.0 sessions between Kafka clients and Kafka brokers. This mechanism enforces the expiry of an authenticated session between the client and the broker after a defined period of time. When a session expires, the client immediately starts a new session by reusing the existing connection rather than dropping it.

Session re-authentication is disabled by default. To enable it, you set a time value for maxSecondsWithoutReauthentication in the oauth listener configuration. The same property is used to configure session re-authentication for OAUTHBEARER and PLAIN authentication. For an example configuration, see Configuring OAuth 2.0 support for Kafka brokers.

Session re-authentication must be supported by the Kafka client libraries used by the client.

Session re-authentication can be used with fast local JWT or introspection endpoint token validation.

Client re-authentication

When the broker’s authenticated session expires, the client must re-authenticate to the existing session by sending a new, valid access token to the broker, without dropping the connection.

If token validation is successful, a new client session is started using the existing connection. If the client fails to re-authenticate, the broker will close the connection if further attempts are made to send or receive messages. Java clients that use Kafka client library 2.2 or later automatically re-authenticate if the re-authentication mechanism is enabled on the broker.

Session re-authentication also applies to refresh tokens, if used. When the session expires, the client refreshes the access token by using its refresh token. The client then uses the new access token to re-authenticate to the existing session.

Session expiry for OAUTHBEARER and PLAIN

When session re-authentication is configured, session expiry works differently for OAUTHBEARER and PLAIN authentication.

For OAUTHBEARER and PLAIN, using the client ID and secret method:

  • The broker’s authenticated session will expire at the configured maxSecondsWithoutReauthentication.

  • The session will expire earlier if the access token expires before the configured time.

For PLAIN using the long-lived access token method:

  • The broker’s authenticated session will expire at the configured maxSecondsWithoutReauthentication.

  • Re-authentication will fail if the access token expires before the configured time. Although session re-authentication is attempted, PLAIN has no mechanism for refreshing tokens.

If maxSecondsWithoutReauthentication is not configured, OAUTHBEARER and PLAIN clients can remain connected to brokers indefinitely, without needing to re-authenticate. Authenticated sessions do not end with access token expiry. However, this can be considered when configuring authorization, for example, by using keycloak authorization or installing a custom authorizer.

Additional resources

9.4.4. OAuth 2.0 Kafka client configuration

A Kafka client is configured with either:

  • The credentials required to obtain a valid access token from an authorization server (client ID and Secret)

  • A valid long-lived access token or refresh token, obtained using tools provided by an authorization server

The only information ever sent to the Kafka broker is an access token. The credentials used to authenticate with the authorization server to obtain the access token are never sent to the broker.

When a client obtains an access token, no further communication with the authorization server is needed.

The simplest mechanism is authentication with a client ID and Secret. Using a long-lived access token, or a long-lived refresh token, adds more complexity because there is an additional dependency on authorization server tools.

Note
 If you are using long-lived access tokens, you may need to configure the client in the authorization server to increase the maximum lifetime of the token.

If the Kafka client is not configured with an access token directly, the client exchanges credentials for an access token during Kafka session initiation by contacting the authorization server. The Kafka client exchanges either:

  • Client ID and Secret

  • Client ID, refresh token, and (optionally) a secret

  • Username and password, with client ID and (optionally) a secret

9.4.5. OAuth 2.0 client authentication flows

OAuth 2.0 authentication flows depend on the underlying Kafka client and Kafka broker configuration. The flows must also be supported by the authorization server used.

The Kafka broker listener configuration determines how clients authenticate using an access token. The client can pass a client ID and secret to request an access token.

If a listener is configured to use PLAIN authentication, the client can authenticate with a client ID and secret or username and access token. These values are passed as the username and password properties of the PLAIN mechanism.

Listener configuration supports the following token validation options:

  • You can use fast local token validation based on JWT signature checking and local token introspection, without contacting an authorization server. The authorization server provides a JWKS endpoint with public certificates that are used to validate signatures on the tokens.

  • You can use a call to a token introspection endpoint provided by an authorization server. Each time a new Kafka broker connection is established, the broker passes the access token received from the client to the authorization server. The Kafka broker checks the response to confirm whether or not the token is valid.

Note
 An authorization server might only allow the use of opaque access tokens, which means that local token validation is not possible.

Kafka client credentials can also be configured for the following types of authentication:

  • Direct local access using a previously generated long-lived access token

  • Contact with the authorization server for a new access token to be issued (using a client ID and a secret, or a refresh token, or a username and a password)

Example client authentication flows using the SASL OAUTHBEARER mechanism

You can use the following communication flows for Kafka authentication using the SASL OAUTHBEARER mechanism.

Client using client ID and secret, with broker delegating validation to authorization server

Client using client ID and secret with broker delegating validation to authorization server

  1. The Kafka client requests an access token from the authorization server using a client ID and secret, and optionally a refresh token. Alternatively, the client may authenticate using a username and a password.

  2. The authorization server generates a new access token.

  3. The Kafka client authenticates with the Kafka broker using the SASL OAUTHBEARER mechanism to pass the access token.

  4. The Kafka broker validates the access token by calling a token introspection endpoint on the authorization server using its own client ID and secret.

  5. A Kafka client session is established if the token is valid.

Client using client ID and secret, with broker performing fast local token validation

Client using client ID and secret with broker performing fast local token validation

  1. The Kafka client authenticates with the authorization server from the token endpoint, using a client ID and secret, and optionally a refresh token. Alternatively, the client may authenticate using a username and a password.

  2. The authorization server generates a new access token.

  3. The Kafka client authenticates with the Kafka broker using the SASL OAUTHBEARER mechanism to pass the access token.

  4. The Kafka broker validates the access token locally using a JWT token signature check, and local token introspection.

Client using long-lived access token, with broker delegating validation to authorization server

Client using long-lived access token with broker delegating validation to authorization server

  1. The Kafka client authenticates with the Kafka broker using the SASL OAUTHBEARER mechanism to pass the long-lived access token.

  2. The Kafka broker validates the access token by calling a token introspection endpoint on the authorization server, using its own client ID and secret.

  3. A Kafka client session is established if the token is valid.

Client using long-lived access token, with broker performing fast local validation

Client using long-lived access token with broker performing fast local validation

  1. The Kafka client authenticates with the Kafka broker using the SASL OAUTHBEARER mechanism to pass the long-lived access token.

  2. The Kafka broker validates the access token locally using a JWT token signature check and local token introspection.

Warning
 Fast local JWT token signature validation is suitable only for short-lived tokens as there is no check with the authorization server if a token has been revoked. Token expiration is written into the token, but revocation can happen at any time, so cannot be accounted for without contacting the authorization server. Any issued token would be considered valid until it expires.
Example client authentication flows using the SASL PLAIN mechanism

You can use the following communication flows for Kafka authentication using the OAuth PLAIN mechanism.

Client using a client ID and secret, with the broker obtaining the access token for the client

Client using a client ID and secret with the broker obtaining the access token for the client

  1. The Kafka client passes a clientId as a username and a secret as a password.

  2. The Kafka broker uses a token endpoint to pass the clientId and secret to the authorization server.

  3. The authorization server returns a fresh access token or an error if the client credentials are not valid.

  4. The Kafka broker validates the token in one of the following ways:

    1. If a token introspection endpoint is specified, the Kafka broker validates the access token by calling the endpoint on the authorization server. A session is established if the token validation is successful.

    2. If local token introspection is used, a request is not made to the authorization server. The Kafka broker validates the access token locally using a JWT token signature check.

Client using a long-lived access token without a client ID and secret

Client using a long-lived access token without a client ID and secret

  1. The Kafka client passes a username and password. The password provides the value of an access token that was obtained manually and configured before running the client.

  2. The password is passed with or without an $accessToken: string prefix depending on whether or not the Kafka broker listener is configured with a token endpoint for authentication.

    1. If the token endpoint is configured, the password should be prefixed by $accessToken: to let the broker know that the password parameter contains an access token rather than a client secret. The Kafka broker interprets the username as the account username.

    2. If the token endpoint is not configured on the Kafka broker listener (enforcing a no-client-credentials mode), the password should provide the access token without the prefix. The Kafka broker interprets the username as the account username. In this mode, the client doesn’t use a client ID and secret, and the password parameter is always interpreted as a raw access token.

  3. The Kafka broker validates the token in one of the following ways:

    1. If a token introspection endpoint is specified, the Kafka broker validates the access token by calling the endpoint on the authorization server. A session is established if token validation is successful.

    2. If local token introspection is used, there is no request made to the authorization server. Kafka broker validates the access token locally using a JWT token signature check.

9.4.6. Configuring OAuth 2.0 authentication

OAuth 2.0 is used for interaction between Kafka clients and Strimzi components.

In order to use OAuth 2.0 for Strimzi, you must:

  1. Configure an OAuth 2.0 authorization server for the Strimzi cluster and Kafka clients

  2. Deploy or update the Kafka cluster with Kafka broker listeners configured to use OAuth 2.0

  3. Update your Java-based Kafka clients to use OAuth 2.0

  4. Update Kafka component clients to use OAuth 2.0

Configuring an OAuth 2.0 authorization server

This procedure describes in general what you need to do to configure an authorization server for integration with Strimzi.

These instructions are not product specific.

The steps are dependent on the chosen authorization server. Consult the product documentation for the authorization server for information on how to set up OAuth 2.0 access.

Note
 If you already have an authorization server deployed, you can skip the deployment step and use your current deployment.
Procedure

  1. Deploy the authorization server to your cluster.

  2. Access the CLI or admin console for the authorization server to configure OAuth 2.0 for Strimzi.

    Now prepare the authorization server to work with Strimzi.

  3. Configure a kafka-broker client.

  4. Configure clients for each Kafka client component of your application.

What to do next

After deploying and configuring the authorization server, configure the Kafka brokers to use OAuth 2.0.

Configuring OAuth 2.0 support for Kafka brokers

This procedure describes how to configure Kafka brokers so that the broker listeners are enabled to use OAuth 2.0 authentication using an authorization server.

We advise use of OAuth 2.0 over an encrypted interface through through a listener with tls: true. Plain listeners are not recommended.

If the authorization server is using certificates signed by the trusted CA and matching the OAuth 2.0 server hostname, TLS connection works using the default settings. Otherwise, you may need to configure the truststore with proper certificates or disable the certificate hostname validation.

When configuring the Kafka broker you have two options for the mechanism used to validate the access token during OAuth 2.0 authentication of the newly connected Kafka client:

Before you start

For more information on the configuration of OAuth 2.0 authentication for Kafka broker listeners, see:

Prerequisites

  • Strimzi and Kafka are running

  • An OAuth 2.0 authorization server is deployed

Procedure

  1. Update the Kafka broker configuration (Kafka.spec.kafka) of your Kafka resource in an editor.

    kubectl edit kafka my-cluster

  2. Configure the Kafka broker listeners configuration.

    The configuration for each type of listener does not have to be the same, as they are independent.

    The examples here show the configuration options as configured for external listeners.

    Example 1: Configuring fast local JWT token validation
    #...
- name: external
  port: 9094
  type: loadbalancer
  tls: true
  authentication:
    type: oauth (1)
    validIssuerUri: <https://<auth-server-address>/auth/realms/external> (2)
    jwksEndpointUri: <https://<auth-server-address>/auth/realms/external/protocol/openid-connect/certs> (3)
    userNameClaim: preferred_username (4)
    maxSecondsWithoutReauthentication: 3600 (5)
    tlsTrustedCertificates: (6)
    - secretName: oauth-server-cert
      certificate: ca.crt
    disableTlsHostnameVerification: true (7)
    jwksExpirySeconds: 360 (8)
    jwksRefreshSeconds: 300 (9)
    jwksMinRefreshPauseSeconds: 1 (10)

    1. Listener type set to oauth.

    2. URI of the token issuer used for authentication.

    3. URI of the JWKS certificate endpoint used for local JWT validation.

    4. The token claim (or key) that contains the actual user name in the token. The user name is the principal used to identify the user. The userNameClaim value will depend on the authentication flow and the authorization server used.

    5. (Optional) Activates the Kafka re-authentication mechanism that enforces session expiry to the same length of time as the access token. If the specified value is less than the time left for the access token to expire, then the client will have to re-authenticate before the actual token expiry. By default, the session does not expire when the access token expires, and the client does not attempt re-authentication.

    6. (Optional) Trusted certificates for TLS connection to the authorization server.

    7. (Optional) Disable TLS hostname verification. Default is false.

    8. The duration the JWKS certificates are considered valid before they expire. Default is 360 seconds. If you specify a longer time, consider the risk of allowing access to revoked certificates.

    9. The period between refreshes of JWKS certificates. The interval must be at least 60 seconds shorter than the expiry interval. Default is 300 seconds.

    10. The minimum pause in seconds between consecutive attempts to refresh JWKS public keys. When an unknown signing key is encountered, the JWKS keys refresh is scheduled outside the regular periodic schedule with at least the specified pause since the last refresh attempt. The refreshing of keys follows the rule of exponential backoff, retrying on unsuccessful refreshes with ever increasing pause, until it reaches jwksRefreshSeconds. The default value is 1.

    Example 2: Configuring token validation using an introspection endpoint
    - name: external
  port: 9094
  type: loadbalancer
  tls: true
  authentication:
    type: oauth
    validIssuerUri: <https://<auth-server-address>/auth/realms/external>
    introspectionEndpointUri: <https://<auth-server-address>/auth/realms/external/protocol/openid-connect/token/introspect> (1)
    clientId: kafka-broker (2)
    clientSecret: (3)
      secretName: my-cluster-oauth
      key: clientSecret
    userNameClaim: preferred_username (4)
    maxSecondsWithoutReauthentication: 3600 (5)

    1. URI of the token introspection endpoint.

    2. Client ID to identify the client.

    3. Client Secret and client ID is used for authentication.

    4. The token claim (or key) that contains the actual user name in the token. The user name is the principal used to identify the user. The userNameClaim value will depend on the authorization server used.

    5. (Optional) Activates the Kafka re-authentication mechanism that enforces session expiry to the same length of time as the access token. If the specified value is less than the time left for the access token to expire, then the client will have to re-authenticate before the actual token expiry. By default, the session does not expire when the access token expires, and the client does not attempt re-authentication.

    Depending on how you apply OAuth 2.0 authentication, and the type of authorization server, there are additional (optional) configuration settings you can use:

      # ...
  authentication:
    type: oauth
    # ...
    checkIssuer: false (1)
    checkAudience: true (2)
    fallbackUserNameClaim: client_id (3)
    fallbackUserNamePrefix: client-account- (4)
    validTokenType: bearer (5)
    userInfoEndpointUri: https://OAUTH-SERVER-ADDRESS/auth/realms/external/protocol/openid-connect/userinfo (6)
    enableOauthBearer: false (7)
    enablePlain: true (8)
    tokenEndpointUri: https://OAUTH-SERVER-ADDRESS/auth/realms/external/protocol/openid-connect/token (9)
    customClaimCheck: "@.custom == 'custom-value'" (10)
    clientAudience: AUDIENCE (11)
    clientScope: SCOPE (12)
    connectTimeoutSeconds: 60 (13)
    readTimeoutSeconds: 60 (14)
    groupsClaim: "$.groups" (15)
    groupsClaimDelimiter: "," (16)

    1. If your authorization server does not provide an iss claim, it is not possible to perform an issuer check. In this situation, set checkIssuer to false and do not specify a validIssuerUri. Default is true.

    2. If your authorization server provides an aud (audience) claim, and you want to enforce an audience check, set checkAudience to true. Audience checks identify the intended recipients of tokens. As a result, the Kafka broker will reject tokens that do not have its clientId in their aud claim. Default is false.

    3. An authorization server may not provide a single attribute to identify both regular users and clients. When a client authenticates in its own name, the server might provide a client ID. When a user authenticates using a username and password, to obtain a refresh token or an access token, the server might provide a username attribute in addition to a client ID. Use this fallback option to specify the username claim (attribute) to use if a primary user ID attribute is not available.

    4. In situations where fallbackUserNameClaim is applicable, it may also be necessary to prevent name collisions between the values of the username claim, and those of the fallback username claim. Consider a situation where a client called producer exists, but also a regular user called producer exists. In order to differentiate between the two, you can use this property to add a prefix to the user ID of the client.

    5. (Only applicable when using introspectionEndpointUri) Depending on the authorization server you are using, the introspection endpoint may or may not return the token type attribute, or it may contain different values. You can specify a valid token type value that the response from the introspection endpoint has to contain.

    6. (Only applicable when using introspectionEndpointUri) The authorization server may be configured or implemented in such a way to not provide any identifiable information in an Introspection Endpoint response. In order to obtain the user ID, you can configure the URI of the userinfo endpoint as a fallback. The userNameClaim, fallbackUserNameClaim, and fallbackUserNamePrefix settings are applied to the response of userinfo endpoint.

    7. Set this to false to disable the OAUTHBEARER mechanism on the listener. At least one of PLAIN or OAUTHBEARER has to be enabled. Default is true.

    8. Set to true to enable PLAIN authentication on the listener, which is supported for clients on all platforms.

    9. Additional configuration for the PLAIN mechanism. If specified, clients can authenticate over PLAIN by passing an access token as the password using an $accessToken: prefix. For production, always use https:// urls.

    10. Additional custom rules can be imposed on the JWT access token during validation by setting this to a JsonPath filter query. If the access token does not contain the necessary data, it is rejected. When using the introspectionEndpointUri, the custom check is applied to the introspection endpoint response JSON.

    11. An audience parameter passed to the token endpoint. An audience is used when obtaining an access token for inter-broker authentication. It is also used in the name of a client for OAuth 2.0 over PLAIN client authentication using a clientId and secret. This only affects the ability to obtain the token, and the content of the token, depending on the authorization server. It does not affect token validation rules by the listener.

    12. A scope parameter passed to the token endpoint. A scope is used when obtaining an access token for inter-broker authentication. It is also used in the name of a client for OAuth 2.0 over PLAIN client authentication using a clientId and secret. This only affects the ability to obtain the token, and the content of the token, depending on the authorization server. It does not affect token validation rules by the listener.

    13. The connect timeout in seconds when connecting to the authorization server. The default value is 60.

    14. The read timeout in seconds when connecting to the authorization server. The default value is 60.

    15. A JsonPath query used to extract groups information from JWT token or introspection endpoint response. Not set by default. This can be used by a custom authorizer to make authorization decisions based on user groups.

    16. A delimiter used to parse groups information when returned as a single delimited string. The default value is ',' (comma).

  3. Save and exit the editor, then wait for rolling updates to complete.

  4. Check the update in the logs or by watching the pod state transitions:

    kubectl logs -f ${POD_NAME} -c ${CONTAINER_NAME}
kubectl get pod -w

    The rolling update configures the brokers to use OAuth 2.0 authentication.

What to do next
Configuring Kafka Java clients to use OAuth 2.0

Configure Kafka producer and consumer APIs to use OAuth 2.0 for interaction with Kafka brokers. Add a callback plugin to your client pom.xml file, then configure your client for OAuth 2.0.

Specify the following in your client configuration:

  • A SASL (Simple Authentication and Security Layer) security protocol:

    • SASL_SSL for authentication over TLS encrypted connections

    • SASL_PLAINTEXT for authentication over unencrypted connections

      Use SASL_SSL for production and SASL_PLAINTEXT for local development only. When using SASL_SSL, additional ssl.truststore configuration is needed. The truststore configuration is required for secure connection (https://) to the OAuth 2.0 authorization server. To verify the OAuth 2.0 authorization server, add the CA certificate for the authorization server to the truststore in your client configuration. You can configure a truststore in PEM or PKCS #12 format.

  • A Kafka SASL mechanism:

    • OAUTHBEARER for credentials exchange using a bearer token

    • PLAIN to pass client credentials (clientId + secret) or an access token

  • A JAAS (Java Authentication and Authorization Service) module that implements the SASL mechanism:

    • org.apache.kafka.common.security.oauthbearer.OAuthBearerLoginModule implements the OAUTHBEARER mechanism

    • org.apache.kafka.common.security.plain.PlainLoginModule implements the PLAIN mechanism

  • SASL authentication properties, which support the following authentication methods:

    • OAuth 2.0 client credentials

    • OAuth 2.0 password grant (deprecated)

    • Access token

    • Refresh token

Add the SASL authentication properties as JAAS configuration (sasl.jaas.config). How you configure the authentication properties depends on the authentication method you are using to access the OAuth 2.0 authorization server. In this procedure, the properties are specified in a properties file, then loaded into the client configuration.

Note
 You can also specify authentication properties as environment variables, or as Java system properties. For Java system properties, you can set them using setProperty and pass them on the command line using the -D option.
Prerequisites

  • Strimzi and Kafka are running

  • An OAuth 2.0 authorization server is deployed and configured for OAuth access to Kafka brokers

  • Kafka brokers are configured for OAuth 2.0

Procedure

  1. Add the client library with OAuth 2.0 support to the pom.xml file for the Kafka client:

    <dependency>
 <groupId>io.strimzi</groupId>
 <artifactId>kafka-oauth-client</artifactId>
 <version>0.11.0</version>
</dependency>

  2. Configure the client properties by specifying the following configuration in a properties file:

    • The security protocol

    • The SASL mechanism

    • The JAAS module and authentication properties according to the method being used

      For example, we can add the following to a client.properties file:

      Client credentials mechanism properties
      security.protocol=SASL_SSL # (1)
sasl.mechanism=OAUTHBEARER # (2)
ssl.truststore.location=/tmp/truststore.p12 (3)
ssl.truststore.password=$STOREPASS
ssl.truststore.type=PKCS12
sasl.jaas.config=org.apache.kafka.common.security.oauthbearer.OAuthBearerLoginModule required \
  oauth.token.endpoint.uri="<token_endpoint_url>" \ # (4)
  oauth.client.id="<client_id>" \ # (5)
  oauth.client.secret="<client_secret>" \ # (6)
  oauth.ssl.truststore.location="/tmp/oauth-truststore.p12" \ (7)
  oauth.ssl.truststore.password="$STOREPASS" \ (8)
  oauth.ssl.truststore.type="PKCS12" \ (9)
  oauth.scope="<scope>" \ # (10)
  oauth.audience="<audience>" ; # (11)

      1. SASL_SSL security protocol for TLS-encrypted connections. Use SASL_PLAINTEXT over unencrypted connections for local development only.

      2. The SASL mechanism specified as OAUTHBEARER or PLAIN.

      3. The truststore configuration for secure access to the Kafka cluster.

      4. URI of the authorization server token endpoint.

      5. Client ID, which is the name used when creating the client in the authorization server.

      6. Client secret created when creating the client in the authorization server.

      7. The location contains the public key certificate (truststore.p12) for the authorization server.

      8. The password for accessing the truststore.

      9. The truststore type.

      10. (Optional) The scope for requesting the token from the token endpoint. An authorization server may require a client to specify the scope.

      11. (Optional) The audience for requesting the token from the token endpoint. An authorization server may require a client to specify the audience.

      Password grants mechanism properties
      security.protocol=SASL_SSL
sasl.mechanism=OAUTHBEARER
ssl.truststore.location=/tmp/truststore.p12
ssl.truststore.password=$STOREPASS
ssl.truststore.type=PKCS12
sasl.jaas.config=org.apache.kafka.common.security.oauthbearer.OAuthBearerLoginModule required \
  oauth.token.endpoint.uri="<token_endpoint_url>" \
  oauth.client.id="<client_id>" \ # (1)
  oauth.client.secret="<client_secret>" \ # (2)
  oauth.password.grant.username="<username>" \ # (3)
  oauth.password.grant.password="<password>" \ # (4)
  oauth.ssl.truststore.location="/tmp/oauth-truststore.p12" \
  oauth.ssl.truststore.password="$STOREPASS" \
  oauth.ssl.truststore.type="PKCS12" \
  oauth.scope="<scope>" \
  oauth.audience="<audience>" ;

      1. Client ID, which is the name used when creating the client in the authorization server.

      2. (Optional) Client secret created when creating the client in the authorization server.

      3. Username for password grant authentication. OAuth password grant configuration (username and password) uses the OAuth 2.0 password grant method. To use password grants, create a user account for a client on your authorization server with limited permissions. The account should act like a service account. Use in environments where user accounts are required for authentication, but consider using a refresh token first.

      4. Password for password grant authentication.

        Note
        		 SASL PLAIN does not support passing a username and password (password grants) using the OAuth 2.0 password grant method.
      Access token properties
      security.protocol=SASL_SSL
sasl.mechanism=OAUTHBEARER
ssl.truststore.location=/tmp/truststore.p12
ssl.truststore.password=$STOREPASS
ssl.truststore.type=PKCS12
sasl.jaas.config=org.apache.kafka.common.security.oauthbearer.OAuthBearerLoginModule required \
  oauth.token.endpoint.uri="<token_endpoint_url>" \
  oauth.access.token="<access_token>" ; # (1)
  oauth.ssl.truststore.location="/tmp/oauth-truststore.p12" \
  oauth.ssl.truststore.password="$STOREPASS" \
  oauth.ssl.truststore.type="PKCS12" \

      1. Long-lived access token for Kafka clients.

      Refresh token properties
      security.protocol=SASL_SSL
sasl.mechanism=OAUTHBEARER
ssl.truststore.location=/tmp/truststore.p12
ssl.truststore.password=$STOREPASS
ssl.truststore.type=PKCS12
sasl.jaas.config=org.apache.kafka.common.security.oauthbearer.OAuthBearerLoginModule required \
  oauth.token.endpoint.uri="<token_endpoint_url>" \
  oauth.client.id="<client_id>" \ # (1)
  oauth.client.secret="<client_secret>" \ # (2)
  oauth.refresh.token="<refresh_token>" ; # (3)
  oauth.ssl.truststore.location="/tmp/oauth-truststore.p12" \
  oauth.ssl.truststore.password="$STOREPASS" \
  oauth.ssl.truststore.type="PKCS12" \

      1. Client ID, which is the name used when creating the client in the authorization server.

      2. (Optional) Client secret created when creating the client in the authorization server.

      3. Long-lived refresh token for Kafka clients.

  3. Input the client properties for OAUTH 2.0 authentication into the Java client code.

    Example showing input of client properties
    Properties props = new Properties();
try (FileReader reader = new FileReader("client.properties", StandardCharsets.UTF_8)) {
  props.load(reader);
}

  4. Verify that the Kafka client can access the Kafka brokers.

Configuring OAuth 2.0 for Kafka components

This procedure describes how to configure Kafka components to use OAuth 2.0 authentication using an authorization server.

You can configure authentication for:

  • Kafka Connect

  • Kafka MirrorMaker

  • Kafka Bridge

In this scenario, the Kafka component and the authorization server are running in the same cluster.

Before you start

For more information on the configuration of OAuth 2.0 authentication for Kafka components, see the KafkaClientAuthenticationOAuth schema reference. The schema reference includes examples of configuration options.

Prerequisites

  • Strimzi and Kafka are running

  • An OAuth 2.0 authorization server is deployed and configured for OAuth access to Kafka brokers

  • Kafka brokers are configured for OAuth 2.0

Procedure

  1. Create a client secret and mount it to the component as an environment variable.

    For example, here we are creating a client Secret for the Kafka Bridge:

    apiVersion: kafka.strimzi.io/v1beta2
kind: Secret
metadata:
 name: my-bridge-oauth
type: Opaque
data:
 clientSecret: MGQ1OTRmMzYtZTllZS00MDY2LWI5OGEtMTM5MzM2NjdlZjQw (1)

    1. The clientSecret key must be in base64 format.

  2. Create or edit the resource for the Kafka component so that OAuth 2.0 authentication is configured for the authentication property.

    For OAuth 2.0 authentication, you can use the following options:

    • Client ID and secret

    • Client ID and refresh token

    • Access token

    • Username and password

    • TLS

    For example, here OAuth 2.0 is assigned to the Kafka Bridge client using a client ID and secret, and TLS:

    apiVersion: kafka.strimzi.io/v1beta2
kind: KafkaBridge
metadata:
  name: my-bridge
spec:
  # ...
  authentication:
    type: oauth (1)
    tokenEndpointUri: https://<auth-server-address>/auth/realms/master/protocol/openid-connect/token (2)
    clientId: kafka-bridge
    clientSecret:
      secretName: my-bridge-oauth
      key: clientSecret
    tlsTrustedCertificates: (3)
    - secretName: oauth-server-cert
      certificate: tls.crt

    1. Authentication type set to oauth.

    2. URI of the token endpoint for authentication.

    3. Trusted certificates for TLS connection to the authorization server.

    Depending on how you apply OAuth 2.0 authentication, and the type of authorization server, there are additional configuration options you can use:

    # ...
spec:
  # ...
  authentication:
    # ...
    disableTlsHostnameVerification: true (1)
    checkAccessTokenType: false (2)
    accessTokenIsJwt: false (3)
    scope: any (4)
    audience: kafka (5)
    connectTimeoutSeconds: 60 (6)
    readTimeoutSeconds: 60 (7)

    1. (Optional) Disable TLS hostname verification. Default is false.

    2. If the authorization server does not return a typ (type) claim inside the JWT token, you can apply checkAccessTokenType: false to skip the token type check. Default is true.

    3. If you are using opaque tokens, you can apply accessTokenIsJwt: false so that access tokens are not treated as JWT tokens.

    4. (Optional) The scope for requesting the token from the token endpoint. An authorization server may require a client to specify the scope. In this case it is any.

    5. (Optional) The audience for requesting the token from the token endpoint. An authorization server may require a client to specify the audience. In this case it is kafka.

    6. (Optional) The connect timeout in seconds when connecting to the authorization server. The default value is 60.

    7. (Optional) The read timeout in seconds when connecting to the authorization server. The default value is 60.

  3. Apply the changes to the deployment of your Kafka resource.

    kubectl apply -f your-file

  4. Check the update in the logs or by watching the pod state transitions:

    kubectl logs -f ${POD_NAME} -c ${CONTAINER_NAME}
kubectl get pod -w

    The rolling updates configure the component for interaction with Kafka brokers using OAuth 2.0 authentication.

9.4.7. Authorization server examples

When choosing an authorization server, consider the features that best support configuration of your chosen authentication flow.

For the purposes of testing OAuth 2.0 with Strimzi, Keycloak and ORY Hydra were implemented as the OAuth 2.0 authorization server.

For more information, see:

9.5. Using OAuth 2.0 token-based authorization

If you are using OAuth 2.0 with Keycloak for token-based authentication, you can also use Keycloak to configure authorization rules to constrain client access to Kafka brokers. Authentication establishes the identity of a user. Authorization decides the level of access for that user.

Strimzi supports the use of OAuth 2.0 token-based authorization through Keycloak Keycloak Authorization Services, which allows you to manage security policies and permissions centrally.

Security policies and permissions defined in Keycloak are used to grant access to resources on Kafka brokers. Users and clients are matched against policies that permit access to perform specific actions on Kafka brokers.

Kafka allows all users full access to brokers by default, and also provides the AclAuthorizer plugin to configure authorization based on Access Control Lists (ACLs).

ZooKeeper stores ACL rules that grant or deny access to resources based on username. However, OAuth 2.0 token-based authorization with Keycloak offers far greater flexibility on how you wish to implement access control to Kafka brokers. In addition, you can configure your Kafka brokers to use OAuth 2.0 authorization and ACLs.

Additional resources

9.5.1. OAuth 2.0 authorization mechanism

OAuth 2.0 authorization in Strimzi uses Keycloak server Authorization Services REST endpoints to extend token-based authentication with Keycloak by applying defined security policies on a particular user, and providing a list of permissions granted on different resources for that user. Policies use roles and groups to match permissions to users. OAuth 2.0 authorization enforces permissions locally based on the received list of grants for the user from Keycloak Authorization Services.

Kafka broker custom authorizer

A Keycloak authorizer (KeycloakRBACAuthorizer) is provided with Strimzi. To be able to use the Keycloak REST endpoints for Authorization Services provided by Keycloak, you configure a custom authorizer on the Kafka broker.

The authorizer fetches a list of granted permissions from the authorization server as needed, and enforces authorization locally on the Kafka Broker, making rapid authorization decisions for each client request.

9.5.2. Configuring OAuth 2.0 authorization support

This procedure describes how to configure Kafka brokers to use OAuth 2.0 authorization using Keycloak Authorization Services.

Before you begin

Consider the access you require or want to limit for certain users. You can use a combination of Keycloak groups, roles, clients, and users to configure access in Keycloak.

Typically, groups are used to match users based on organizational departments or geographical locations. And roles are used to match users based on their function.

With Keycloak, you can store users and groups in LDAP, whereas clients and roles cannot be stored this way. Storage and access to user data may be a factor in how you choose to configure authorization policies.

Note
 Super users always have unconstrained access to a Kafka broker regardless of the authorization implemented on the Kafka broker.
Prerequisites

  • Strimzi must be configured to use OAuth 2.0 with Keycloak for token-based authentication. You use the same Keycloak server endpoint when you set up authorization.

  • OAuth 2.0 authentication must be configured with the maxSecondsWithoutReauthentication option to enable re-authentication.

Procedure

  1. Access the Keycloak Admin Console or use the Keycloak Admin CLI to enable Authorization Services for the Kafka broker client you created when setting up OAuth 2.0 authentication.

  2. Use Authorization Services to define resources, authorization scopes, policies, and permissions for the client.

  3. Bind the permissions to users and clients by assigning them roles and groups.

  4. Configure the Kafka brokers to use Keycloak authorization by updating the Kafka broker configuration (Kafka.spec.kafka) of your Kafka resource in an editor.

    kubectl edit kafka my-cluster

  5. Configure the Kafka broker kafka configuration to use keycloak authorization, and to be able to access the authorization server and Authorization Services.

    For example:

    apiVersion: kafka.strimzi.io/v1beta2
kind: Kafka
metadata:
  name: my-cluster
spec:
  kafka:
    # ...
    authorization:
      type: keycloak (1)
      tokenEndpointUri: <https://<auth-server-address>/auth/realms/external/protocol/openid-connect/token> (2)
      clientId: kafka (3)
      delegateToKafkaAcls: false (4)
      disableTlsHostnameVerification: false (5)
      superUsers: (6)
      - CN=fred
      - sam
      - CN=edward
      tlsTrustedCertificates: (7)
      - secretName: oauth-server-cert
        certificate: ca.crt
      grantsRefreshPeriodSeconds: 60 (8)
      grantsRefreshPoolSize: 5 (9)
      connectTimeoutSeconds: 60 (10)
      readTimeoutSeconds: 60 (11)
    #...

    1. Type keycloak enables Keycloak authorization.

    2. URI of the Keycloak token endpoint. For production, always use https:// urls. When you configure token-based oauth authentication, you specify a jwksEndpointUri as the URI for local JWT validation. The hostname for the tokenEndpointUri URI must be the same.

    3. The client ID of the OAuth 2.0 client definition in Keycloak that has Authorization Services enabled. Typically, kafka is used as the ID.

    4. (Optional) Delegate authorization to Kafka AclAuthorizer if access is denied by Keycloak Authorization Services policies. Default is false.

    5. (Optional) Disable TLS hostname verification. Default is false.

    6. (Optional) Designated super users.

    7. (Optional) Trusted certificates for TLS connection to the authorization server.

    8. (Optional) The time between two consecutive grants refresh runs. That is the maximum time for active sessions to detect any permissions changes for the user on Keycloak. The default value is 60.

    9. (Optional) The number of threads to use to refresh (in parallel) the grants for the active sessions. The default value is 5.

    10. (Optional) The connect timeout in seconds when connecting to the Keycloak token endpoint. The default value is 60.

    11. (Optional) The read timeout in seconds when connecting to the Keycloak token endpoint. The default value is 60.

  6. Save and exit the editor, then wait for rolling updates to complete.

  7. Check the update in the logs or by watching the pod state transitions:

    kubectl logs -f ${POD_NAME} -c kafka
kubectl get pod -w

    The rolling update configures the brokers to use OAuth 2.0 authorization.

  8. Verify the configured permissions by accessing Kafka brokers as clients or users with specific roles, making sure they have the necessary access, or do not have the access they are not supposed to have.

9.5.3. Managing policies and permissions in Keycloak Authorization Services

This section describes the authorization models used by Keycloak Authorization Services and Kafka, and defines the important concepts in each model.

To grant permissions to access Kafka, you can map Keycloak Authorization Services objects to Kafka resources by creating an OAuth client specification in Keycloak. Kafka permissions are granted to user accounts or service accounts using Keycloak Authorization Services rules.

Examples are shown of the different user permissions required for common Kafka operations, such as creating and listing topics.

Kafka and Keycloak authorization models overview

Kafka and Keycloak Authorization Services use different authorization models.

Kafka authorization model

Kafka’s authorization model uses resource types. When a Kafka client performs an action on a broker, the broker uses the configured KeycloakRBACAuthorizer to check the client’s permissions, based on the action and resource type.

Kafka uses five resource types to control access: Topic, Group, Cluster, TransactionalId, and DelegationToken. Each resource type has a set of available permissions.

Topic

  • Create

  • Write

  • Read

  • Delete

  • Describe

  • DescribeConfigs

  • Alter

  • AlterConfigs

Group

  • Read

  • Describe

  • Delete

Cluster

  • Create

  • Describe

  • Alter

  • DescribeConfigs

  • AlterConfigs

  • IdempotentWrite

  • ClusterAction

TransactionalId

  • Describe

  • Write

DelegationToken

  • Describe

Keycloak Authorization Services model

The Keycloak Authorization Services model has four concepts for defining and granting permissions: resources, authorization scopes, policies, and permissions.

Resources

A resource is a set of resource definitions that are used to match resources with permitted actions. A resource might be an individual topic, for example, or all topics with names starting with the same prefix. A resource definition is associated with a set of available authorization scopes, which represent a set of all actions available on the resource. Often, only a subset of these actions is actually permitted.

Authorization scopes

An authorization scope is a set of all the available actions on a specific resource definition. When you define a new resource, you add scopes from the set of all scopes.

Policies

A policy is an authorization rule that uses criteria to match against a list of accounts. Policies can match:

  • Service accounts based on client ID or roles

  • User accounts based on username, groups, or roles.

Permissions

A permission grants a subset of authorization scopes on a specific resource definition to a set of users.

Additional resources
Map Keycloak Authorization Services to the Kafka authorization model

The Kafka authorization model is used as a basis for defining the Keycloak roles and resources that will control access to Kafka.

To grant Kafka permissions to user accounts or service accounts, you first create an OAuth client specification in Keycloak for the Kafka broker. You then specify Keycloak Authorization Services rules on the client. Typically, the client id of the OAuth client that represents the broker is kafka. The example configuration files provided with Strimzi use kafka as the OAuth client id.

Note

If you have multiple Kafka clusters, you can use a single OAuth client (kafka) for all of them. This gives you a single, unified space in which to define and manage authorization rules. However, you can also use different OAuth client ids (for example, my-cluster-kafka or cluster-dev-kafka) and define authorization rules for each cluster within each client configuration.

The kafka client definition must have the Authorization Enabled option enabled in the Keycloak Admin Console.

All permissions exist within the scope of the kafka client. If you have different Kafka clusters configured with different OAuth client IDs, they each need a separate set of permissions even though they’re part of the same Keycloak realm.

When the Kafka client uses OAUTHBEARER authentication, the Keycloak authorizer (KeycloakRBACAuthorizer) uses the access token of the current session to retrieve a list of grants from the Keycloak server. To retrieve the grants, the authorizer evaluates the Keycloak Authorization Services policies and permissions.

Authorization scopes for Kafka permissions

An initial Keycloak configuration usually involves uploading authorization scopes to create a list of all possible actions that can be performed on each Kafka resource type. This step is performed once only, before defining any permissions. You can add authorization scopes manually instead of uploading them.

Authorization scopes must contain all the possible Kafka permissions regardless of the resource type:

  • Create

  • Write

  • Read

  • Delete

  • Describe

  • Alter

  • DescribeConfig

  • AlterConfig

  • ClusterAction

  • IdempotentWrite

Note

If you’re certain you won’t need a permission (for example, IdempotentWrite), you can omit it from the list of authorization scopes. However, that permission won’t be available to target on Kafka resources.
Resource patterns for permissions checks

Resource patterns are used for pattern matching against the targeted resources when performing permission checks. The general pattern format is RESOURCE-TYPE:PATTERN-NAME.

The resource types mirror the Kafka authorization model. The pattern allows for two matching options:

  • Exact matching (when the pattern does not end with *)

  • Prefix matching (when the pattern ends with *)

Example patterns for resources
Topic:my-topic
Topic:orders-*
Group:orders-*
Cluster:*

Additionally, the general pattern format can be prefixed by kafka-cluster:CLUSTER-NAME followed by a comma, where CLUSTER-NAME refers to the metadata.name in the Kafka custom resource.

Example patterns for resources with cluster prefix
kafka-cluster:my-cluster,Topic:*
kafka-cluster:*,Group:b_*

When the kafka-cluster prefix is missing, it is assumed to be kafka-cluster:*.

When defining a resource, you can associate it with a list of possible authorization scopes which are relevant to the resource. Set whatever actions make sense for the targeted resource type.

Though you may add any authorization scope to any resource, only the scopes supported by the resource type are considered for access control.

Policies for applying access permission

Policies are used to target permissions to one or more user accounts or service accounts. Targeting can refer to:

  • Specific user or service accounts

  • Realm roles or client roles

  • User groups

  • JavaScript rules to match a client IP address

A policy is given a unique name and can be reused to target multiple permissions to multiple resources.

Permissions to grant access

Use fine-grained permissions to pull together the policies, resources, and authorization scopes that grant access to users.

The name of each permission should clearly define which permissions it grants to which users. For example, Dev Team B can read from topics starting with x.

Additional resources
Example permissions required for Kafka operations

The following examples demonstrate the user permissions required for performing common operations on Kafka.

Create a topic

To create a topic, the Create permission is required for the specific topic, or for Cluster:kafka-cluster.

bin/kafka-topics.sh --create --topic my-topic \
  --bootstrap-server my-cluster-kafka-bootstrap:9092 --command-config=/tmp/config.properties
List topics

If a user has the Describe permission on a specified topic, the topic is listed.

bin/kafka-topics.sh --list \
  --bootstrap-server my-cluster-kafka-bootstrap:9092 --command-config=/tmp/config.properties
Display topic details

To display a topic’s details, Describe and DescribeConfigs permissions are required on the topic.

bin/kafka-topics.sh --describe --topic my-topic \
  --bootstrap-server my-cluster-kafka-bootstrap:9092 --command-config=/tmp/config.properties
Produce messages to a topic

To produce messages to a topic, Describe and Write permissions are required on the topic.

If the topic hasn’t been created yet, and topic auto-creation is enabled, the permissions to create a topic are required.

bin/kafka-console-producer.sh  --topic my-topic \
  --bootstrap-server my-cluster-kafka-bootstrap:9092 --producer.config=/tmp/config.properties
Consume messages from a topic

To consume messages from a topic, Describe and Read permissions are required on the topic. Consuming from the topic normally relies on storing the consumer offsets in a consumer group, which requires additional Describe and Read permissions on the consumer group.

Two resources are needed for matching. For example:

Topic:my-topic
Group:my-group-*
bin/kafka-console-consumer.sh --topic my-topic --group my-group-1 --from-beginning \
  --bootstrap-server my-cluster-kafka-bootstrap:9092 --consumer.config /tmp/config.properties
Produce messages to a topic using an idempotent producer

As well as the permissions for producing to a topic, an additional IdempotentWrite permission is required on the Cluster:kafka-cluster resource.

Two resources are needed for matching. For example:

Topic:my-topic
Cluster:kafka-cluster
bin/kafka-console-producer.sh  --topic my-topic \
  --bootstrap-server my-cluster-kafka-bootstrap:9092 --producer.config=/tmp/config.properties --producer-property enable.idempotence=true --request-required-acks -1
List consumer groups

When listing consumer groups, only the groups on which the user has the Describe permissions are returned. Alternatively, if the user has the Describe permission on the Cluster:kafka-cluster, all the consumer groups are returned.

bin/kafka-consumer-groups.sh --list \
  --bootstrap-server my-cluster-kafka-bootstrap:9092 --command-config=/tmp/config.properties
Display consumer group details

To display a consumer group’s details, the Describe permission is required on the group and the topics associated with the group.

bin/kafka-consumer-groups.sh --describe --group my-group-1 \
  --bootstrap-server my-cluster-kafka-bootstrap:9092 --command-config=/tmp/config.properties
Change topic configuration

To change a topic’s configuration, the Describe and Alter permissions are required on the topic.

bin/kafka-topics.sh --alter --topic my-topic --partitions 2 \
  --bootstrap-server my-cluster-kafka-bootstrap:9092 --command-config=/tmp/config.properties
Display Kafka broker configuration

In order to use kafka-configs.sh to get a broker’s configuration, the DescribeConfigs permission is required on the Cluster:kafka-cluster.

bin/kafka-configs.sh --entity-type brokers --entity-name 0 --describe --all \
  --bootstrap-server my-cluster-kafka-bootstrap:9092 --command-config=/tmp/config.properties
Change Kafka broker configuration

To change a Kafka broker’s configuration, DescribeConfigs and AlterConfigs permissions are required on Cluster:kafka-cluster.

bin/kafka-configs --entity-type brokers --entity-name 0 --alter --add-config log.cleaner.threads=2 \
  --bootstrap-server my-cluster-kafka-bootstrap:9092 --command-config=/tmp/config.properties
Delete a topic

To delete a topic, the Describe and Delete permissions are required on the topic.

bin/kafka-topics.sh --delete --topic my-topic \
  --bootstrap-server my-cluster-kafka-bootstrap:9092 --command-config=/tmp/config.properties
Select a lead partition

To run leader selection for topic partitions, the Alter permission is required on the Cluster:kafka-cluster.

bin/kafka-leader-election.sh --topic my-topic --partition 0 --election-type PREFERRED  /
  --bootstrap-server my-cluster-kafka-bootstrap:9092 --admin.config /tmp/config.properties
Reassign partitions

To generate a partition reassignment file, Describe permissions are required on the topics involved.

bin/kafka-reassign-partitions.sh --topics-to-move-json-file /tmp/topics-to-move.json --broker-list "0,1" --generate \
  --bootstrap-server my-cluster-kafka-bootstrap:9092 --command-config /tmp/config.properties > /tmp/partition-reassignment.json

To execute the partition reassignment, Describe and Alter permissions are required on Cluster:kafka-cluster. Also, Describe permissions are required on the topics involved.

bin/kafka-reassign-partitions.sh --reassignment-json-file /tmp/partition-reassignment.json --execute \
  --bootstrap-server my-cluster-kafka-bootstrap:9092 --command-config /tmp/config.properties

To verify partition reassignment, Describe, and AlterConfigs permissions are required on Cluster:kafka-cluster, and on each of the topics involved.

bin/kafka-reassign-partitions.sh --reassignment-json-file /tmp/partition-reassignment.json --verify \
  --bootstrap-server my-cluster-kafka-bootstrap:9092 --command-config /tmp/config.properties

9.5.4. Trying Keycloak Authorization Services

This example explains how to use Keycloak Authorization Services with keycloak authorization. Use Keycloak Authorization Services to enforce access restrictions on Kafka clients. Keycloak Authorization Services use authorization scopes, policies and permissions to define and apply access control to resources.

Keycloak Authorization Services REST endpoints provide a list of granted permissions on resources for authenticated users. The list of grants (permissions) is fetched from the Keycloak server as the first action after an authenticated session is established by the Kafka client. The list is refreshed in the background so that changes to the grants are detected. Grants are cached and enforced locally on the Kafka broker for each user session to provide fast authorization decisions.

Strimzi provides example configuration files. These include the following example files for setting up Keycloak:

kafka-ephemeral-oauth-single-keycloak-authz.yaml

An example Kafka custom resource configured for OAuth 2.0 token-based authorization using Keycloak. You can use the custom resource to deploy a Kafka cluster that uses keycloak authorization and token-based oauth authentication.

kafka-authz-realm.json

An example Keycloak realm configured with sample groups, users, roles and clients. You can import the realm into a Keycloak instance to set up fine-grained permissions to access Kafka.

If you want to try the example with Keycloak, use these files to perform the tasks outlined in this section in the order shown.

  1. Accessing the Keycloak Admin Console

  2. Deploying a Kafka cluster with Keycloak authorization

  3. Preparing TLS connectivity for a CLI Kafka client session

  4. Checking authorized access to Kafka using a CLI Kafka client session

Authentication

When you configure token-based oauth authentication, you specify a jwksEndpointUri as the URI for local JWT validation. When you configure keycloak authorization, you specify a tokenEndpointUri as the URI of the Keycloak token endpoint. The hostname for both URIs must be the same.

Targeted permissions with group or role policies

In Keycloak, confidential clients with service accounts enabled can authenticate to the server in their own name using a client ID and a secret. This is convenient for microservices that typically act in their own name, and not as agents of a particular user (like a web site). Service accounts can have roles assigned like regular users. They cannot, however, have groups assigned. As a consequence, if you want to target permissions to microservices using service accounts, you cannot use group policies, and should instead use role policies. Conversely, if you want to limit certain permissions only to regular user accounts where authentication with a username and password is required, you can achieve that as a side effect of using the group policies rather than the role policies. This is what is used in this example for permissions that start with ClusterManager. Performing cluster management is usually done interactively using CLI tools. It makes sense to require the user to log in before using the resulting access token to authenticate to the Kafka broker. In this case, the access token represents the specific user, rather than the client application.

Accessing the Keycloak Admin Console

Set up Keycloak, then connect to its Admin Console and add the preconfigured realm. Use the example kafka-authz-realm.json file to import the realm. You can check the authorization rules defined for the realm in the Admin Console. The rules grant access to the resources on the Kafka cluster configured to use the example Keycloak realm.

Prerequisites

  • A running Kubernetes cluster.

  • The Strimzi examples/security/keycloak-authorization/kafka-authz-realm.json file that contains the preconfigured realm.

Procedure

  1. Install the Keycloak server using the Keycloak Operator as described in Installing the Keycloak Operator in the Keycloak documentation.

  2. Wait until the Keycloak instance is running.

  3. Get the external hostname to be able to access the Admin Console.

    NS=sso
kubectl get ingress keycloak -n $NS

    In this example, we assume the Keycloak server is running in the sso namespace.

  4. Get the password for the admin user.

    kubectl get -n $NS pod keycloak-0 -o yaml | less

    The password is stored as a secret, so get the configuration YAML file for the Keycloak instance to identify the name of the secret (secretKeyRef.name).

  5. Use the name of the secret to obtain the clear text password.

    SECRET_NAME=credential-keycloak
kubectl get -n $NS secret $SECRET_NAME -o yaml | grep PASSWORD | awk '{print $2}' | base64 -D

    In this example, we assume the name of the secret is credential-keycloak.

  6. Log in to the Admin Console with the username admin and the password you obtained.

    Use https://HOSTNAME to access the Kubernetes ingress.

    You can now upload the example realm to Keycloak using the Admin Console.

  7. Click Add Realm to import the example realm.

  8. Add the examples/security/keycloak-authorization/kafka-authz-realm.json file, and then click Create.

    You now have kafka-authz as your current realm in the Admin Console.

    The default view displays the Master realm.

  9. In the Keycloak Admin Console, go to Clients > kafka > Authorization > Settings and check that Decision Strategy is set to Affirmative.

    An affirmative policy means that at least one policy must be satisfied for a client to access the Kafka cluster.

  10. In the Keycloak Admin Console, go to Groups, Users, Roles and Clients to view the realm configuration.

    Groups

    Groups are used to create user groups and set user permissions. Groups are sets of users with a name assigned. They are used to compartmentalize users into geographical, organizational or departmental units. Groups can be linked to an LDAP identity provider. You can make a user a member of a group through a custom LDAP server admin user interface, for example, to grant permissions on Kafka resources.

    Users

    Users are used to create users. For this example, alice and bob are defined. alice is a member of the ClusterManager group and bob is a member of ClusterManager-my-cluster group. Users can be stored in an LDAP identity provider.

    Roles

    Roles mark users or clients as having certain permissions. Roles are a concept analogous to groups. They are usually used to tag users with organizational roles and have the requisite permissions. Roles cannot be stored in an LDAP identity provider. If LDAP is a requirement, you can use groups instead, and add Keycloak roles to the groups so that when users are assigned a group they also get a corresponding role.

    Clients

    Clients can have specific configurations. For this example, kafka, kafka-cli, team-a-client, and team-b-client clients are configured.

    • The kafka client is used by Kafka brokers to perform the necessary OAuth 2.0 communication for access token validation. This client also contains the authorization services resource definitions, policies, and authorization scopes used to perform authorization on the Kafka brokers. The authorization configuration is defined in the kafka client from the Authorization tab, which becomes visible when Authorization Enabled is switched on from the Settings tab.

    • The kafka-cli client is a public client that is used by the Kafka command line tools when authenticating with username and password to obtain an access token or a refresh token.

    • The team-a-client and team-b-client clients are confidential clients representing services with partial access to certain Kafka topics.

  11. In the Keycloak Admin Console, go to Authorization > Permissions to see the granted permissions that use the resources and policies defined for the realm.

    For example, the kafka client has the following permissions:

    Dev Team A can write to topics that start with x_ on any cluster
Dev Team B can read from topics that start with x_ on any cluster
Dev Team B can update consumer group offsets that start with x_ on any cluster
ClusterManager of my-cluster Group has full access to cluster config on my-cluster
ClusterManager of my-cluster Group has full access to consumer groups on my-cluster
ClusterManager of my-cluster Group has full access to topics on my-cluster
    Dev Team A

    The Dev Team A realm role can write to topics that start with x_ on any cluster. This combines a resource called Topic:x_*, Describe and Write scopes, and the Dev Team A policy. The Dev Team A policy matches all users that have a realm role called Dev Team A.

    Dev Team B

    The Dev Team B realm role can read from topics that start with x_ on any cluster. This combines Topic:x_*, Group:x_* resources, Describe and Read scopes, and the Dev Team B policy. The Dev Team B policy matches all users that have a realm role called Dev Team B. Matching users and clients have the ability to read from topics, and update the consumed offsets for topics and consumer groups that have names starting with x_.

Deploying a Kafka cluster with Keycloak authorization

Deploy a Kafka cluster configured to connect to the Keycloak server. Use the example kafka-ephemeral-oauth-single-keycloak-authz.yaml file to deploy the Kafka cluster as a Kafka custom resource. The example deploys a single-node Kafka cluster with keycloak authorization and oauth authentication.

Prerequisites

  • The Keycloak authorization server is deployed to your Kubernetes cluster and loaded with the example realm.

  • The Cluster Operator is deployed to your Kubernetes cluster.

  • The Strimzi examples/security/keycloak-authorization/kafka-ephemeral-oauth-single-keycloak-authz.yaml custom resource.

Procedure

  1. Use the hostname of the Keycloak instance you deployed to prepare a truststore certificate for Kafka brokers to communicate with the Keycloak server.

    SSO_HOST=SSO-HOSTNAME
SSO_HOST_PORT=$SSO_HOST:443
STOREPASS=storepass

echo "Q" | openssl s_client -showcerts -connect $SSO_HOST_PORT 2>/dev/null | awk ' /BEGIN CERTIFICATE/,/END CERTIFICATE/ { print $0 } ' > /tmp/sso.pem

    The certificate is required as Kubernetes ingress is used to make a secure (HTTPS) connection.

    Usually there is not one single certificate, but a certificate chain. You only have to provide the top-most issuer CA, which is listed last in the /tmp/sso.pem file. You can extract it manually or using the following commands:

    Example command to extract the top CA certificate in a certificate chain
    split -p "-----BEGIN CERTIFICATE-----" sso.pem sso-
for f in $(ls sso-*); do mv $f $f.pem; done
cp $(ls sso-* | sort -r | head -n 1) sso-ca.crt
    Note
    		 A trusted CA certificate is normally obtained from a trusted source, and not by using the openssl command.

  2. Deploy the certificate to Kubernetes as a secret.

    kubectl create secret generic oauth-server-cert --from-file=/tmp/sso-ca.crt -n $NS

  3. Set the hostname as an environment variable

    SSO_HOST=SSO-HOSTNAME

  4. Create and deploy the example Kafka cluster.

    cat examples/security/keycloak-authorization/kafka-ephemeral-oauth-single-keycloak-authz.yaml | sed -E 's#\${SSO_HOST}'"#$SSO_HOST#" | kubectl create -n $NS -f -
Preparing TLS connectivity for a CLI Kafka client session

Create a new pod for an interactive CLI session. Set up a truststore with a Keycloak certificate for TLS connectivity. The truststore is to connect to Keycloak and the Kafka broker.

Prerequisites

  • The Keycloak authorization server is deployed to your Kubernetes cluster and loaded with the example realm.

    In the Keycloak Admin Console, check the roles assigned to the clients are displayed in Clients > Service Account Roles.

  • The Kafka cluster configured to connect with Keycloak is deployed to your Kubernetes cluster.

Procedure

  1. Run a new interactive pod container using the Strimzi Kafka image to connect to a running Kafka broker.

    NS=sso
kubectl run -ti --restart=Never --image=quay.io/strimzi/kafka:latest-kafka-3.3.2 kafka-cli -n $NS -- /bin/sh
    Note
    		 If kubectl times out waiting on the image download, subsequent attempts may result in an AlreadyExists error.

  2. Attach to the pod container.

    kubectl attach -ti kafka-cli -n $NS

  3. Use the hostname of the Keycloak instance to prepare a certificate for client connection using TLS.

    SSO_HOST=SSO-HOSTNAME
SSO_HOST_PORT=$SSO_HOST:443
STOREPASS=storepass

echo "Q" | openssl s_client -showcerts -connect $SSO_HOST_PORT 2>/dev/null | awk ' /BEGIN CERTIFICATE/,/END CERTIFICATE/ { print $0 } ' > /tmp/sso.pem

    Usually there is not one single certificate, but a certificate chain. You only have to provide the top-most issuer CA, which is listed last in the /tmp/sso.pem file. You can extract it manually or using the following command:

    Example command to extract the top CA certificate in a certificate chain
    split -p "-----BEGIN CERTIFICATE-----" sso.pem sso-
for f in $(ls sso-*); do mv $f $f.pem; done
cp $(ls sso-* | sort -r | head -n 1) sso-ca.crt
    Note
    		 A trusted CA certificate is normally obtained from a trusted source, and not by using the openssl command.

  4. Create a truststore for TLS connection to the Kafka brokers.

    keytool -keystore /tmp/truststore.p12 -storetype pkcs12 -alias sso -storepass $STOREPASS -import -file /tmp/sso-ca.crt -noprompt

  5. Use the Kafka bootstrap address as the hostname of the Kafka broker and the tls listener port (9093) to prepare a certificate for the Kafka broker.

    KAFKA_HOST_PORT=my-cluster-kafka-bootstrap:9093
STOREPASS=storepass

echo "Q" | openssl s_client -showcerts -connect $KAFKA_HOST_PORT 2>/dev/null | awk ' /BEGIN CERTIFICATE/,/END CERTIFICATE/ { print $0 } ' > /tmp/my-cluster-kafka.pem

    The obtained .pem file is usually not one single certificate, but a certificate chain. You only have to provide the top-most issuer CA, which is listed last in the /tmp/my-cluster-kafka.pem file. You can extract it manually or using the following command:

    Example command to extract the top CA certificate in a certificate chain
    split -p "-----BEGIN CERTIFICATE-----" /tmp/my-cluster-kafka.pem kafka-
for f in $(ls kafka-*); do mv $f $f.pem; done
cp $(ls kafka-* | sort -r | head -n 1) my-cluster-kafka-ca.crt
    Note
    		 A trusted CA certificate is normally obtained from a trusted source, and not by using the openssl command. For this example we assume the client is running in a pod in the same namespace where the Kafka cluster was deployed. If the client is accessing the Kafka cluster from outside the Kubernetes cluster, you would have to first determine the bootstrap address. In that case you can also get the cluster certificate directly from the Kubernetes secret, and there is no need for openssl. For more information, see Setting up client access to a Kafka cluster.

  6. Add the certificate for the Kafka broker to the truststore.

    keytool -keystore /tmp/truststore.p12 -storetype pkcs12 -alias my-cluster-kafka -storepass $STOREPASS -import -file /tmp/my-cluster-kafka-ca.crt -noprompt

    Keep the session open to check authorized access.

Checking authorized access to Kafka using a CLI Kafka client session

Check the authorization rules applied through the Keycloak realm using an interactive CLI session. Apply the checks using Kafka’s example producer and consumer clients to create topics with user and service accounts that have different levels of access.

Use the team-a-client and team-b-client clients to check the authorization rules. Use the alice admin user to perform additional administrative tasks on Kafka.

The Strimzi Kafka image used in this example contains Kafka producer and consumer binaries.

Prerequisites
Setting up client and admin user configuration

  1. Prepare a Kafka configuration file with authentication properties for the team-a-client client.

    SSO_HOST=SSO-HOSTNAME

cat > /tmp/team-a-client.properties << EOF
security.protocol=SASL_SSL
ssl.truststore.location=/tmp/truststore.p12
ssl.truststore.password=$STOREPASS
ssl.truststore.type=PKCS12
sasl.mechanism=OAUTHBEARER
sasl.jaas.config=org.apache.kafka.common.security.oauthbearer.OAuthBearerLoginModule required \
  oauth.client.id="team-a-client" \
  oauth.client.secret="team-a-client-secret" \
  oauth.ssl.truststore.location="/tmp/truststore.p12" \
  oauth.ssl.truststore.password="$STOREPASS" \
  oauth.ssl.truststore.type="PKCS12" \
  oauth.token.endpoint.uri="https://$SSO_HOST/auth/realms/kafka-authz/protocol/openid-connect/token" ;
sasl.login.callback.handler.class=io.strimzi.kafka.oauth.client.JaasClientOauthLoginCallbackHandler
EOF

    The SASL OAUTHBEARER mechanism is used. This mechanism requires a client ID and client secret, which means the client first connects to the Keycloak server to obtain an access token. The client then connects to the Kafka broker and uses the access token to authenticate.

  2. Prepare a Kafka configuration file with authentication properties for the team-b-client client.

    cat > /tmp/team-b-client.properties << EOF
security.protocol=SASL_SSL
ssl.truststore.location=/tmp/truststore.p12
ssl.truststore.password=$STOREPASS
ssl.truststore.type=PKCS12
sasl.mechanism=OAUTHBEARER
sasl.jaas.config=org.apache.kafka.common.security.oauthbearer.OAuthBearerLoginModule required \
  oauth.client.id="team-b-client" \
  oauth.client.secret="team-b-client-secret" \
  oauth.ssl.truststore.location="/tmp/truststore.p12" \
  oauth.ssl.truststore.password="$STOREPASS" \
  oauth.ssl.truststore.type="PKCS12" \
  oauth.token.endpoint.uri="https://$SSO_HOST/auth/realms/kafka-authz/protocol/openid-connect/token" ;
sasl.login.callback.handler.class=io.strimzi.kafka.oauth.client.JaasClientOauthLoginCallbackHandler
EOF

  3. Authenticate admin user alice by using curl and performing a password grant authentication to obtain a refresh token.

    USERNAME=alice
PASSWORD=alice-password

GRANT_RESPONSE=$(curl -X POST "https://$SSO_HOST/auth/realms/kafka-authz/protocol/openid-connect/token" -H 'Content-Type: application/x-www-form-urlencoded' -d "grant_type=password&username=$USERNAME&password=$PASSWORD&client_id=kafka-cli&scope=offline_access" -s -k)

REFRESH_TOKEN=$(echo $GRANT_RESPONSE | awk -F "refresh_token\":\"" '{printf $2}' | awk -F "\"" '{printf $1}')

    The refresh token is an offline token that is long-lived and does not expire.

  4. Prepare a Kafka configuration file with authentication properties for the admin user alice.

    cat > /tmp/alice.properties << EOF
security.protocol=SASL_SSL
ssl.truststore.location=/tmp/truststore.p12
ssl.truststore.password=$STOREPASS
ssl.truststore.type=PKCS12
sasl.mechanism=OAUTHBEARER
sasl.jaas.config=org.apache.kafka.common.security.oauthbearer.OAuthBearerLoginModule required \
  oauth.refresh.token="$REFRESH_TOKEN" \
  oauth.client.id="kafka-cli" \
  oauth.ssl.truststore.location="/tmp/truststore.p12" \
  oauth.ssl.truststore.password="$STOREPASS" \
  oauth.ssl.truststore.type="PKCS12" \
  oauth.token.endpoint.uri="https://$SSO_HOST/auth/realms/kafka-authz/protocol/openid-connect/token" ;
sasl.login.callback.handler.class=io.strimzi.kafka.oauth.client.JaasClientOauthLoginCallbackHandler
EOF

    The kafka-cli public client is used for the oauth.client.id in the sasl.jaas.config. Since it’s a public client it does not require a secret. The client authenticates with the refresh token that was authenticated in the previous step. The refresh token requests an access token behind the scenes, which is then sent to the Kafka broker for authentication.

Producing messages with authorized access

Use the team-a-client configuration to check that you can produce messages to topics that start with a_ or x_.

  1. Write to topic my-topic.

    bin/kafka-console-producer.sh --bootstrap-server my-cluster-kafka-bootstrap:9093 --topic my-topic \
  --producer.config=/tmp/team-a-client.properties
First message

    This request returns a Not authorized to access topics: [my-topic] error.

    team-a-client has a Dev Team A role that gives it permission to perform any supported actions on topics that start with a_, but can only write to topics that start with x_. The topic named my-topic matches neither of those rules.

  2. Write to topic a_messages.

    bin/kafka-console-producer.sh --bootstrap-server my-cluster-kafka-bootstrap:9093 --topic a_messages \
  --producer.config /tmp/team-a-client.properties
First message
Second message

    Messages are produced to Kafka successfully.

  3. Press CTRL+C to exit the CLI application.

  4. Check the Kafka container log for a debug log of Authorization GRANTED for the request.

    kubectl logs my-cluster-kafka-0 -f -n $NS
Consuming messages with authorized access

Use the team-a-client configuration to consume messages from topic a_messages.

  1. Fetch messages from topic a_messages.

    bin/kafka-console-consumer.sh --bootstrap-server my-cluster-kafka-bootstrap:9093 --topic a_messages \
  --from-beginning --consumer.config /tmp/team-a-client.properties

    The request returns an error because the Dev Team A role for team-a-client only has access to consumer groups that have names starting with a_.

  2. Update the team-a-client properties to specify the custom consumer group it is permitted to use.

    bin/kafka-console-consumer.sh --bootstrap-server my-cluster-kafka-bootstrap:9093 --topic a_messages \
  --from-beginning --consumer.config /tmp/team-a-client.properties --group a_consumer_group_1

    The consumer receives all the messages from the a_messages topic.

Administering Kafka with authorized access

The team-a-client is an account without any cluster-level access, but it can be used with some administrative operations.

  1. List topics.

    bin/kafka-topics.sh --bootstrap-server my-cluster-kafka-bootstrap:9093 --command-config /tmp/team-a-client.properties --list

    The a_messages topic is returned.

  2. List consumer groups.

    bin/kafka-consumer-groups.sh --bootstrap-server my-cluster-kafka-bootstrap:9093 --command-config /tmp/team-a-client.properties --list

    The a_consumer_group_1 consumer group is returned.

    Fetch details on the cluster configuration.

    bin/kafka-configs.sh --bootstrap-server my-cluster-kafka-bootstrap:9093 --command-config /tmp/team-a-client.properties \
  --entity-type brokers --describe --entity-default

    The request returns an error because the operation requires cluster level permissions that team-a-client does not have.

Using clients with different permissions

Use the team-b-client configuration to produce messages to topics that start with b_.

  1. Write to topic a_messages.

    bin/kafka-console-producer.sh --bootstrap-server my-cluster-kafka-bootstrap:9093 --topic a_messages \
  --producer.config /tmp/team-b-client.properties
Message 1

    This request returns a Not authorized to access topics: [a_messages] error.

  2. Write to topic b_messages.

    bin/kafka-console-producer.sh --bootstrap-server my-cluster-kafka-bootstrap:9093 --topic b_messages \
  --producer.config /tmp/team-b-client.properties
Message 1
Message 2
Message 3

    Messages are produced to Kafka successfully.

  3. Write to topic x_messages.

    bin/kafka-console-producer.sh --bootstrap-server my-cluster-kafka-bootstrap:9093 --topic x_messages \
  --producer.config /tmp/team-b-client.properties
Message 1

    A Not authorized to access topics: [x_messages] error is returned, The team-b-client can only read from topic x_messages.

  4. Write to topic x_messages using team-a-client.

    bin/kafka-console-producer.sh --bootstrap-server my-cluster-kafka-bootstrap:9093 --topic x_messages \
  --producer.config /tmp/team-a-client.properties
Message 1

    This request returns a Not authorized to access topics: [x_messages] error. The team-a-client can write to the x_messages topic, but it does not have a permission to create a topic if it does not yet exist. Before team-a-client can write to the x_messages topic, an admin power user must create it with the correct configuration, such as the number of partitions and replicas.

Managing Kafka with an authorized admin user

Use admin user alice to manage Kafka. alice has full access to manage everything on any Kafka cluster.

  1. Create the x_messages topic as alice.

    bin/kafka-topics.sh --bootstrap-server my-cluster-kafka-bootstrap:9093 --command-config /tmp/alice.properties \
  --topic x_messages --create --replication-factor 1 --partitions 1

    The topic is created successfully.

  2. List all topics as alice.

    bin/kafka-topics.sh --bootstrap-server my-cluster-kafka-bootstrap:9093 --command-config /tmp/alice.properties --list
bin/kafka-topics.sh --bootstrap-server my-cluster-kafka-bootstrap:9093 --command-config /tmp/team-a-client.properties --list
bin/kafka-topics.sh --bootstrap-server my-cluster-kafka-bootstrap:9093 --command-config /tmp/team-b-client.properties --list

    Admin user alice can list all the topics, whereas team-a-client and team-b-client can only list the topics they have access to.

    The Dev Team A and Dev Team B roles both have Describe permission on topics that start with x_, but they cannot see the other team’s topics because they do not have Describe permissions on them.

  3. Use the team-a-client to produce messages to the x_messages topic:

    bin/kafka-console-producer.sh --bootstrap-server my-cluster-kafka-bootstrap:9093 --topic x_messages \
  --producer.config /tmp/team-a-client.properties
Message 1
Message 2
Message 3

    As alice created the x_messages topic, messages are produced to Kafka successfully.

  4. Use the team-b-client to produce messages to the x_messages topic.

    bin/kafka-console-producer.sh --bootstrap-server my-cluster-kafka-bootstrap:9093 --topic x_messages \
  --producer.config /tmp/team-b-client.properties
Message 4
Message 5

    This request returns a Not authorized to access topics: [x_messages] error.

  5. Use the team-b-client to consume messages from the x_messages topic:

    bin/kafka-console-consumer.sh --bootstrap-server my-cluster-kafka-bootstrap:9093 --topic x_messages \
  --from-beginning --consumer.config /tmp/team-b-client.properties --group x_consumer_group_b

    The consumer receives all the messages from the x_messages topic.

  6. Use the team-a-client to consume messages from the x_messages topic.

    bin/kafka-console-consumer.sh --bootstrap-server my-cluster-kafka-bootstrap:9093 --topic x_messages \
  --from-beginning --consumer.config /tmp/team-a-client.properties --group x_consumer_group_a

    This request returns a Not authorized to access topics: [x_messages] error.

  7. Use the team-a-client to consume messages from a consumer group that begins with a_.

    bin/kafka-console-consumer.sh --bootstrap-server my-cluster-kafka-bootstrap:9093 --topic x_messages \
  --from-beginning --consumer.config /tmp/team-a-client.properties --group a_consumer_group_a

    This request returns a Not authorized to access topics: [x_messages] error.

    Dev Team A has no Read access on topics that start with a x_.

  8. Use alice to produce messages to the x_messages topic.

    bin/kafka-console-consumer.sh --bootstrap-server my-cluster-kafka-bootstrap:9093 --topic x_messages \
  --from-beginning --consumer.config /tmp/alice.properties

    Messages are produced to Kafka successfully.

    alice can read from or write to any topic.

  9. Use alice to read the cluster configuration.

    bin/kafka-configs.sh --bootstrap-server my-cluster-kafka-bootstrap:9093 --command-config /tmp/alice.properties \
  --entity-type brokers --describe --entity-default

    The cluster configuration for this example is empty.

Additional resources

10. Introducing metrics

You can use Prometheus and Grafana to monitor your Strimzi deployment.

You can monitor your Strimzi deployment by viewing key metrics on dashboards and setting up alerts that trigger under certain conditions. Metrics are available for each of the components of Strimzi.

You can also collect metrics specific to oauth authentication and opa or keycloak authorization. You do this by setting the enableMetrics property to true in the listener configuration of the Kafka resource. For example, set enableMetrics to true in spec.kafka.listeners.authentication and spec.kafka.authorization. Similarly, you can enable metrics for oauth authentication in the KafkaBridge, KafkaConnect, KafkaMirrorMaker, and KafkaMirrorMaker2 custom resources.

To provide metrics information, Strimzi uses Prometheus rules and Grafana dashboards.

When configured with a set of rules for each component of Strimzi, Prometheus consumes key metrics from the pods that are running in your cluster. Grafana then visualizes those metrics on dashboards. Strimzi includes example Grafana dashboards that you can customize to suit your deployment.

Depending on your requirements, you can:

With Prometheus and Grafana set up, you can use the example Grafana dashboards provided by Strimzi for monitoring.

Additionally, you can configure your deployment to track messages end-to-end by setting up distributed tracing.

Note
 Strimzi provides example installation files for Prometheus and Grafana. You can use these files as a starting point when trying out monitoring of Strimzi. For further support, try engaging with the Prometheus and Grafana developer communities.
Supporting documentation for metrics and monitoring tools

For more information on the metrics and monitoring tools, refer to the supporting documentation:

10.1. Monitoring consumer lag with Kafka Exporter

Kafka Exporter is an open source project to enhance monitoring of Apache Kafka brokers and clients. You can configure the Kafka resource to deploy Kafka Exporter with your Kafka cluster. Kafka Exporter extracts additional metrics data from Kafka brokers related to offsets, consumer groups, consumer lag, and topics. The metrics data is used, for example, to help identify slow consumers. Lag data is exposed as Prometheus metrics, which can then be presented in Grafana for analysis.

Kafka Exporter reads from the __consumer_offsets topic, which stores information on committed offsets for consumer groups. For Kafka Exporter to be able to work properly, consumer groups needs to be in use.

A Grafana dashboard for Kafka Exporter is one of a number of example Grafana dashboards provided by Strimzi.

Important
 Kafka Exporter provides only additional metrics related to consumer lag and consumer offsets. For regular Kafka metrics, you have to configure the Prometheus metrics in Kafka brokers.

Consumer lag indicates the difference in the rate of production and consumption of messages. Specifically, consumer lag for a given consumer group indicates the delay between the last message in the partition and the message being currently picked up by that consumer.

The lag reflects the position of the consumer offset in relation to the end of the partition log.

Consumer lag between the producer and consumer offset

Consumer lag

This difference is sometimes referred to as the delta between the producer offset and consumer offset: the read and write positions in the Kafka broker topic partitions.

Suppose a topic streams 100 messages a second. A lag of 1000 messages between the producer offset (the topic partition head) and the last offset the consumer has read means a 10-second delay.

The importance of monitoring consumer lag

For applications that rely on the processing of (near) real-time data, it is critical to monitor consumer lag to check that it does not become too big. The greater the lag becomes, the further the process moves from the real-time processing objective.

Consumer lag, for example, might be a result of consuming too much old data that has not been purged, or through unplanned shutdowns.

Reducing consumer lag

Use the Grafana charts to analyze lag and to check if actions to reduce lag are having an impact on an affected consumer group. If, for example, Kafka brokers are adjusted to reduce lag, the dashboard will show the Lag by consumer group chart going down and the Messages consumed per minute chart going up.

Typical actions to reduce lag include:

  • Scaling-up consumer groups by adding new consumers

  • Increasing the retention time for a message to remain in a topic

  • Adding more disk capacity to increase the message buffer

Actions to reduce consumer lag depend on the underlying infrastructure and the use cases Strimzi is supporting. For instance, a lagging consumer is less likely to benefit from the broker being able to service a fetch request from its disk cache. And in certain cases, it might be acceptable to automatically drop messages until a consumer has caught up.

10.2. Monitoring Cruise Control operations

Cruise Control monitors Kafka brokers in order to track the utilization of brokers, topics, and partitions. Cruise Control also provides a set of metrics for monitoring its own performance.

The Cruise Control metrics reporter collects raw metrics data from Kafka brokers. The data is produced to topics that are automatically created by Cruise Control. The metrics are used to generate optimization proposals for Kafka clusters.

Cruise Control metrics are available for real-time monitoring of Cruise Control operations. For example, you can use Cruise Control metrics to monitor the status of rebalancing operations that are running or provide alerts on any anomalies that are detected in an operation’s performance.

You expose Cruise Control metrics by enabling the Prometheus JMX Exporter in the Cruise Control configuration.

Note
 For a full list of available Cruise Control metrics, which are known as sensors, see the Cruise Control documentation.

10.2.1. Exposing Cruise Control metrics

If you want to expose metrics on Cruise Control operations, configure the Kafka resource to deploy Cruise Control and enable Prometheus metrics in the deployment. You can use your own configuration or use the example kafka-cruise-control-metrics.yaml file provided by Strimzi.

You add the configuration to the metricsConfig of the CruiseControl property in the Kafka resource. The configuration enables the Prometheus JMX Exporter to expose Cruise Control metrics through an HTTP endpoint. The HTTP endpoint is scraped by the Prometheus server.

Example metrics configuration for Cruise Control
  apiVersion: kafka.strimzi.io/v1beta2
  kind: Kafka
  metadata:
    name: my-cluster
  Spec:
    # ...
    cruiseControl:
      # ...
      metricsConfig:
        type: jmxPrometheusExporter
        valueFrom:
          configMapKeyRef:
            name: cruise-control-metrics
            key: metrics-config.yml
  ---
  kind: ConfigMap
  apiVersion: v1
  metadata:
    name: cruise-control-metrics
    labels:
      app: strimzi
  data:
    metrics-config.yml: |
    # metrics configuration...

10.2.2. Viewing Cruise Control metrics

After you expose the Cruise Control metrics, you can use Prometheus or another suitable monitoring system to view information on the metrics data. Strimzi provides an example Grafana dashboard to display visualizations of Cruise Control metrics. The dashboard is a JSON file called strimzi-cruise-control.json. The exposed metrics provide the monitoring data when you enable the Grafana dashboard.

Monitoring balancedness scores

Cruise Control metrics include a balancedness score. Balancedness is the measure of how evenly a workload is distributed in a Kafka cluster.

The Cruise Control metric for balancedness score (balancedness-score) might differ from the balancedness score in the KafkaRebalance resource. Cruise Control calculates each score using anomaly.detection.goals which might not be the same as the default.goals used in the KafkaRebalance resource. The anomaly.detection.goals are specified in the spec.cruiseControl.config of the Kafka custom resource.

Note

Refreshing the KafkaRebalance resource fetches an optimization proposal. The latest cached optimization proposal is fetched if one of the following conditions applies:

  • KafkaRebalance goals match the goals configured in the default.goals section of the Kafka resource

  • KafkaRebalance goals are not specified

Otherwise, Cruise Control generates a new optimization proposal based on KafkaRebalance goals. If new proposals are generated with each refresh, this can impact performance monitoring.
Alerts on anomaly detection

Cruise control’s anomaly detector provides metrics data for conditions that block the generation of optimization goals, such as broker failures. If you want more visibility, you can use the metrics provided by the anomaly detector to set up alerts and send out notifications. You can set up Cruise Control’s anomaly notifier to route alerts based on these metrics through a specified notification channel. Alternatively, you can set up Prometheus to scrape the metrics data provided by the anomaly detector and generate alerts. Prometheus Alertmanager can then route the alerts generated by Prometheus.

The Cruise Control documentation provides information on AnomalyDetector metrics and the anomaly notifier.

10.3. Example metrics files

You can find example Grafana dashboards and other metrics configuration files in the example configuration files provided by Strimzi.

Example metrics files provided with Strimzi
metrics
├── grafana-dashboards (1)
│   ├── strimzi-cruise-control.json
│   ├── strimzi-kafka-bridge.json
│   ├── strimzi-kafka-connect.json
│   ├── strimzi-kafka-exporter.json
│   ├── strimzi-kafka-mirror-maker-2.json
│   ├── strimzi-kafka.json
│   ├── strimzi-operators.json
│   └── strimzi-zookeeper.json
├── grafana-install
│   └── grafana.yaml (2)
├── prometheus-additional-properties
│   └── prometheus-additional.yaml (3)
├── prometheus-alertmanager-config
│   └── alert-manager-config.yaml (4)
├── prometheus-install
│    ├── alert-manager.yaml (5)
│    ├── prometheus-rules.yaml (6)
│    ├── prometheus.yaml (7)
│    ├── strimzi-pod-monitor.yaml (8)
├── kafka-bridge-metrics.yaml (9)
├── kafka-connect-metrics.yaml (10)
├── kafka-cruise-control-metrics.yaml (11)
├── kafka-metrics.yaml (12)
└── kafka-mirror-maker-2-metrics.yaml (13)

  1. Example Grafana dashboards for the different Strimzi components.

  2. Installation file for the Grafana image.

  3. Additional configuration to scrape metrics for CPU, memory and disk volume usage, which comes directly from the Kubernetes cAdvisor agent and kubelet on the nodes.

  4. Hook definitions for sending notifications through Alertmanager.

  5. Resources for deploying and configuring Alertmanager.

  6. Alerting rules examples for use with Prometheus Alertmanager (deployed with Prometheus).

  7. Installation resource file for the Prometheus image.

  8. PodMonitor definitions translated by the Prometheus Operator into jobs for the Prometheus server to be able to scrape metrics data directly from pods.

  9. Kafka Bridge resource with metrics enabled.

  10. Metrics configuration that defines Prometheus JMX Exporter relabeling rules for Kafka Connect.

  11. Metrics configuration that defines Prometheus JMX Exporter relabeling rules for Cruise Control.

  12. Metrics configuration that defines Prometheus JMX Exporter relabeling rules for Kafka and ZooKeeper.

  13. Metrics configuration that defines Prometheus JMX Exporter relabeling rules for Kafka Mirror Maker 2.0.

10.3.1. Example Prometheus metrics configuration

Strimzi uses the Prometheus JMX Exporter to expose metrics through an HTTP endpoint, which can be scraped by the Prometheus server.

Grafana dashboards are dependent on Prometheus JMX Exporter relabeling rules, which are defined for Strimzi components in the custom resource configuration.

A label is a name-value pair. Relabeling is the process of writing a label dynamically. For example, the value of a label may be derived from the name of a Kafka server and client ID.

Strimzi provides example custom resource configuration YAML files with relabeling rules. When deploying Prometheus metrics configuration, you can can deploy the example custom resource or copy the metrics configuration to your own custom resource definition.

Table 4. Example custom resources with metrics configuration
Component Custom resource Example YAML file

Kafka and ZooKeeper

Kafka

kafka-metrics.yaml

Kafka Connect

KafkaConnect

kafka-connect-metrics.yaml

Kafka MirrorMaker 2.0

KafkaMirrorMaker2

kafka-mirror-maker-2-metrics.yaml

Kafka Bridge

KafkaBridge

kafka-bridge-metrics.yaml

Cruise Control

Kafka

kafka-cruise-control-metrics.yaml

10.3.2. Example Prometheus rules for alert notifications

Example Prometheus rules for alert notifications are provided with the example metrics configuration files provided by Strimzi. The rules are specified in the example prometheus-rules.yaml file for use in a Prometheus deployment.

Alerting rules provide notifications about specific conditions observed in metrics. Rules are declared on the Prometheus server, but Prometheus Alertmanager is responsible for alert notifications.

Prometheus alerting rules describe conditions using PromQL expressions that are continuously evaluated.

When an alert expression becomes true, the condition is met and the Prometheus server sends alert data to the Alertmanager. Alertmanager then sends out a notification using the communication method configured for its deployment.

General points about the alerting rule definitions:

  • A for property is used with the rules to determine the period of time a condition must persist before an alert is triggered.

  • A tick is a basic ZooKeeper time unit, which is measured in milliseconds and configured using the tickTime parameter of Kafka.spec.zookeeper.config. For example, if ZooKeeper tickTime=3000, 3 ticks (3 x 3000) equals 9000 milliseconds.

  • The availability of the ZookeeperRunningOutOfSpace metric and alert is dependent on the Kubernetes configuration and storage implementation used. Storage implementations for certain platforms may not be able to supply the information on available space required for the metric to provide an alert.

Alertmanager can be configured to use email, chat messages or other notification methods. Adapt the default configuration of the example rules according to your specific needs.

Example altering rules

The prometheus-rules.yaml file contains example rules for the following components:

  • Kafka

  • ZooKeeper

  • Entity Operator

  • Kafka Connect

  • Kafka Bridge

  • MirrorMaker

  • Kafka Exporter

A description of each of the example rules is provided in the file.

10.3.3. Example Grafana dashboards

If you deploy Prometheus to provide metrics, you can use the example Grafana dashboards provided with Strimzi to monitor Strimzi components.

Example dashboards are provided in the examples/metrics/grafana-dashboards directory as JSON files.

All dashboards provide JVM metrics, as well as metrics specific to the component. For example, the Grafana dashboard for Strimzi operators provides information on the number of reconciliations or custom resources they are processing.

The example dashboards don’t show all the metrics supported by Kafka. The dashboards are populated with a representative set of metrics for monitoring.

Table 5. Example Grafana dashboard files
Component Example JSON file

Strimzi operators

strimzi-operators.json

Kafka

strimzi-kafka.json

ZooKeeper

strimzi-zookeeper.json

Kafka Connect

strimzi-kafka-connect.json

Kafka MirrorMaker 2.0

strimzi-kafka-mirror-maker-2.json

Kafka Bridge

strimzi-kafka-bridge.json

Cruise Control

strimzi-cruise-control.json

Kafka Exporter

strimzi-kafka-exporter.json
Note
 When metrics are not available to the Kafka Exporter, because there is no traffic in the cluster yet, the Kafka Exporter Grafana dashboard will show N/A for numeric fields and No data to show for graphs.

10.4. Using Prometheus with Strimzi

You can use Prometheus to provide monitoring data for the example Grafana dashboards provided with Strimzi.

To expose metrics in Prometheus format, you add configuration to a custom resource. You will also need to make sure that the metrics are scraped by your monitoring stack. Prometheus and Prometheus Alertmanager are used in the examples provided by Strimzi, but you can use also other compatible tools.

  1. Deploying Prometheus metrics configuration

  2. Setting up Prometheus

  3. Deploying Prometheus Alertmanager

10.4.1. Deploying Prometheus metrics configuration

Deploy Prometheus metrics configuration to use Prometheus with Strimzi. Use the metricsConfig property to enable and configure Prometheus metrics.

You can use your own configuration or the example custom resource configuration files provided with Strimzi.

  • kafka-metrics.yaml

  • kafka-connect-metrics.yaml

  • kafka-mirror-maker-2-metrics.yaml

  • kafka-bridge-metrics.yaml

  • kafka-cruise-control-metrics.yaml

The example configuration files have relabeling rules and the configuration required to enable Prometheus metrics. Prometheus scrapes metrics from target HTTP endpoints. The example files are a good way to try Prometheus with Strimzi.

To apply the relabeling rules and metrics configuration, do one of the following:

  • Copy the example configuration to your own custom resources

  • Deploy the custom resource with the metrics configuration

If you want to include Kafka Exporter metrics, add kafkaExporter configuration to your Kafka resource.

Important
 Kafka Exporter provides only additional metrics related to consumer lag and consumer offsets. For regular Kafka metrics, you have to configure the Prometheus metrics in Kafka brokers.

This procedure shows how to deploy Prometheus metrics configuration in the Kafka resource. The process is the same when using the example files for other resources.

Procedure

  1. Deploy the example custom resource with the Prometheus configuration.

    For example, for each Kafka resource you apply the kafka-metrics.yaml file.

    Deploying the example configuration
    kubectl apply -f kafka-metrics.yaml

    Alternatively, you can copy the example configuration in kafka-metrics.yaml to your own Kafka resource.

    Copying the example configuration
    kubectl edit kafka <kafka-configuration-file>

    Copy the metricsConfig property and the ConfigMap it references to your Kafka resource.

    Example metrics configuration for Kafka
    apiVersion: kafka.strimzi.io/v1beta2
kind: Kafka
metadata:
  name: my-cluster
spec:
  kafka:
    # ...
    metricsConfig: (1)
      type: jmxPrometheusExporter
      valueFrom:
        configMapKeyRef:
          name: kafka-metrics
          key: kafka-metrics-config.yml
---
kind: ConfigMap (2)
apiVersion: v1
metadata:
  name: kafka-metrics
  labels:
    app: strimzi
data:
  kafka-metrics-config.yml: |
  # metrics configuration...

    1. Copy the metricsConfig property that references the ConfigMap that contains metrics configuration.

    2. Copy the whole ConfigMap that specifies the metrics configuration.

    Note

    For Kafka Bridge, you specify the enableMetrics property and set it to true.

    apiVersion: kafka.strimzi.io/v1beta2
kind: KafkaBridge
metadata:
  name: my-bridge
spec:
  # ...
  bootstrapServers: my-cluster-kafka:9092
  http:
    # ...
  enableMetrics: true
  # ...

  2. To deploy Kafka Exporter, add kafkaExporter configuration.

    kafkaExporter configuration is only specified in the Kafka resource.

    Example configuration for deploying Kafka Exporter
    apiVersion: kafka.strimzi.io/v1beta2
kind: Kafka
metadata:
  name: my-cluster
spec:
  # ...
  kafkaExporter:
    image: my-registry.io/my-org/my-exporter-cluster:latest (1)
    groupRegex: ".*" (2)
    topicRegex: ".*" (3)
    resources: (4)
      requests:
        cpu: 200m
        memory: 64Mi
      limits:
        cpu: 500m
        memory: 128Mi
    logging: debug (5)
    enableSaramaLogging: true (6)
    template: (7)
      pod:
        metadata:
          labels:
            label1: value1
        imagePullSecrets:
          - name: my-docker-credentials
        securityContext:
          runAsUser: 1000001
          fsGroup: 0
        terminationGracePeriodSeconds: 120
    readinessProbe: (8)
      initialDelaySeconds: 15
      timeoutSeconds: 5
    livenessProbe: (9)
      initialDelaySeconds: 15
      timeoutSeconds: 5
# ...

    1. ADVANCED OPTION: Container image configuration, which is recommended only in special situations.

    2. A regular expression to specify the consumer groups to include in the metrics.

    3. A regular expression to specify the topics to include in the metrics.

    4. CPU and memory resources to reserve.

    5. Logging configuration, to log messages with a given severity (debug, info, warn, error, fatal) or above.

    6. Boolean to enable Sarama logging, a Go client library used by Kafka Exporter.

    7. Customization of deployment templates and pods.

    8. Healthcheck readiness probes.

    9. Healthcheck liveness probes.

Note
 For Kafka Exporter to be able to work properly, consumer groups need to be in use.
Additional resources

10.4.2. Setting up Prometheus

Prometheus provides an open source set of components for systems monitoring and alert notification.

We describe here how you can use the CoreOS Prometheus Operator to run and manage a Prometheus server that is suitable for use in production environments, but with the correct configuration you can run any Prometheus server.

Note

The Prometheus server configuration uses service discovery to discover the pods in the cluster from which it gets metrics. For this feature to work correctly, the service account used for running the Prometheus service pod must have access to the API server so it can retrieve the pod list.

For more information, see Discovering services.
Prometheus configuration

Strimzi provides example configuration files for the Prometheus server.

A Prometheus YAML file is provided for deployment:

  • prometheus.yaml

Additional Prometheus-related configuration is also provided in the following files:

  • prometheus-additional.yaml

  • prometheus-rules.yaml

  • strimzi-pod-monitor.yaml

For Prometheus to obtain monitoring data:

Then use the configuration files to:

Alerting rules

The prometheus-rules.yaml file provides example alerting rule examples for use with Alertmanager.

Prometheus resources

When you apply the Prometheus configuration, the following resources are created in your Kubernetes cluster and managed by the Prometheus Operator:

  • A ClusterRole that grants permissions to Prometheus to read the health endpoints exposed by the Kafka and ZooKeeper pods, cAdvisor and the kubelet for container metrics.

  • A ServiceAccount for the Prometheus pods to run under.

  • A ClusterRoleBinding which binds the ClusterRole to the ServiceAccount.

  • A Deployment to manage the Prometheus Operator pod.

  • A PodMonitor to manage the configuration of the Prometheus pod.

  • A Prometheus to manage the configuration of the Prometheus pod.

  • A PrometheusRule to manage alerting rules for the Prometheus pod.

  • A Secret to manage additional Prometheus settings.

  • A Service to allow applications running in the cluster to connect to Prometheus (for example, Grafana using Prometheus as datasource).

Deploying the CoreOS Prometheus Operator

To deploy the Prometheus Operator to your Kafka cluster, apply the YAML bundle resources file from the Prometheus CoreOS repository.

Procedure

  1. Download the bundle.yaml resources file from the repository.

    On Linux, use:

    curl -s https://raw.githubusercontent.com/coreos/prometheus-operator/master/bundle.yaml | sed -e '/[[:space:]]*namespace: [a-zA-Z0-9-]*$/s/namespace:[[:space:]]*[a-zA-Z0-9-]*$/namespace: my-namespace/' > prometheus-operator-deployment.yaml

    On MacOS, use:

    curl -s https://raw.githubusercontent.com/coreos/prometheus-operator/master/bundle.yaml | sed -e '' '/[[:space:]]*namespace: [a-zA-Z0-9-]*$/s/namespace:[[:space:]]*[a-zA-Z0-9-]*$/namespace: my-namespace/' > prometheus-operator-deployment.yaml

    • Replace the example namespace with your own.

    • Use the latest master release as shown, or choose a release that is compatible with your version of Kubernetes (see the Kubernetes compatibility matrix). The master release of the Prometheus Operator works with Kubernetes 1.18+.

      Note
      		 If using OpenShift, specify a release of the OpenShift fork of the Prometheus Operator repository.

  2. (Optional) If it is not required, you can manually remove the spec.template.spec.securityContext property from the prometheus-operator-deployment.yaml file.

  3. Deploy the Prometheus Operator:

    kubectl create -f prometheus-operator-deployment.yaml
Deploying Prometheus

Use Prometheus to obtain monitoring data in your Kafka cluster.

You can use your own Prometheus deployment or deploy Prometheus using the example metrics configuration files provided by Strimzi. The example files include a configuration file for a Prometheus deployment and files for Prometheus-related resources:

  • examples/metrics/prometheus-install/prometheus.yaml

  • examples/metrics/prometheus-install/prometheus-rules.yaml

  • examples/metrics/prometheus-install/strimzi-pod-monitor.yaml

  • examples/metrics/prometheus-additional-properties/prometheus-additional.yaml

The deployment process creates a ClusterRoleBinding and discovers an Alertmanager instance in the namespace specified for the deployment.

Note
 By default, the Prometheus Operator only supports jobs that include an endpoints role for service discovery. Targets are discovered and scraped for each endpoint port address. For endpoint discovery, the port address may be derived from service (role: service) or pod (role: pod) discovery.
Prerequisites
Procedure

  1. Modify the Prometheus installation file (prometheus.yaml) according to the namespace Prometheus is going to be installed into:

    On Linux, use:

    sed -i 's/namespace: .*/namespace: my-namespace/' prometheus.yaml

    On MacOS, use:

    sed -i '' 's/namespace: .*/namespace: my-namespace/' prometheus.yaml

  2. Edit the PodMonitor resource in strimzi-pod-monitor.yaml to define Prometheus jobs that will scrape the metrics data from pods.

    Update the namespaceSelector.matchNames property with the namespace where the pods to scrape the metrics from are running.

    PodMonitor is used to scrape data directly from pods for Apache Kafka, ZooKeeper, Operators, the Kafka Bridge and Cruise Control.

  3. Edit the prometheus.yaml installation file to include additional configuration for scraping metrics directly from nodes.

    The Grafana dashboards provided show metrics for CPU, memory and disk volume usage, which come directly from the Kubernetes cAdvisor agent and kubelet on the nodes.

    The Prometheus Operator does not have a monitoring resource like PodMonitor for scraping the nodes, so the prometheus-additional.yaml file contains the additional configuration needed.

    1. Create a Secret resource from the configuration file (prometheus-additional.yaml in the examples/metrics/prometheus-additional-properties directory):

      kubectl apply -f prometheus-additional.yaml

    2. Edit the additionalScrapeConfigs property in the prometheus.yaml file to include the name of the Secret and the prometheus-additional.yaml file.

  4. Deploy the Prometheus resources:

    kubectl apply -f strimzi-pod-monitor.yaml
kubectl apply -f prometheus-rules.yaml
kubectl apply -f prometheus.yaml

10.4.3. Deploying Alertmanager

Use Alertmanager to route alerts to a notification service. Prometheus Alertmanager is a component for handling alerts and routing them to a notification service. Alertmanager supports an essential aspect of monitoring, which is to be notified of conditions that indicate potential issues based on alerting rules.

You can use the example metrics configuration files provided by Strimzi to deploy Alertmanager to send notifications to a Slack channel. A configuration file defines the resources for deploying Alertmanager:

  • examples/metrics/prometheus-install/alert-manager.yaml

An additional configuration file provides the hook definitions for sending notifications from your Kafka cluster:

  • examples/metrics/prometheus-alertmanager-config/alert-manager-config.yaml

The following resources are defined on deployment:

  • An Alertmanager to manage the Alertmanager pod.

  • A Secret to manage the configuration of the Alertmanager.

  • A Service to provide an easy to reference hostname for other services to connect to Alertmanager (such as Prometheus).

Prerequisites
Procedure

  1. Create a Secret resource from the Alertmanager configuration file (alert-manager-config.yaml in the examples/metrics/prometheus-alertmanager-config directory):

    kubectl apply -f alert-manager-config.yaml

  2. Update the alert-manager-config.yaml file to replace the:

    • slack_api_url property with the actual value of the Slack API URL related to the application for the Slack workspace

    • channel property with the actual Slack channel on which to send notifications

  3. Deploy Alertmanager:

    kubectl apply -f alert-manager.yaml

10.4.4. Using metrics with Minikube

When adding Prometheus and Grafana servers to an Apache Kafka deployment using Minikube, the memory available to the virtual machine should be increased (to 4 GB of RAM, for example, instead of the default 2 GB).

For information on how to increase the default amount of memory, see Installing a local Kubernetes cluster with Minikube

Additional resources

10.5. Enabling the example Grafana dashboards

Use Grafana to provide visualizations of Prometheus metrics on customizable dashboards.

You can use your own Grafana deployment or deploy Grafana using the example metrics configuration files provided by Strimzi. The example files include a configuration file for a Grafana deployment

  • examples/metrics/grafana-install/grafana.yaml

Strimzi also provides example dashboard configuration files for Grafana in JSON format.

  • examples/metrics/grafana-dashboards

This procedure uses the example Grafana configuration file and example dashboards.

The example dashboards are a good starting point for monitoring key metrics, but they don’t show all the metrics supported by Kafka. You can modify the example dashboards or add other metrics, depending on your infrastructure.

Note
 No alert notification rules are defined.

When accessing a dashboard, you can use the port-forward command to forward traffic from the Grafana pod to the host. The name of the Grafana pod is different for each user.

Prerequisites
Procedure

  1. Deploy Grafana.

    kubectl apply -f grafana.yaml

  2. Get the details of the Grafana service.

    kubectl get service grafana

    For example:

    NAME TYPE CLUSTER-IP PORT(S)

    grafana

    ClusterIP

    172.30.123.40

    3000/TCP

    Note the port number for port forwarding.

  3. Use port-forward to redirect the Grafana user interface to localhost:3000:

    kubectl port-forward svc/grafana 3000:3000

  4. In a web browser, access the Grafana login screen using the URL http://localhost:3000.

    The Grafana Log In page appears.

  5. Enter your user name and password, and then click Log In.

    The default Grafana user name and password are both admin. After logging in for the first time, you can change the password.

  6. In Configuration > Data Sources, add Prometheus as a data source.

    • Specify a name

    • Add Prometheus as the type

    • Specify a Prometheus server URL (http://prometheus-operated:9090)

      Save and test the connection when you have added the details.

  7. Click the + icon and then click Import.

  8. In examples/metrics/grafana-dashboards, copy the JSON of the dashboard to import.

  9. Paste the JSON into the text box, and then click Load.

  10. Repeat steps 7-9 for the other example Grafana dashboards.

The imported Grafana dashboards are available to view from the Dashboards home page.

11. Introducing distributed tracing

Distributed tracing tracks the progress of transactions between applications in a distributed system. In a microservices architecture, tracing tracks the progress of transactions between services. Trace data is useful for monitoring application performance and investigating issues with target systems and end-user applications.

In Strimzi, tracing facilitates the end-to-end tracking of messages: from source systems to Kafka, and then from Kafka to target systems and applications. Distributed tracing complements the monitoring of metrics in Grafana dashboards, as well as the component loggers.

Support for tracing is built in to the following Kafka components:

  • MirrorMaker to trace messages from a source cluster to a target cluster

  • Kafka Connect to trace messages consumed and produced by Kafka Connect

  • Kafka Bridge to trace messages between Kafka and HTTP client applications

Tracing is not supported for Kafka brokers.

You enable and configure tracing for these components through their custom resources. You add tracing configuration using spec.template properties.

You enable tracing by specifying a tracing type using the spec.tracing.type property:

opentelemetry

Specify type: opentelemetry to use OpenTelemetry. By Default, OpenTelemetry uses the OTLP (OpenTelemetry Protocol) exporter and endpoint to get trace data. You can specify other tracing systems supported by OpenTelemetry, including Jaeger tracing. To do this, you change the OpenTelemetry exporter and endpoint in the tracing configuration.

jaeger

Specify type:jaeger to use OpenTracing and the Jaeger client to get trace data.

Note

Support for type: jaeger tracing is deprecated. The Jaeger clients are now retired and the OpenTracing project archived. As such, we cannot guarantee their support for future Kafka versions. If possible, we will maintain the support for type: jaeger tracing until June 2023 and remove it afterwards. Please migrate to OpenTelemetry as soon as possible.

11.1. Tracing options

Use OpenTelemetry or OpenTracing (deprecated) with the Jaeger tracing system.

OpenTelemetry and OpenTracing provide API specifications that are independent from the tracing or monitoring system.

You use the APIs to instrument application code for tracing.

  • Instrumented applications generate traces for individual requests across the distributed system.

  • Traces are composed of spans that define specific units of work over time.

Jaeger is a tracing system for microservices-based distributed systems.

  • Jaeger implements the tracing APIs and provides client libraries for instrumentation.

  • The Jaeger user interface allows you to query, filter, and analyze trace data.

The Jaeger user interface showing a simple query

The Jaeger user interface showing a simple query

Additional resources

11.2. Environment variables for tracing

Use environment variables when you are enabling tracing for Kafka components or initializing a tracer for Kafka clients.

Tracing environment variables are subject to change. For the latest information, see the OpenTelemetry documentation and OpenTracing documentation.

The following tables describe the key environment variables for setting up a tracer.

Table 6. OpenTelemetry environment variables
Property Required Description

OTEL_SERVICE_NAME

Yes

The name of the Jaeger tracing service for OpenTelemetry.

OTEL_EXPORTER_JAEGER_ENDPOINT

Yes

The exporter used for tracing.

OTEL_TRACES_EXPORTER

Yes

The exporter used for tracing. Set to otlp by default. If using Jaeger tracing, you need to set this environment variable as jaeger. If you are using another tracing implementation, specify the exporter used.
Table 7. OpenTracing environment variables
Property Required Description

JAEGER_SERVICE_NAME

Yes

The name of the Jaeger tracer service.

JAEGER_AGENT_HOST

No

The hostname for communicating with the jaeger-agent through the User Datagram Protocol (UDP).

JAEGER_AGENT_PORT

No

The port used for communicating with the jaeger-agent through UDP.

11.3. Setting up distributed tracing

Enable distributed tracing in Kafka components by specifying a tracing type in the custom resource. Instrument tracers in Kafka clients for end-to-end tracking of messages.

To set up distributed tracing, follow these procedures in order:

11.3.1. Prerequisites

Before setting up distributed tracing, make sure Jaeger backend components are deployed to your Kubernetes cluster. We recommend using the Jaeger operator for deploying Jaeger on your Kubernetes cluster.

For deployment instructions, see the Jaeger documentation.

Note

Setting up tracing for applications and systems beyond Strimzi is outside the scope of this content.

11.3.2. Enabling tracing in MirrorMaker, Kafka Connect, and Kafka Bridge resources

Distributed tracing is supported for MirrorMaker, MirrorMaker 2.0, Kafka Connect, and the Strimzi Kafka Bridge. Configure the custom resource of the component to specify and enable a tracer service.

Enabling tracing in a resource triggers the following events:

  • Interceptor classes are updated in the integrated consumers and producers of the component.

  • For MirrorMaker, MirrorMaker 2.0, and Kafka Connect, the tracing agent initializes a tracer based on the tracing configuration defined in the resource.

  • For the Kafka Bridge, a tracer based on the tracing configuration defined in the resource is initialized by the Kafka Bridge itself.

You can enable tracing that uses OpenTelemetry or OpenTracing.

Tracing in MirrorMaker and MirrorMaker 2.0

For MirrorMaker and MirrorMaker 2.0, messages are traced from the source cluster to the target cluster. The trace data records messages entering and leaving the MirrorMaker or MirrorMaker 2.0 component.

Tracing in Kafka Connect

For Kafka Connect, only messages produced and consumed by Kafka Connect are traced. To trace messages sent between Kafka Connect and external systems, you must configure tracing in the connectors for those systems.

Tracing in the Kafka Bridge

For the Kafka Bridge, messages produced and consumed by the Kafka Bridge are traced. Incoming HTTP requests from client applications to send and receive messages through the Kafka Bridge are also traced. To have end-to-end tracing, you must configure tracing in your HTTP clients.

Procedure

Perform these steps for each KafkaMirrorMaker, KafkaMirrorMaker2, KafkaConnect, and KafkaBridge resource.

  1. In the spec.template property, configure the tracer service.

    • Use the tracing environment variables as template configuration properties.

    • For OpenTelemetry, set the spec.tracing.type property to opentelemetry.

    • For OpenTracing, set the spec.tracing.type property to jaeger.

    Example tracing configuration for Kafka Connect using OpenTelemetry
    apiVersion: kafka.strimzi.io/v1beta2
kind: KafkaConnect
metadata:
  name: my-connect-cluster
spec:
  #...
  template:
    connectContainer:
      env:
        - name: OTEL_SERVICE_NAME
          value: my-otel-service
        - name: OTEL_EXPORTER_OTLP_ENDPOINT
          value: "http://otlp-host:4317"
  tracing:
    type: opentelemetry
  #...
    Example tracing configuration for MirrorMaker using OpenTelemetry
    apiVersion: kafka.strimzi.io/v1beta2
kind: KafkaMirrorMaker
metadata:
  name: my-mirror-maker
spec:
  #...
  template:
    mirrorMakerContainer:
      env:
        - name: OTEL_SERVICE_NAME
          value: my-otel-service
        - name: OTEL_EXPORTER_OTLP_ENDPOINT
          value: "http://otlp-host:4317"
  tracing:
    type: opentelemetry
#...
    Example tracing configuration for MirrorMaker 2.0 using OpenTelemetry
    apiVersion: kafka.strimzi.io/v1beta2
kind: KafkaMirrorMaker2
metadata:
  name: my-mm2-cluster
spec:
  #...
  template:
    connectContainer:
      env:
        - name: OTEL_SERVICE_NAME
          value: my-otel-service
        - name: OTEL_EXPORTER_OTLP_ENDPOINT
          value: "http://otlp-host:4317"
  tracing:
    type: opentelemetry
#...
    Example tracing configuration for the Kafka Bridge using OpenTelemetry
    apiVersion: kafka.strimzi.io/v1beta2
kind: KafkaBridge
metadata:
  name: my-bridge
spec:
  #...
  template:
    bridgeContainer:
      env:
        - name: OTEL_SERVICE_NAME
          value: my-otel-service
        - name: OTEL_EXPORTER_OTLP_ENDPOINT
          value: "http://otlp-host:4317"
  tracing:
    type: opentelemetry
#...
    Example tracing configuration for Kafka Connect using OpenTracing
    apiVersion: kafka.strimzi.io/v1beta2
kind: KafkaConnect
metadata:
  name: my-connect-cluster
spec:
  #...
  template:
    connectContainer:
      env:
        - name: JAEGER_SERVICE_NAME
          value: my-jaeger-service
        - name: JAEGER_AGENT_HOST
          value: jaeger-agent-name
        - name: JAEGER_AGENT_PORT
          value: "6831"
  tracing:
    type: jaeger
  #...
    Example tracing configuration for MirrorMaker using OpenTracing
    apiVersion: kafka.strimzi.io/v1beta2
kind: KafkaMirrorMaker
metadata:
  name: my-mirror-maker
spec:
  #...
  template:
    mirrorMakerContainer:
      env:
        - name: JAEGER_SERVICE_NAME
          value: my-jaeger-service
        - name: JAEGER_AGENT_HOST
          value: jaeger-agent-name
        - name: JAEGER_AGENT_PORT
          value: "6831"
  tracing:
    type: jaeger
#...
    Example tracing configuration for MirrorMaker 2.0 using OpenTracing
    apiVersion: kafka.strimzi.io/v1beta2
kind: KafkaMirrorMaker2
metadata:
  name: my-mm2-cluster
spec:
  #...
  template:
    connectContainer:
      env:
        - name: JAEGER_SERVICE_NAME
          value: my-jaeger-service
        - name: JAEGER_AGENT_HOST
          value: jaeger-agent-name
        - name: JAEGER_AGENT_PORT
          value: "6831"
  tracing:
    type: jaeger
#...
    Example tracing configuration for the Kafka Bridge using OpenTracing
    apiVersion: kafka.strimzi.io/v1beta2
kind: KafkaBridge
metadata:
  name: my-bridge
spec:
  #...
  template:
    bridgeContainer:
      env:
        - name: JAEGER_SERVICE_NAME
          value: my-jaeger-service
        - name: JAEGER_AGENT_HOST
          value: jaeger-agent-name
        - name: JAEGER_AGENT_PORT
          value: "6831"
  tracing:
    type: jaeger
#...

  2. Create or update the resource:

    kubectl apply -f <resource_configuration_file>

11.3.3. Initializing tracing for Kafka clients

Initialize a tracer, then instrument your client applications for distributed tracing. You can instrument Kafka producer and consumer clients, and Kafka Streams API applications. You can initialize a tracer for OpenTracing or OpenTelemetry.

Configure and initialize a tracer using a set of tracing environment variables.

Procedure

In each client application add the dependencies for the tracer:

  1. Add the Maven dependencies to the pom.xml file for the client application:

    Dependencies for OpenTelemetry
    <dependency>
    <groupId>io.opentelemetry</groupId>
    <artifactId>opentelemetry-sdk-extension-autoconfigure</artifactId>
    <version>1.18.0-alpha</version>
</dependency>
<dependency>
  <groupId>io.opentelemetry.instrumentation</groupId>
  <artifactId>opentelemetry-kafka-clients-{OpenTelemetryKafkaClient}</artifactId>
  <version>1.18.0-alpha</version>
</dependency>
<dependency>
  <groupId>io.opentelemetry</groupId>
  <artifactId>opentelemetry-exporter-otlp</artifactId>
  <version>1.18.0</version>
</dependency>
    Dependencies for OpenTracing
    <dependency>
    <groupId>io.jaegertracing</groupId>
    <artifactId>jaeger-client</artifactId>
    <version>1.3.2</version>
</dependency>
<dependency>
  <groupId>io.opentracing.contrib</groupId>
  <artifactId>opentracing-kafka-client</artifactId>
  <version>0.1.15</version>
</dependency>

  2. Define the configuration of the tracer using the tracing environment variables.

  3. Create a tracer, which is initialized with the environment variables:

    Creating a tracer for OpenTelemetry
    OpenTelemetry ot = GlobalOpenTelemetry.get();
    Creating a tracer for OpenTracing
    Tracer tracer = Configuration.fromEnv().getTracer();

  4. Register the tracer as a global tracer:

    GlobalTracer.register(tracer);

  5. Instrument your client:

11.3.4. Instrumenting producers and consumers for tracing

Instrument application code to enable tracing in Kafka producers and consumers. Use a decorator pattern or interceptors to instrument your Java producer and consumer application code for tracing. You can then record traces when messages are produced or retrieved from a topic.

OpenTelemetry and OpenTracing instrumentation projects provide classes that support instrumentation of producers and consumers.

Decorator instrumentation

For decorator instrumentation, create a modified producer or consumer instance for tracing. Decorator instrumentation is different for OpenTelemetry and OpenTracing.

Interceptor instrumentation

For interceptor instrumentation, add the tracing capability to the consumer or producer configuration. Interceptor instrumentation is the same for OpenTelemetry and OpenTracing.

Prerequisites

  • You have initialized tracing for the client.

    You enable instrumentation in producer and consumer applications by adding the tracing JARs as dependencies to your project.

Procedure

Perform these steps in the application code of each producer and consumer application. Instrument your client application code using either a decorator pattern or interceptors.

  • To use a decorator pattern, create a modified producer or consumer instance to send or receive messages.

    You pass the original KafkaProducer or KafkaConsumer class.

    Example decorator instrumentation for OpenTelemetry
    // Producer instance
Producer < String, String > op = new KafkaProducer < > (
    configs,
    new StringSerializer(),
    new StringSerializer()
    );
    Producer < String, String > producer = tracing.wrap(op);
KafkaTracing tracing = KafkaTracing.create(GlobalOpenTelemetry.get());
producer.send(...);

//consumer instance
Consumer<String, String> oc = new KafkaConsumer<>(
    configs,
    new StringDeserializer(),
    new StringDeserializer()
    );
    Consumer<String, String> consumer = tracing.wrap(oc);
consumer.subscribe(Collections.singleton("mytopic"));
ConsumerRecords<Integer, String> records = consumer.poll(1000);
ConsumerRecord<Integer, String> record = ...
SpanContext spanContext = TracingKafkaUtils.extractSpanContext(record.headers(), tracer);
    Example decorator instrumentation for OpenTracing
    //producer instance
KafkaProducer<Integer, String> producer = new KafkaProducer<>(senderProps);
TracingKafkaProducer<Integer, String> tracingProducer = new TracingKafkaProducer<>(producer, tracer);
TracingKafkaProducer.send(...)

//consumer instance
KafkaConsumer<Integer, String> consumer = new KafkaConsumer<>(consumerProps);
TracingKafkaConsumer<Integer, String> tracingConsumer = new TracingKafkaConsumer<>(consumer, tracer);
tracingConsumer.subscribe(Collections.singletonList("mytopic"));
ConsumerRecords<Integer, String> records = tracingConsumer.poll(1000);
ConsumerRecord<Integer, String> record = ...
SpanContext spanContext = TracingKafkaUtils.extractSpanContext(record.headers(), tracer);

  • To use interceptors, set the interceptor class in the producer or consumer configuration.

    You use the KafkaProducer and KafkaConsumer classes in the usual way. The TracingProducerInterceptor and TracingConsumerInterceptor interceptor classes take care of the tracing capability.

    Example producer configuration using interceptors
    senderProps.put(ProducerConfig.INTERCEPTOR_CLASSES_CONFIG,
    TracingProducerInterceptor.class.getName());

KafkaProducer<Integer, String> producer = new KafkaProducer<>(senderProps);
producer.send(...);
    Example consumer configuration using interceptors
    consumerProps.put(ConsumerConfig.INTERCEPTOR_CLASSES_CONFIG,
    TracingConsumerInterceptor.class.getName());

KafkaConsumer<Integer, String> consumer = new KafkaConsumer<>(consumerProps);
consumer.subscribe(Collections.singletonList("messages"));
ConsumerRecords<Integer, String> records = consumer.poll(1000);
ConsumerRecord<Integer, String> record = ...
SpanContext spanContext = TracingKafkaUtils.extractSpanContext(record.headers(), tracer);

11.3.5. Instrumenting Kafka Streams applications for tracing

Instrument application code to enable tracing in Kafka Streams API applications. Use a decorator pattern or interceptors to instrument your Kafka Streams API applications for tracing. You can then record traces when messages are produced or retrieved from a topic.

Decorator instrumentation

For decorator instrumentation, create a modified Kafka Streams instance for tracing. The OpenTracing instrumentation project provides a TracingKafkaClientSupplier class that supports instrumentation of Kafka Streams. You create a wrapped instance of the TracingKafkaClientSupplier supplier interface, which provides tracing instrumentation for Kafka Streams. For OpenTelemetry, the process is the same but you need to create a custom TracingKafkaClientSupplier class to provide the support.

Interceptor instrumentation

For interceptor instrumentation, add the tracing capability to the Kafka Streams producer and consumer configuration.

Prerequisites

  • You have initialized tracing for the client.

    You enable instrumentation in Kafka Streams applications by adding the tracing JARs as dependencies to your project.

  • To instrument Kafka Streams with OpenTelemetry, you’ll need to write a custom TracingKafkaClientSupplier.

  • The custom TracingKafkaClientSupplier can extend Kafka’s DefaultKafkaClientSupplier, overriding the producer and consumer creation methods to wrap the instances with the telemetry-related code.

    Example custom TracingKafkaClientSupplier
    private class TracingKafkaClientSupplier extends DefaultKafkaClientSupplier {
    @Override
    public Producer<byte[], byte[]> getProducer(Map<String, Object> config) {
        KafkaTelemetry telemetry = KafkaTelemetry.create(GlobalOpenTelemetry.get());
        return telemetry.wrap(super.getProducer(config));
    }

    @Override
    public Consumer<byte[], byte[]> getConsumer(Map<String, Object> config) {
        KafkaTelemetry telemetry = KafkaTelemetry.create(GlobalOpenTelemetry.get());
        return telemetry.wrap(super.getConsumer(config));
    }

    @Override
    public Consumer<byte[], byte[]> getRestoreConsumer(Map<String, Object> config) {
        return this.getConsumer(config);
    }

    @Override
    public Consumer<byte[], byte[]> getGlobalConsumer(Map<String, Object> config) {
        return this.getConsumer(config);
    }
}
Procedure

Perform these steps for each Kafka Streams API application.

  • To use a decorator pattern, create an instance of the TracingKafkaClientSupplier supplier interface, then provide the supplier interface to KafkaStreams.

    Example decorator instrumentation
    KafkaClientSupplier supplier = new TracingKafkaClientSupplier(tracer);
KafkaStreams streams = new KafkaStreams(builder.build(), new StreamsConfig(config), supplier);
streams.start();

  • To use interceptors, set the interceptor class in the Kafka Streams producer and consumer configuration.

    The TracingProducerInterceptor and TracingConsumerInterceptor interceptor classes take care of the tracing capability.

    Example producer and consumer configuration using interceptors
    props.put(StreamsConfig.PRODUCER_PREFIX + ProducerConfig.INTERCEPTOR_CLASSES_CONFIG, TracingProducerInterceptor.class.getName());
props.put(StreamsConfig.CONSUMER_PREFIX + ConsumerConfig.INTERCEPTOR_CLASSES_CONFIG, TracingConsumerInterceptor.class.getName());

11.3.6. Introducing a different OpenTelemetry tracing system

Instead of the default OTLP system, you can specify other tracing systems that are supported by OpenTelemetry. You do this by adding the required artifacts to the Kafka image provided with Strimzi. Any required implementation specific environment variables must also be set. You then enable the new tracing implementation using the OTEL_TRACES_EXPORTER environment variable.

This procedure shows how to implement Zipkin tracing.

Procedure

  1. Add the tracing artifacts to the /opt/kafka/libs/ directory of the Strimzi Kafka image.

    You can use the Kafka container image on the Container Registry as a base image for creating a new custom image.

    OpenTelemetry artifact for Zipkin
    io.opentelemetry:opentelemetry-exporter-zipkin

  2. Set the tracing exporter and endpoint for the new tracing implementation.

    Example Zikpin tracer configuration
    apiVersion: kafka.strimzi.io/v1beta2
kind: KafkaMirrorMaker2
metadata:
  name: my-mm2-cluster
spec:
  #...
  template:
    connectContainer:
      env:
        - name: OTEL_SERVICE_NAME
          value: my-zipkin-service
        - name: OTEL_EXPORTER_ZIPKIN_ENDPOINT
          value: http://zipkin-exporter-host-name:9411/api/v2/spans # (1)
        - name: OTEL_TRACES_EXPORTER
          value: zipkin # (2)
  tracing:
    type: opentelemetry
#...

    1. Specifies the Zipkin endpoint to connect to.

    2. The Zipkin exporter.

11.3.7. Custom span names

A tracing span is a logical unit of work in Jaeger, with an operation name, start time, and duration. Spans have built-in names, but you can specify custom span names in your Kafka client instrumentation where used.

Specifying custom span names is optional and only applies when using a decorator pattern in producer and consumer client instrumentation or Kafka Streams instrumentation.

Specifying span names for OpenTelemetry

Custom span names cannot be specified directly with OpenTelemetry. Instead, you retrieve span names by adding code to your client application to extract additional tags and attributes.

Example code to extract attributes
//Defines attribute extraction for a producer
private static class ProducerAttribExtractor implements AttributesExtractor < ProducerRecord < ? , ? > , Void > {
    @Override
    public void onStart(AttributesBuilder attributes, ProducerRecord < ? , ? > producerRecord) {
        set(attributes, AttributeKey.stringKey("prod_start"), "prod1");
    }
    @Override
    public void onEnd(AttributesBuilder attributes, ProducerRecord < ? , ? > producerRecord, @Nullable Void unused, @Nullable Throwable error) {
        set(attributes, AttributeKey.stringKey("prod_end"), "prod2");
    }
}
//Defines attribute extraction for a consumer
private static class ConsumerAttribExtractor implements AttributesExtractor < ConsumerRecord < ? , ? > , Void > {
    @Override
    public void onStart(AttributesBuilder attributes, ConsumerRecord < ? , ? > producerRecord) {
        set(attributes, AttributeKey.stringKey("con_start"), "con1");
    }
    @Override
    public void onEnd(AttributesBuilder attributes, ConsumerRecord < ? , ? > producerRecord, @Nullable Void unused, @Nullable Throwable error) {
        set(attributes, AttributeKey.stringKey("con_end"), "con2");
    }
}
//Extracts the attributes
public static void main(String[] args) throws Exception {
        Map < String, Object > configs = new HashMap < > (Collections.singletonMap(ProducerConfig.BOOTSTRAP_SERVERS_CONFIG, "localhost:9092"));
        System.setProperty("otel.traces.exporter", "jaeger");
        System.setProperty("otel.service.name", "myapp1");
        KafkaTracing tracing = KafkaTracing.newBuilder(GlobalOpenTelemetry.get())
            .addProducerAttributesExtractors(new ProducerAttribExtractor())
            .addConsumerAttributesExtractors(new ConsumerAttribExtractor())
            .build();
Specifying span names for OpenTracing

To specify custom span names for OpenTracing, pass a BiFunction object as an additional argument when instrumenting producers and consumers.

For more information on built-in names and specifying custom span names to instrument client application code in a decorator pattern, see the OpenTracing Apache Kafka client instrumentation.

12. Upgrading Strimzi

Strimzi can be upgraded to version latest to take advantage of new features and enhancements, performance improvements, and security options.

As part of the upgrade, you upgrade Kafka to the latest supported version. Each Kafka release introduces new features, improvements, and bug fixes to your Strimzi deployment.

Strimzi can be downgraded to the previous version if you encounter issues with the newer version.

Released versions of Strimzi are available from the GitHub releases page.

Upgrade downtime and availability

If topics are configured for high availability, upgrading Strimzi should not cause any downtime for consumers and producers that publish and read data from those topics. Highly available topics have a replication factor of at least 3 and partitions distributed evenly among the brokers.

Upgrading Strimzi triggers rolling updates, where all brokers are restarted in turn, at different stages of the process. During rolling updates, not all brokers are online, so overall cluster availability is temporarily reduced. A reduction in cluster availability increases the chance that a broker failure will result in lost messages.

12.1. Strimzi upgrade paths

Two Strimzi upgrade paths are possible.

Incremental upgrade

Upgrading Strimzi from the previous minor version to version latest.

Multi-version upgrade

Upgrading Strimzi from an old version to version latest within a single upgrade (skipping one or more intermediate versions).

For example, upgrading from Strimzi 0.24.0 directly to Strimzi latest.

Note
 A multi-version Strimzi upgrade might still support the current version of a Kafka deployment.

12.1.1. Supported Kafka versions

Decide which Kafka version to upgrade to before starting the Strimzi upgrade process. You can review supported Kafka versions in the Supported versions table.

  • The Operators column lists all released Strimzi versions (the Strimzi version is often called the "Operator version").

  • The Kafka versions column lists the supported Kafka versions for each Strimzi version.

You can only use a Kafka version supported by the version of Strimzi you are using. You can upgrade to a higher Kafka version as long as it is supported by your version of Strimzi. In some cases, you can also downgrade to a previous supported Kafka version.

12.1.2. Upgrading from a Strimzi version earlier than 0.22

If you are upgrading to the latest version of Strimzi from a version prior to version 0.22, do the following:

  1. Upgrade Strimzi to version 0.22 following the standard sequence.

  2. Convert Strimzi custom resources to v1beta2 using the API conversion tool provided with Strimzi.

  3. Do one of the following:

    • Upgrade Strimzi to a version between 0.23 and 0.26 (where the ControlPlaneListener feature gate is disabled by default).

    • Upgrade Strimzi to a version between 0.27 and 0.31 (where the ControlPlaneListener feature gate is enabled by default) with the ControlPlaneListener feature gate disabled.

  4. Enable the ControlPlaneListener feature gate.

  5. Upgrade to Strimzi latest following the standard sequence.

Strimzi custom resources started using the v1beta2 API version in release 0.22. CRDs and custom resources must be converted before upgrading to Strimzi 0.23 or newer. For information on using the API conversion tool, see the Strimzi 0.24.0 upgrade documentation.

Note
 As an alternative to first upgrading to version 0.22, you can install the custom resources from version 0.22 and then convert the resources.

The ControlPlaneListener feature is now permanently enabled in Strimzi. You must upgrade to a version of Strimzi where it is disabled, then enable it using the STRIMZI_FEATURE_GATES environment variable in the Cluster Operator configuration.

Disabling the ControlPlaneListener feature gate
env:
  - name: STRIMZI_FEATURE_GATES
    value: -ControlPlaneListener
Enabling the ControlPlaneListener feature gate
env:
  - name: STRIMZI_FEATURE_GATES
    value: +ControlPlaneListener

12.2. Required upgrade sequence

To upgrade brokers and clients without downtime, you must complete the Strimzi upgrade procedures in the following order:

  1. Make sure your Kubernetes cluster version is supported.

    Strimzi latest requires Kubernetes 1.19 and later.

    You can upgrade Kubernetes with minimal downtime.

  2. Upgrade the Cluster Operator.

  3. Upgrade all Kafka brokers and client applications to the latest supported Kafka version.

  4. Optional: Upgrade consumers and Kafka Streams applications to use the incremental cooperative rebalance protocol for partition rebalances.

12.3. Upgrading Kubernetes with minimal downtime

If you are upgrading Kubernetes, refer to the Kubernetes upgrade documentation to check the upgrade path and the steps to upgrade your nodes correctly. Before upgrading Kubernetes, check the supported versions for your version of Strimzi.

When performing your upgrade, you’ll want to keep your Kafka clusters available.

You can employ one of the following strategies:

  1. Configuring pod disruption budgets

  2. Rolling pods by one of these methods:

    1. Using the Strimzi Drain Cleaner

    2. Manually by applying an annotation to your pod

You have to configure the pod disruption budget before using one of the methods to roll your pods.

For Kafka to stay operational, topics must also be replicated for high availability. This requires topic configuration that specifies a replication factor of at least 3 and a minimum number of in-sync replicas to 1 less than the replication factor.

Kafka topic replicated for high availability
apiVersion: kafka.strimzi.io/v1beta2
kind: KafkaTopic
metadata:
  name: my-topic
  labels:
    strimzi.io/cluster: my-cluster
spec:
  partitions: 1
  replicas: 3
  config:
    # ...
    min.insync.replicas: 2
    # ...

In a highly available environment, the Cluster Operator maintains a minimum number of in-sync replicas for topics during the upgrade process so that there is no downtime.

12.3.1. Rolling pods using the Strimzi Drain Cleaner

You can use the Strimzi Drain Cleaner tool to evict nodes during an upgrade. The Strimzi Drain Cleaner annotates pods with a rolling update pod annotation. This informs the Cluster Operator to perform a rolling update of an evicted pod.

A pod disruption budget allows only a specified number of pods to be unavailable at a given time. During planned maintenance of Kafka broker pods, a pod disruption budget ensures Kafka continues to run in a highly available environment.

You specify a pod disruption budget using a template customization for a Kafka component. By default, pod disruption budgets allow only a single pod to be unavailable at a given time.

To do this, you set maxUnavailable to 0 (zero). Reducing the maximum pod disruption budget to zero prevents voluntary disruptions, so pods must be evicted manually.

Specifying a pod disruption budget
apiVersion: kafka.strimzi.io/v1beta2
kind: Kafka
metadata:
  name: my-cluster
  namespace: myproject
spec:
  kafka:
    # ...
    template:
      podDisruptionBudget:
        maxUnavailable: 0
# ...

12.3.2. Rolling pods manually while keeping topics available

During an upgrade, you can trigger a manual rolling update of pods through the Cluster Operator. Using Pod resources, rolling updates restart the pods of resources with new pods. As with using the Strimzi Drain Cleaner, you’ll need to set the maxUnavailable value to zero for the pod disruption budget.

You need to watch the pods that need to be drained. You then add a pod annotation to make the update.

Here, the annotation updates a Kafka broker.

Performing a manual rolling update on a Kafka broker pod
kubectl annotate pod <cluster_name>-kafka-<index> strimzi.io/manual-rolling-update=true

You replace <cluster_name> with the name of the cluster. Kafka broker pods are named <cluster-name>-kafka-<index>, where <index> starts at zero and ends at the total number of replicas minus one. For example, my-cluster-kafka-0.

Additional resources

12.4. Upgrading the Cluster Operator

Use the same method to upgrade the Cluster Operator as the initial method of deployment.

Using installation files

If you deployed the Cluster Operator using the installation YAML files, perform your upgrade by modifying the Operator installation files, as described in Upgrading the Cluster Operator.

Using the OperatorHub.io

If you deployed Strimzi from OperatorHub.io, use the Operator Lifecycle Manager (OLM) to change the update channel for the Strimzi operators to a new Strimzi version.

Updating the channel starts one of the following types of upgrade, depending on your chosen upgrade strategy:

  • An automatic upgrade is initiated

  • A manual upgrade that requires approval before installation begins

Note
 If you subscribe to the stable channel, you can get automatic updates without changing channels. However, enabling automatic updates is not recommended because of the potential for missing any pre-installation upgrade steps. Use automatic upgrades only on version-specific channels.

For more information on using OperatorHub.io to upgrade Operators, see the Operator Lifecycle Manager documentation.

Using a Helm chart

If you deployed the Cluster Operator using a Helm chart, use helm upgrade.

The helm upgrade command does not upgrade the Custom Resource Definitions for Helm. Install the new CRDs manually after upgrading the Cluster Operator. You can access the CRDs from the GitHub releases page or find them in the crd subdirectory inside the Helm Chart.

12.4.1. Upgrading the Cluster Operator returns Kafka version error

If you upgrade the Cluster Operator and get an unsupported Kafka version error, your Kafka cluster deployment has an older Kafka version that is not supported by the new operator version. This error applies to all installation methods.

If this error occurs, upgrade Kafka to a supported Kafka version. Change the spec.kafka.version in the Kafka resource to the supported version.

You can use kubectl to check for error messages like this in the status of the Kafka resource.

Checking the Kafka status for errors
kubectl get kafka <kafka_cluster_name> -n <namespace> -o jsonpath='{.status.conditions}'

Replace <kafka_cluster_name> with the name of your Kafka cluster and <namespace> with the Kubernetes namespace where the pod is running.

12.4.2. Upgrading the Cluster Operator using installation files

This procedure describes how to upgrade a Cluster Operator deployment to use Strimzi latest.

Follow this procedure if you deployed the Cluster Operator using the installation YAML files.

The availability of Kafka clusters managed by the Cluster Operator is not affected by the upgrade operation.

Note
 Refer to the documentation supporting a specific version of Strimzi for information on how to upgrade to that version.
Prerequisites
Procedure

  1. Take note of any configuration changes made to the existing Cluster Operator resources (in the /install/cluster-operator directory). Any changes will be overwritten by the new version of the Cluster Operator.

  2. Update your custom resources to reflect the supported configuration options available for Strimzi version latest.

  3. Update the Cluster Operator.

    1. Modify the installation files for the new Cluster Operator version according to the namespace the Cluster Operator is running in.

      On Linux, use:

      sed -i 's/namespace: .*/namespace: my-cluster-operator-namespace/' install/cluster-operator/*RoleBinding*.yaml

      On MacOS, use:

      sed -i '' 's/namespace: .*/namespace: my-cluster-operator-namespace/' install/cluster-operator/*RoleBinding*.yaml

    2. If you modified one or more environment variables in your existing Cluster Operator Deployment, edit the install/cluster-operator/060-Deployment-strimzi-cluster-operator.yaml file to use those environment variables.

  4. When you have an updated configuration, deploy it along with the rest of the installation resources:

    kubectl replace -f install/cluster-operator

    Wait for the rolling updates to complete.

  5. If the new Operator version no longer supports the Kafka version you are upgrading from, the Cluster Operator returns an error message to say the version is not supported. Otherwise, no error message is returned.

    • If the error message is returned, upgrade to a Kafka version that is supported by the new Cluster Operator version:

      1. Edit the Kafka custom resource.

      2. Change the spec.kafka.version property to a supported Kafka version.

    • If the error message is not returned, go to the next step. You will upgrade the Kafka version later.

  6. Get the image for the Kafka pod to ensure the upgrade was successful:

    kubectl get pods my-cluster-kafka-0 -o jsonpath='{.spec.containers[0].image}'

    The image tag shows the new Operator version. For example:

    quay.io/strimzi/kafka:latest-kafka-3.3.2

Your Cluster Operator was upgraded to version latest but the version of Kafka running in the cluster it manages is unchanged.

Following the Cluster Operator upgrade, you must perform a Kafka upgrade.

12.5. Upgrading Kafka

After you have upgraded your Cluster Operator to latest, the next step is to upgrade all Kafka brokers to the latest supported version of Kafka.

Kafka upgrades are performed by the Cluster Operator through rolling updates of the Kafka brokers.

The Cluster Operator initiates rolling updates based on the Kafka cluster configuration.

If Kafka.spec.kafka.config contains…​ The Cluster Operator initiates…​

Both the inter.broker.protocol.version and the log.message.format.version.

A single rolling update. After the update, the inter.broker.protocol.version must be updated manually, followed by log.message.format.version. Changing each will trigger a further rolling update.

Either the inter.broker.protocol.version or the log.message.format.version.

Two rolling updates.

No configuration for the inter.broker.protocol.version or the log.message.format.version.

Two rolling updates.
Important
 From Kafka 3.0.0, when the inter.broker.protocol.version is set to 3.0 or higher, the log.message.format.version option is ignored and doesn’t need to be set. The log.message.format.version property for brokers and the message.format.version property for topics are deprecated and will be removed in a future release of Kafka.

As part of the Kafka upgrade, the Cluster Operator initiates rolling updates for ZooKeeper.

  • A single rolling update occurs even if the ZooKeeper version is unchanged.

  • Additional rolling updates occur if the new version of Kafka requires a new ZooKeeper version.

12.5.1. Kafka versions

Kafka’s log message format version and inter-broker protocol version specify, respectively, the log format version appended to messages and the version of the Kafka protocol used in a cluster. To ensure the correct versions are used, the upgrade process involves making configuration changes to existing Kafka brokers and code changes to client applications (consumers and producers).

The following table shows the differences between Kafka versions:

Table 8. Kafka version differences
Kafka version Inter-broker protocol version Log message format version ZooKeeper version

3.2.0

3.2

3.2

3.6.3

3.2.1

3.2

3.2

3.6.3

3.2.3

3.2

3.2

3.6.3

3.3.1

3.3

3.3

3.6.3

3.3.2

3.3

3.3

3.6.3
Inter-broker protocol version

In Kafka, the network protocol used for inter-broker communication is called the inter-broker protocol. Each version of Kafka has a compatible version of the inter-broker protocol. The minor version of the protocol typically increases to match the minor version of Kafka, as shown in the preceding table.

The inter-broker protocol version is set cluster wide in the Kafka resource. To change it, you edit the inter.broker.protocol.version property in Kafka.spec.kafka.config.

Log message format version

When a producer sends a message to a Kafka broker, the message is encoded using a specific format. The format can change between Kafka releases, so messages specify which version of the message format they were encoded with.

The properties used to set a specific message format version are as follows:

  • message.format.version property for topics

  • log.message.format.version property for Kafka brokers

From Kafka 3.0.0, the message format version values are assumed to match the inter.broker.protocol.version and don’t need to be set. The values reflect the Kafka version used.

When upgrading to Kafka 3.0.0 or higher, you can remove these settings when you update the inter.broker.protocol.version. Otherwise, set the message format version based on the Kafka version you are upgrading to.

The default value of message.format.version for a topic is defined by the log.message.format.version that is set on the Kafka broker. You can manually set the message.format.version of a topic by modifying its topic configuration.

12.5.2. Strategies for upgrading clients

The right approach to upgrading your client applications (including Kafka Connect connectors) depends on your particular circumstances.

Consuming applications need to receive messages in a message format that they understand. You can ensure that this is the case in one of two ways:

  • By upgrading all the consumers for a topic before upgrading any of the producers.

  • By having the brokers down-convert messages to an older format.

Using broker down-conversion puts extra load on the brokers, so it is not ideal to rely on down-conversion for all topics for a prolonged period of time. For brokers to perform optimally they should not be down converting messages at all.

Broker down-conversion is configured in two ways:

  • The topic-level message.format.version configures it for a single topic.

  • The broker-level log.message.format.version is the default for topics that do not have the topic-level message.format.version configured.

Messages published to a topic in a new-version format will be visible to consumers, because brokers perform down-conversion when they receive messages from producers, not when they are sent to consumers.

Common strategies you can use to upgrade your clients are described as follows. Other strategies for upgrading client applications are also possible.

Important
 The steps outlined in each strategy change slightly when upgrading to Kafka 3.0.0 or later. From Kafka 3.0.0, the message format version values are assumed to match the inter.broker.protocol.version and don’t need to be set.
Broker-level consumers first strategy

  1. Upgrade all the consuming applications.

  2. Change the broker-level log.message.format.version to the new version.

  3. Upgrade all the producing applications.

This strategy is straightforward, and avoids any broker down-conversion. However, it assumes that all consumers in your organization can be upgraded in a coordinated way, and it does not work for applications that are both consumers and producers. There is also a risk that, if there is a problem with the upgraded clients, new-format messages might get added to the message log so that you cannot revert to the previous consumer version.

Topic-level consumers first strategy

For each topic:

  1. Upgrade all the consuming applications.

  2. Change the topic-level message.format.version to the new version.

  3. Upgrade all the producing applications.

This strategy avoids any broker down-conversion, and means you can proceed on a topic-by-topic basis. It does not work for applications that are both consumers and producers of the same topic. Again, it has the risk that, if there is a problem with the upgraded clients, new-format messages might get added to the message log.

Topic-level consumers first strategy with down conversion

For each topic:

  1. Change the topic-level message.format.version to the old version (or rely on the topic defaulting to the broker-level log.message.format.version).

  2. Upgrade all the consuming and producing applications.

  3. Verify that the upgraded applications function correctly.

  4. Change the topic-level message.format.version to the new version.

This strategy requires broker down-conversion, but the load on the brokers is minimized because it is only required for a single topic (or small group of topics) at a time. It also works for applications that are both consumers and producers of the same topic. This approach ensures that the upgraded producers and consumers are working correctly before you commit to using the new message format version.

The main drawback of this approach is that it can be complicated to manage in a cluster with many topics and applications.

Note
 It is also possible to apply multiple strategies. For example, for the first few applications and topics the "per-topic consumers first, with down conversion" strategy can be used. When this has proved successful another, more efficient strategy can be considered acceptable to use instead.

12.5.3. Kafka version and image mappings

When upgrading Kafka, consider your settings for the STRIMZI_KAFKA_IMAGES environment variable and the Kafka.spec.kafka.version property.

  • Each Kafka resource can be configured with a Kafka.spec.kafka.version.

  • The Cluster Operator’s STRIMZI_KAFKA_IMAGES environment variable provides a mapping between the Kafka version and the image to be used when that version is requested in a given Kafka resource.

    • If Kafka.spec.kafka.image is not configured, the default image for the given version is used.

    • If Kafka.spec.kafka.image is configured, the default image is overridden.

Warning
 The Cluster Operator cannot validate that an image actually contains a Kafka broker of the expected version. Take care to ensure that the given image corresponds to the given Kafka version.

12.5.4. Upgrading Kafka brokers and client applications

Upgrade a Strimzi Kafka cluster to the latest supported Kafka version and inter-broker protocol version.

You should also choose a strategy for upgrading clients. Kafka clients are upgraded in step 6 of this procedure.

Prerequisites

  • The Cluster Operator is up and running.

  • Before you upgrade the Strimzi Kafka cluster, check that the Kafka.spec.kafka.config properties of the Kafka resource do not contain configuration options that are not supported in the new Kafka version.

Procedure

  1. Update the Kafka cluster configuration:

    kubectl edit kafka <my_cluster>

  2. If configured, check that the inter.broker.protocol.version and log.message.format.version properties are set to the current version.

    For example, the current version is 3.2 if upgrading from Kafka version 3.2.3 to 3.3.2:

    kind: Kafka
spec:
  # ...
  kafka:
    version: 3.2.3
    config:
      log.message.format.version: "3.2"
      inter.broker.protocol.version: "3.2"
      # ...

    If log.message.format.version and inter.broker.protocol.version are not configured, Strimzi automatically updates these versions to the current defaults after the update to the Kafka version in the next step.

    Note
    		 The value of log.message.format.version and inter.broker.protocol.version must be strings to prevent them from being interpreted as floating point numbers.

  3. Change the Kafka.spec.kafka.version to specify the new Kafka version; leave the log.message.format.version and inter.broker.protocol.version at the defaults for the current Kafka version.

    Note

    Changing the kafka.version ensures that all brokers in the cluster will be upgraded to start using the new broker binaries. During this process, some brokers are using the old binaries while others have already upgraded to the new ones. Leaving the inter.broker.protocol.version unchanged at the current setting ensures that the brokers can continue to communicate with each other throughout the upgrade.

    For example, if upgrading from Kafka 3.2.3 to 3.3.2:

    apiVersion: kafka.strimzi.io/v1beta2
kind: Kafka
spec:
  # ...
  kafka:
    version: 3.3.2 (1)
    config:
      log.message.format.version: "3.2" (2)
      inter.broker.protocol.version: "3.2" (3)
      # ...

    1. Kafka version is changed to the new version.

    2. Message format version is unchanged.

    3. Inter-broker protocol version is unchanged.

    Warning
    		 You cannot downgrade Kafka if the inter.broker.protocol.version for the new Kafka version changes. The inter-broker protocol version determines the schemas used for persistent metadata stored by the broker, including messages written to __consumer_offsets. The downgraded cluster will not understand the messages.

  4. If the image for the Kafka cluster is defined in the Kafka custom resource, in Kafka.spec.kafka.image, update the image to point to a container image with the new Kafka version.

    See Kafka version and image mappings

  5. Save and exit the editor, then wait for rolling updates to complete.

    Check the progress of the rolling updates by watching the pod state transitions:

    kubectl get pods my-cluster-kafka-0 -o jsonpath='{.spec.containers[0].image}'

    The rolling updates ensure that each pod is using the broker binaries for the new version of Kafka.

  6. Depending on your chosen strategy for upgrading clients, upgrade all client applications to use the new version of the client binaries.

    If required, set the version property for Kafka Connect and MirrorMaker as the new version of Kafka:

    1. For Kafka Connect, update KafkaConnect.spec.version.

    2. For MirrorMaker, update KafkaMirrorMaker.spec.version.

    3. For MirrorMaker 2.0, update KafkaMirrorMaker2.spec.version.

  7. If configured, update the Kafka resource to use the new inter.broker.protocol.version version. Otherwise, go to step 9.

    For example, if upgrading to Kafka 3.3.2:

    apiVersion: kafka.strimzi.io/v1beta2
kind: Kafka
spec:
  # ...
  kafka:
    version: 3.3.2
    config:
      log.message.format.version: "3.2"
      inter.broker.protocol.version: "3.3"
      # ...

  8. Wait for the Cluster Operator to update the cluster.

  9. If configured, update the Kafka resource to use the new log.message.format.version version. Otherwise, go to step 10.

    For example, if upgrading to Kafka 3.3.2:

    apiVersion: kafka.strimzi.io/v1beta2
kind: Kafka
spec:
  # ...
  kafka:
    version: 3.3.2
    config:
      log.message.format.version: "3.3"
      inter.broker.protocol.version: "3.3"
      # ...
    Important
    		 From Kafka 3.0.0, when the inter.broker.protocol.version is set to 3.0 or higher, the log.message.format.version option is ignored and doesn’t need to be set.

  10. Wait for the Cluster Operator to update the cluster.

    • The Kafka cluster and clients are now using the new Kafka version.

    • The brokers are configured to send messages using the inter-broker protocol version and message format version of the new version of Kafka.

Following the Kafka upgrade, if required, you can upgrade consumers to use the incremental cooperative rebalance protocol.

12.6. Switching to FIPS mode when upgrading Strimzi

Upgrade Strimzi to run in FIPS mode on FIPS-enabled Kubernetes clusters. Until Strimzi 0.33, running on FIPS-enabled Kubernetes clusters was possible only by disabling FIPS mode using the FIPS_MODE environment variable. From release 0.33, Strimzi supports FIPS mode. If you run Strimzi on a FIPS-enabled Kubernetes cluster with the FIPS_MODE set to disabled, you can enable it by following this procedure.

Prerequisites

  • FIPS-enabled Kubernetes cluster

  • An existing Cluster Operator deployment with the FIPS_MODE environment variable set to disabled

Procedure

  1. Upgrade the Cluster Operator to version 0.33 or newer but keep the FIPS_MODE environment variable set to disabled.

  2. If you initially deployed a Strimzi version older than 0.30, it might use old encryption and digest algorithms in its PKCS #12 stores, which are not supported with FIPS enabled. To recreate the certificates with updated algorithms, renew the cluster and clients CA certificates.

    1. To renew the CAs generated by the Cluster Operator, add the force-renew annotation to the CA secrets to trigger a renewal.

    2. To renew your own CAs, add the new certificate to the CA secret and update the ca-cert-generation annotation with a higher incremental value to capture the update.

  3. If you use SCRAM-SHA-512 authentication, check the password length of your users. If they are less than 32 characters long, generate a new password in one of the following ways:

    1. Delete the user secret so that the User Operator generates a new one with a new password of sufficient length.

    2. If you provided your password using the .spec.authentication.password properties of the KafkaUser custom resource, update the password in the Kubernetes secret referenced in the same password configuration. Don’t forget to update your clients to use the new passwords.

  4. Ensure that the CA certificates are using the correct algorithms and the SCRAM-SHA-512 passwords are of sufficient length. You can then enable the FIPS mode.

  5. Remove the FIPS_MODE environment variable from the Cluster Operator deployment. This restarts the Cluster Operator and rolls all the operands to enable the FIPS mode. After the restart is complete, all Kafka clusters now run with FIPS mode enabled.

12.7. Upgrading consumers to cooperative rebalancing

You can upgrade Kafka consumers and Kafka Streams applications to use the incremental cooperative rebalance protocol for partition rebalances instead of the default eager rebalance protocol. The new protocol was added in Kafka 2.4.0.

Consumers keep their partition assignments in a cooperative rebalance and only revoke them at the end of the process, if needed to achieve a balanced cluster. This reduces the unavailability of the consumer group or Kafka Streams application.

Note
 Upgrading to the incremental cooperative rebalance protocol is optional. The eager rebalance protocol is still supported.
Prerequisites
Procedure

To upgrade a Kafka consumer to use the incremental cooperative rebalance protocol:

  1. Replace the Kafka clients .jar file with the new version.

  2. In the consumer configuration, append cooperative-sticky to the partition.assignment.strategy. For example, if the range strategy is set, change the configuration to range, cooperative-sticky.

  3. Restart each consumer in the group in turn, waiting for the consumer to rejoin the group after each restart.

  4. Reconfigure each consumer in the group by removing the earlier partition.assignment.strategy from the consumer configuration, leaving only the cooperative-sticky strategy.

  5. Restart each consumer in the group in turn, waiting for the consumer to rejoin the group after each restart.

To upgrade a Kafka Streams application to use the incremental cooperative rebalance protocol:

  1. Replace the Kafka Streams .jar file with the new version.

  2. In the Kafka Streams configuration, set the upgrade.from configuration parameter to the Kafka version you are upgrading from (for example, 2.3).

  3. Restart each of the stream processors (nodes) in turn.

  4. Remove the upgrade.from configuration parameter from the Kafka Streams configuration.

  5. Restart each consumer in the group in turn.

13. Downgrading Strimzi

If you are encountering issues with the version of Strimzi you upgraded to, you can revert your installation to the previous version.

If you used the YAML installation files to install Strimzi, you can use the YAML installation files from the previous release to perform the following downgrade procedures:

  1. Downgrading the Cluster Operator to a previous version

  2. Downgrading Kafka

If the previous version of Strimzi does not support the version of Kafka you are using, you can also downgrade Kafka as long as the log message format versions appended to messages match.

Warning
 If you deployed Strimzi using another installation method, use a supported approach to downgrade Strimzi. Do not use the downgrade instructions provided here. For example, if you installed Strimzi using the Operator Lifecycle Manager (OLM), you can downgrade by changing the deployment channel to an earlier version of Strimzi.

13.1. Downgrading the Cluster Operator to a previous version

If you are encountering issues with Strimzi, you can revert your installation.

This procedure describes how to downgrade a Cluster Operator deployment to a previous version.

Prerequisites
Procedure

  1. Take note of any configuration changes made to the existing Cluster Operator resources (in the /install/cluster-operator directory). Any changes will be overwritten by the previous version of the Cluster Operator.

  2. Revert your custom resources to reflect the supported configuration options available for the version of Strimzi you are downgrading to.

  3. Update the Cluster Operator.

    1. Modify the installation files for the previous version according to the namespace the Cluster Operator is running in.

      On Linux, use:

      sed -i 's/namespace: .*/namespace: my-cluster-operator-namespace/' install/cluster-operator/*RoleBinding*.yaml

      On MacOS, use:

      sed -i '' 's/namespace: .*/namespace: my-cluster-operator-namespace/' install/cluster-operator/*RoleBinding*.yaml

    2. If you modified one or more environment variables in your existing Cluster Operator Deployment, edit the install/cluster-operator/060-Deployment-strimzi-cluster-operator.yaml file to use those environment variables.

  4. When you have an updated configuration, deploy it along with the rest of the installation resources:

    kubectl replace -f install/cluster-operator

    Wait for the rolling updates to complete.

  5. Get the image for the Kafka pod to ensure the downgrade was successful:

    kubectl get pod my-cluster-kafka-0 -o jsonpath='{.spec.containers[0].image}'

    The image tag shows the new Strimzi version followed by the Kafka version. For example, NEW-STRIMZI-VERSION-kafka-CURRENT-KAFKA-VERSION.

Your Cluster Operator was downgraded to the previous version.

13.2. Downgrading Kafka

Kafka version downgrades are performed by the Cluster Operator.

13.2.1. Kafka version compatibility for downgrades

Kafka downgrades are dependent on compatible current and target Kafka versions, and the state at which messages have been logged.

You cannot revert to the previous Kafka version if that version does not support any of the inter.broker.protocol.version settings which have ever been used in that cluster, or messages have been added to message logs that use a newer log.message.format.version.

The inter.broker.protocol.version determines the schemas used for persistent metadata stored by the broker, such as the schema for messages written to __consumer_offsets. If you downgrade to a version of Kafka that does not understand an inter.broker.protocol.version that has ever been previously used in the cluster the broker will encounter data it cannot understand.

If the target downgrade version of Kafka has:

  • The same log.message.format.version as the current version, the Cluster Operator downgrades by performing a single rolling restart of the brokers.

  • A different log.message.format.version, downgrading is only possible if the running cluster has always had log.message.format.version set to the version used by the downgraded version. This is typically only the case if the upgrade procedure was aborted before the log.message.format.version was changed. In this case, the downgrade requires:

    • Two rolling restarts of the brokers if the interbroker protocol of the two versions is different

    • A single rolling restart if they are the same

Downgrading is not possible if the new version has ever used a log.message.format.version that is not supported by the previous version, including when the default value for log.message.format.version is used. For example, this resource can be downgraded to Kafka version 3.2.3 because the log.message.format.version has not been changed:

apiVersion: kafka.strimzi.io/v1beta2
kind: Kafka
spec:
  # ...
  kafka:
    version: 3.3.2
    config:
      log.message.format.version: "3.2"
      # ...

The downgrade would not be possible if the log.message.format.version was set at "3.3" or a value was absent, so that the parameter took the default value for a 3.3.2 broker of 3.3.

Important
 From Kafka 3.0.0, when the inter.broker.protocol.version is set to 3.0 or higher, the log.message.format.version option is ignored and doesn’t need to be set.

13.2.2. Downgrading Kafka brokers and client applications

Downgrade a Strimzi Kafka cluster to a lower (previous) version of Kafka, such as downgrading from 3.3.2 to 3.2.3.

Prerequisites

  • The Cluster Operator is up and running.

  • Before you downgrade the Strimzi Kafka cluster, check the following for the Kafka resource:

    • IMPORTANT: Compatibility of Kafka versions.

    • Kafka.spec.kafka.config does not contain options that are not supported by the Kafka version being downgraded to.

    • Kafka.spec.kafka.config has a log.message.format.version and inter.broker.protocol.version that is supported by the Kafka version being downgraded to.

      From Kafka 3.0.0, when the inter.broker.protocol.version is set to 3.0 or higher, the log.message.format.version option is ignored and doesn’t need to be set.

Procedure

  1. Update the Kafka cluster configuration.

    kubectl edit kafka KAFKA-CONFIGURATION-FILE

  2. Change the Kafka.spec.kafka.version to specify the previous version.

    For example, if downgrading from Kafka 3.3.2 to 3.2.3:

    apiVersion: kafka.strimzi.io/v1beta2
kind: Kafka
spec:
  # ...
  kafka:
    version: 3.2.3 (1)
    config:
      log.message.format.version: "3.2" (2)
      inter.broker.protocol.version: "3.2" (3)
      # ...

    1. Kafka version is changed to the previous version.

    2. Message format version is unchanged.

    3. Inter-broker protocol version is unchanged.

    Note
    		 The value of log.message.format.version and inter.broker.protocol.version must be strings to prevent them from being interpreted as floating point numbers.

  3. If the image for the Kafka version is different from the image defined in STRIMZI_KAFKA_IMAGES for the Cluster Operator, update Kafka.spec.kafka.image.

    See Kafka version and image mappings

  4. Save and exit the editor, then wait for rolling updates to complete.

    Check the update in the logs or by watching the pod state transitions:

    kubectl logs -f CLUSTER-OPERATOR-POD-NAME | grep -E "Kafka version downgrade from [0-9.]+ to [0-9.]+, phase ([0-9]+) of \1 completed"
    kubectl get pod -w

    Check the Cluster Operator logs for an INFO level message:

    Reconciliation #NUM(watch) Kafka(NAMESPACE/NAME): Kafka version downgrade from FROM-VERSION to TO-VERSION, phase 1 of 1 completed

  5. Downgrade all client applications (consumers) to use the previous version of the client binaries.

    The Kafka cluster and clients are now using the previous Kafka version.

  6. If you are reverting back to a version of Strimzi earlier than 0.22, which uses ZooKeeper for the storage of topic metadata, delete the internal topic store topics from the Kafka cluster.

    kubectl run kafka-admin -ti --image=quay.io/strimzi/kafka:latest-kafka-3.3.2 --rm=true --restart=Never -- ./bin/kafka-topics.sh --bootstrap-server localhost:9092 --topic __strimzi-topic-operator-kstreams-topic-store-changelog --delete && ./bin/kafka-topics.sh --bootstrap-server localhost:9092 --topic __strimzi_store_topic --delete
Additional resources

14. Finding information on Kafka restarts

After the Cluster Operator restarts a Kafka pod in a Kubernetes cluster, it emits a Kubernetes event into the pod’s namespace explaining why the pod restarted. For help in understanding cluster behavior, you can check restart events from the command line.

Tip
 You can export and monitor restart events using metrics collection tools like Prometheus. Use the metrics tool with an event exporter that can export the output in a suitable format.

14.1. Reasons for a restart event

The Cluster Operator initiates a restart event for a specific reason. You can check the reason by fetching information on the restart event.

The reason given depends on whether you are using StrimziPodSet or StatefulSet resources for the creation and management of pods.

Table 9. Restart reasons
StrimziPodSet StatefulSet Description

CaCertHasOldGeneration

CaCertHasOldGeneration

The pod is still using a server certificate signed with an old CA, so needs to be restarted as part of the certificate update.

CaCertRemoved

CaCertRemoved

Expired CA certificates have been removed, and the pod is restarted to run with the current certificates.

CaCertRenewed

CaCertRenewed

CA certificates have been renewed, and the pod is restarted to run with the updated certificates.

ClientCaCertKeyReplaced

ClientCaCertKeyReplaced

The key used to sign clients CA certificates has been replaced, and the pod is being restarted as part of the CA renewal process.

ClusterCaCertKeyReplaced

ClusterCaCertKeyReplaced

The key used to sign the cluster’s CA certificates has been replaced, and the pod is being restarted as part of the CA renewal process.

ConfigChangeRequiresRestart

ConfigChangeRequiresRestart

Some Kafka configuration properties are changed dynamically, but others require that the broker be restarted.

CustomListenerCaCertChanged

CustomListenerCaCertChanged

The CA certificate used to secure the Kafka network listeners has changed, and the pod is restarted to use it.

FileSystemResizeNeeded

FileSystemResizeNeeded

The file system size has been increased, and a restart is needed to apply it.

KafkaCertificatesChanged

KafkaCertificatesChanged

One or more TLS certificates used by the Kafka broker have been updated, and a restart is needed to use them.

ManualRollingUpdate

ManualRollingUpdate

A user annotated the pod, or the StatefulSet` or StrimziPodSet set it belongs to, to trigger a restart.

PodForceRestartOnError

PodForceRestartOnError

An error occurred that requires a pod restart to rectify.

PodHasOldRevision

JbodVolumesChanged

A disk was added or removed from the Kafka volumes, and a restart is needed to apply the change. When using StrimziPodSet resources, the same reason is given if the pod needs to be recreated.

PodHasOldRevision

PodHasOldGeneration

The StatefulSet or StrimziPodSet that the pod is a member of has been updated, so the pod needs to be recreated. When using StrimziPodSet resources, the same reason is given if a disk was added or removed from the Kafka volumes.

PodStuck

PodStuck

The pod is still pending, and is not scheduled or cannot be scheduled, so the operator has restarted the pod in a final attempt to get it running.

PodUnresponsive

PodUnresponsive

Strimzi was unable to connect to the pod, which can indicate a broker not starting correctly, so the operator restarted it in an attempt to resolve the issue.

14.2. Restart event filters

When checking restart events from the command line, you can specify a field-selector to filter on Kubernetes event fields.

The following fields are available when filtering events with field-selector.

regardingObject.kind

The object that was restarted, and for restart events, the kind is always Pod.

regarding.namespace

The namespace that the pod belongs to.

regardingObject.name

The pod’s name, for example, strimzi-cluster-kafka-0.

regardingObject.uid

The unique ID of the pod.

reason

The reason the pod was restarted, for example, JbodVolumesChanged.

reportingController

The reporting component is always strimzi.io/cluster-operator for Strimzi restart events.

source

source is an older version of reportingController. The reporting component is always strimzi.io/cluster-operator for Strimzi restart events.

type

The event type, which is either Warning or Normal. For Strimzi restart events, the type is Normal.

Note
 In older versions of Kubernetes, the fields using the regarding prefix might use an involvedObject prefix instead. reportingController was previously called reportingComponent.

14.3. Checking Kafka restarts

Use a kubectl command to list restart events initiated by the Cluster Operator. Filter restart events emitted by the Cluster Operator by setting the Cluster Operator as the reporting component using the reportingController or source event fields.

Prerequisites

  • The Cluster Operator is running in the Kubernetes cluster.

Procedure

  1. Get all restart events emitted by the Cluster Operator:

    kubectl -n kafka get events --field-selector reportingController=strimzi.io/cluster-operator
    Example showing events returned
    LAST SEEN   TYPE     REASON                   OBJECT                        MESSAGE
2m          Normal   CaCertRenewed            pod/strimzi-cluster-kafka-0   CA certificate renewed
58m         Normal   PodForceRestartOnError   pod/strimzi-cluster-kafka-1   Pod needs to be forcibly restarted due to an error
5m47s       Normal   ManualRollingUpdate      pod/strimzi-cluster-kafka-2   Pod was manually annotated to be rolled

    You can also specify a reason or other field-selector options to constrain the events returned.

    Here, a specific reason is added:

    kubectl -n kafka get events --field-selector reportingController=strimzi.io/cluster-operator,reason=PodForceRestartOnError

  2. Use an output format, such as YAML, to return more detailed information about one or more events.

    kubectl -n kafka get events --field-selector reportingController=strimzi.io/cluster-operator,reason=PodForceRestartOnError -o yaml
    Example showing detailed events output
    apiVersion: v1
items:
- action: StrimziInitiatedPodRestart
  apiVersion: v1
  eventTime: "2022-05-13T00:22:34.168086Z"
  firstTimestamp: null
  involvedObject:
      kind: Pod
      name: strimzi-cluster-kafka-1
      namespace: kafka
  kind: Event
  lastTimestamp: null
  message: Pod needs to be forcibly restarted due to an error
  metadata:
      creationTimestamp: "2022-05-13T00:22:34Z"
      generateName: strimzi-event
      name: strimzi-eventwppk6
      namespace: kafka
      resourceVersion: "432961"
      uid: 29fcdb9e-f2cf-4c95-a165-a5efcd48edfc
  reason: PodForceRestartOnError
  reportingController: strimzi.io/cluster-operator
  reportingInstance: strimzi-cluster-operator-6458cfb4c6-6bpdp
  source: {}
  type: Normal
kind: List
metadata:
  resourceVersion: ""
  selfLink: ""

The following fields are deprecated, so they are not populated for these events:

  • firstTimestamp

  • lastTimestamp

  • source

15. Tuning Kafka configuration

Use configuration properties to optimize the performance of Kafka brokers, producers and consumers.

A minimum set of configuration properties is required, but you can add or adjust properties to change how producers and consumers interact with Kafka brokers. For example, you can tune latency and throughput of messages so that clients can respond to data in real time.

You might start by analyzing metrics to gauge where to make your initial configurations, then make incremental changes and further comparisons of metrics until you have the configuration you need.

For more information about Apache Kafka configuration properties, see the Apache Kafka documentation.

15.1. Tools that help with tuning

The following tools help with Kafka tuning:

  • Cruise Control generates optimization proposals that you can use to assess and implement a cluster rebalance

  • Kafka Static Quota plugin sets limits on brokers

  • Rack configuration spreads broker partitions across racks and allows consumers to fetch data from the nearest replica

15.2. Managed broker configurations

When you deploy Strimzi on Kubernetes, you can specify broker configuration through the config property of the Kafka custom resource. However, certain broker configuration options are managed directly by Strimzi.

As such, you cannot configure the following options:

  • broker.id to specify the ID of the Kafka broker

  • log.dirs directories for log data

  • zookeeper.connect configuration to connect Kafka with ZooKeeper

  • listeners to expose the Kafka cluster to clients

  • authorization mechanisms to allow or decline actions executed by users

  • authentication mechanisms to prove the identity of users requiring access to Kafka

Broker IDs start from 0 (zero) and correspond to the number of broker replicas. Log directories are mounted to /var/lib/kafka/data/kafka-logIDX based on the spec.kafka.storage configuration in the Kafka custom resource. IDX is the Kafka broker pod index.

For a list of exclusions, see the KafkaClusterSpec schema reference.

15.3. Kafka broker configuration tuning

Use configuration properties to optimize the performance of Kafka brokers. You can use standard Kafka broker configuration options, except for properties managed directly by Strimzi.

15.3.1. Basic broker configuration

A typical broker configuration will include settings for properties related to topics, threads and logs.

Basic broker configuration properties
# ...
num.partitions=1
default.replication.factor=3
offsets.topic.replication.factor=3
transaction.state.log.replication.factor=3
transaction.state.log.min.isr=2
log.retention.hours=168
log.segment.bytes=1073741824
log.retention.check.interval.ms=300000
num.network.threads=3
num.io.threads=8
num.recovery.threads.per.data.dir=1
socket.send.buffer.bytes=102400
socket.receive.buffer.bytes=102400
socket.request.max.bytes=104857600
group.initial.rebalance.delay.ms=0
zookeeper.connection.timeout.ms=6000
# ...

15.3.2. Replicating topics for high availability

Basic topic properties set the default number of partitions and replication factor for topics, which will apply to topics that are created without these properties being explicitly set, including when topics are created automatically.

# ...
num.partitions=1
auto.create.topics.enable=false
default.replication.factor=3
min.insync.replicas=2
replica.fetch.max.bytes=1048576
# ...

For high availability environments, it is advisable to increase the replication factor to at least 3 for topics and set the minimum number of in-sync replicas required to 1 less than the replication factor.

The auto.create.topics.enable property is enabled by default so that topics that do not already exist are created automatically when needed by producers and consumers. If you are using automatic topic creation, you can set the default number of partitions for topics using num.partitions. Generally, however, this property is disabled so that more control is provided over topics through explicit topic creation.

For data durability, you should also set min.insync.replicas in your topic configuration and message delivery acknowledgments using acks=all in your producer configuration.

Use replica.fetch.max.bytes to set the maximum size, in bytes, of messages fetched by each follower that replicates the leader partition. Change this value according to the average message size and throughput. When considering the total memory allocation required for read/write buffering, the memory available must also be able to accommodate the maximum replicated message size when multiplied by all followers.

The delete.topic.enable property is enabled by default to allow topics to be deleted. In a production environment, you should disable this property to avoid accidental topic deletion, resulting in data loss. You can, however, temporarily enable it and delete topics and then disable it again.

Note
 When running Strimzi on Kubernetes, the Topic Operator can provide operator-style topic management. You can use the KafkaTopic resource to create topics. For topics created using the KafkaTopic resource, the replication factor is set using spec.replicas. If delete.topic.enable is enabled, you can also delete topics using the KafkaTopic resource. 
# ...
auto.create.topics.enable=false
delete.topic.enable=true
# ...

15.3.3. Internal topic settings for transactions and commits

If you are using transactions to enable atomic writes to partitions from producers, the state of the transactions is stored in the internal __transaction_state topic. By default, the brokers are configured with a replication factor of 3 and a minimum of 2 in-sync replicas for this topic, which means that a minimum of three brokers are required in your Kafka cluster.

# ...
transaction.state.log.replication.factor=3
transaction.state.log.min.isr=2
# ...

Similarly, the internal __consumer_offsets topic, which stores consumer state, has default settings for the number of partitions and replication factor.

# ...
offsets.topic.num.partitions=50
offsets.topic.replication.factor=3
# ...

Do not reduce these settings in production. You can increase the settings in a production environment. As an exception, you might want to reduce the settings in a single-broker test environment.

15.3.4. Improving request handling throughput by increasing I/O threads

Network threads handle requests to the Kafka cluster, such as produce and fetch requests from client applications. Produce requests are placed in a request queue. Responses are placed in a response queue.

The number of network threads per listener should reflect the replication factor and the levels of activity from client producers and consumers interacting with the Kafka cluster. If you are going to have a lot of requests, you can increase the number of threads, using the amount of time threads are idle to determine when to add more threads.

To reduce congestion and regulate the request traffic, you can limit the number of requests allowed in the request queue. When the request queue is full, all incoming traffic is blocked.

I/O threads pick up requests from the request queue to process them. Adding more threads can improve throughput, but the number of CPU cores and disk bandwidth imposes a practical upper limit. At a minimum, the number of I/O threads should equal the number of storage volumes.

# ...
num.network.threads=3 (1)
queued.max.requests=500 (2)
num.io.threads=8 (3)
num.recovery.threads.per.data.dir=4 (4)
# ...

  1. The number of network threads for the Kafka cluster.

  2. The number of requests allowed in the request queue.

  3. The number of I/O threads for a Kafka broker.

  4. The number of threads used for log loading at startup and flushing at shutdown. Try setting to a value of at least the number of cores.

Configuration updates to the thread pools for all brokers might occur dynamically at the cluster level. These updates are restricted to between half the current size and twice the current size.

Tip

The following Kafka broker metrics can help with working out the number of threads required:

  • kafka.network:type=SocketServer,name=NetworkProcessorAvgIdlePercent provides metrics on the average time network threads are idle as a percentage.

  • kafka.server:type=KafkaRequestHandlerPool,name=RequestHandlerAvgIdlePercent provides metrics on the average time I/O threads are idle as a percentage.

If there is 0% idle time, all resources are in use, which means that adding more threads might be beneficial. When idle time goes below 30%, performance may start to suffer.

If threads are slow or limited due to the number of disks, you can try increasing the size of the buffers for network requests to improve throughput:

# ...
replica.socket.receive.buffer.bytes=65536
# ...

And also increase the maximum number of bytes Kafka can receive:

# ...
socket.request.max.bytes=104857600
# ...

15.3.5. Increasing bandwidth for high latency connections

Kafka batches data to achieve reasonable throughput over high-latency connections from Kafka to clients, such as connections between datacenters. However, if high latency is a problem, you can increase the size of the buffers for sending and receiving messages.

# ...
socket.send.buffer.bytes=1048576
socket.receive.buffer.bytes=1048576
# ...

You can estimate the optimal size of your buffers using a bandwidth-delay product calculation, which multiplies the maximum bandwidth of the link (in bytes/s) with the round-trip delay (in seconds) to give an estimate of how large a buffer is required to sustain maximum throughput.

15.3.6. Managing logs with data retention policies

Kafka uses logs to store message data. Logs are a series of segments associated with various indexes. New messages are written to an active segment, and never subsequently modified. Segments are read when serving fetch requests from consumers. Periodically, the active segment is rolled to become read-only and a new active segment is created to replace it. There is only a single segment active at a time. Older segments are retained until they are eligible for deletion.

Configuration at the broker level sets the maximum size in bytes of a log segment and the amount of time in milliseconds before an active segment is rolled:

# ...
log.segment.bytes=1073741824
log.roll.ms=604800000
# ...

You can override these settings at the topic level using segment.bytes and segment.ms. Whether you need to lower or raise these values depends on the policy for segment deletion. A larger size means the active segment contains more messages and is rolled less often. Segments also become eligible for deletion less often.

You can set time-based or size-based log retention and cleanup policies so that logs are kept manageable. Depending on your requirements, you can use log retention configuration to delete old segments. If log retention policies are used, non-active log segments are removed when retention limits are reached. Deleting old segments bounds the storage space required for the log so you do not exceed disk capacity.

For time-based log retention, you set a retention period based on hours, minutes and milliseconds. The retention period is based on the time messages were appended to the segment.

The milliseconds configuration has priority over minutes, which has priority over hours. The minutes and milliseconds configuration is null by default, but the three options provide a substantial level of control over the data you wish to retain. Preference should be given to the milliseconds configuration, as it is the only one of the three properties that is dynamically updateable.

# ...
log.retention.ms=1680000
# ...

If log.retention.ms is set to -1, no time limit is applied to log retention, so all logs are retained. Disk usage should always be monitored, but the -1 setting is not generally recommended as it can lead to issues with full disks, which can be hard to rectify.

For size-based log retention, you set a maximum log size (of all segments in the log) in bytes:

# ...
log.retention.bytes=1073741824
# ...

In other words, a log will typically have approximately log.retention.bytes/log.segment.bytes segments once it reaches a steady state. When the maximum log size is reached, older segments are removed.

A potential issue with using a maximum log size is that it does not take into account the time messages were appended to a segment. You can use time-based and size-based log retention for your cleanup policy to get the balance you need. Whichever threshold is reached first triggers the cleanup.

If you wish to add a time delay before a segment file is deleted from the system, you can add the delay using log.segment.delete.delay.ms for all topics at the broker level or file.delete.delay.ms for specific topics in the topic configuration.

# ...
log.segment.delete.delay.ms=60000
# ...

15.3.7. Removing log data with cleanup policies

The method of removing older log data is determined by the log cleaner configuration.

The log cleaner is enabled for the broker by default:

# ...
log.cleaner.enable=true
# ...

The log cleaner needs to be enabled if you are using log compaction cleanup policy. You can set the cleanup policy at the topic or broker level. Broker-level configuration is the default for topics that do not have policy set.

You can set policy to delete logs, compact logs, or do both:

# ...
log.cleanup.policy=compact,delete
# ...

The delete policy corresponds to managing logs with data retention policies. It is suitable when data does not need to be retained forever. The compact policy guarantees to keep the most recent message for each message key. Log compaction is suitable where message values are changeable, and you want to retain the latest update.

If cleanup policy is set to delete logs, older segments are deleted based on log retention limits. Otherwise, if the log cleaner is not enabled, and there are no log retention limits, the log will continue to grow.

If cleanup policy is set for log compaction, the head of the log operates as a standard Kafka log, with writes for new messages appended in order. In the tail of a compacted log, where the log cleaner operates, records will be deleted if another record with the same key occurs later in the log. Messages with null values are also deleted. If you’re not using keys, you can’t use compaction because keys are needed to identify related messages. While Kafka guarantees that the latest messages for each key will be retained, it does not guarantee that the whole compacted log will not contain duplicates.

Image of compaction showing key value writes
Figure 3. Log showing key value writes with offset positions before compaction

Using keys to identify messages, Kafka compaction keeps the latest message (with the highest offset) for a specific message key, eventually discarding earlier messages that have the same key. In other words, the message in its latest state is always available and any out-of-date records of that particular message are eventually removed when the log cleaner runs. You can restore a message back to a previous state.

Records retain their original offsets even when surrounding records get deleted. Consequently, the tail can have non-contiguous offsets. When consuming an offset that’s no longer available in the tail, the record with the next higher offset is found.

Image of compaction after log cleanup
Figure 4. Log after compaction

If you choose only a compact policy, your log can still become arbitrarily large. In which case, you can set policy to compact and delete logs. If you choose to compact and delete, first the log data is compacted, removing records with a key in the head of the log. After which, data that falls before the log retention threshold is deleted.

Image of compaction with retention point
Figure 5. Log retention point and compaction point

You set the frequency the log is checked for cleanup in milliseconds:

# ...
log.retention.check.interval.ms=300000
# ...

Adjust the log retention check interval in relation to the log retention settings. Smaller retention sizes might require more frequent checks.

The frequency of cleanup should be often enough to manage the disk space, but not so often it affects performance on a topic.

You can also set a time in milliseconds to put the cleaner on standby if there are no logs to clean:

# ...
log.cleaner.backoff.ms=15000
# ...

If you choose to delete older log data, you can set a period in milliseconds to retain the deleted data before it is purged:

# ...
log.cleaner.delete.retention.ms=86400000
# ...

The deleted data retention period gives time to notice the data is gone before it is irretrievably deleted.

To delete all messages related to a specific key, a producer can send a tombstone message. A tombstone has a null value and acts as a marker to tell a consumer the value is deleted. After compaction, only the tombstone is retained, which must be for a long enough period for the consumer to know that the message is deleted. When older messages are deleted, having no value, the tombstone key is also deleted from the partition.

15.3.8. Managing disk utilization

There are many other configuration settings related to log cleanup, but of particular importance is memory allocation.

The deduplication property specifies the total memory for cleanup across all log cleaner threads. You can set an upper limit on the percentage of memory used through the buffer load factor.

# ...
log.cleaner.dedupe.buffer.size=134217728
log.cleaner.io.buffer.load.factor=0.9
# ...

Each log entry uses exactly 24 bytes, so you can work out how many log entries the buffer can handle in a single run and adjust the setting accordingly.

If possible, consider increasing the number of log cleaner threads if you are looking to reduce the log cleaning time:

# ...
log.cleaner.threads=8
# ...

If you are experiencing issues with 100% disk bandwidth usage, you can throttle the log cleaner I/O so that the sum of the read/write operations is less than a specified double value based on the capabilities of the disks performing the operations:

# ...
log.cleaner.io.max.bytes.per.second=1.7976931348623157E308
# ...

15.3.9. Handling large message sizes

The default batch size for messages is 1MB, which is optimal for maximum throughput in most use cases. Kafka can accommodate larger batches at a reduced throughput, assuming adequate disk capacity.

Large message sizes are handled in four ways:

  1. Producer-side message compression writes compressed messages to the log.

  2. Reference-based messaging sends only a reference to data stored in some other system in the message’s value.

  3. Inline messaging splits messages into chunks that use the same key, which are then combined on output using a stream-processor like Kafka Streams.

  4. Broker and producer/consumer client application configuration built to handle larger message sizes.

The reference-based messaging and message compression options are recommended and cover most situations. With any of these options, care must be take to avoid introducing performance issues.

Producer-side compression

For producer configuration, you specify a compression.type, such as Gzip, which is then applied to batches of data generated by the producer. Using the broker configuration compression.type=producer, the broker retains whatever compression the producer used. Whenever producer and topic compression do not match, the broker has to compress batches again prior to appending them to the log, which impacts broker performance.

Compression also adds additional processing overhead on the producer and decompression overhead on the consumer, but includes more data in a batch, so is often beneficial to throughput when message data compresses well.

Combine producer-side compression with fine-tuning of the batch size to facilitate optimum throughput. Using metrics helps to gauge the average batch size needed.

Reference-based messaging

Reference-based messaging is useful for data replication when you do not know how big a message will be. The external data store must be fast, durable, and highly available for this configuration to work. Data is written to the data store and a reference to the data is returned. The producer sends a message containing the reference to Kafka. The consumer gets the reference from the message and uses it to fetch the data from the data store.

Image of reference-based messaging flow
Figure 6. Reference-based messaging flow

As the message passing requires more trips, end-to-end latency will increase. Another significant drawback of this approach is there is no automatic clean up of the data in the external system when the Kafka message gets cleaned up. A hybrid approach would be to only send large messages to the data store and process standard-sized messages directly.

Inline messaging

Inline messaging is complex, but it does not have the overhead of depending on external systems like reference-based messaging.

The producing client application has to serialize and then chunk the data if the message is too big. The producer then uses the Kafka ByteArraySerializer or similar to serialize each chunk again before sending it. The consumer tracks messages and buffers chunks until it has a complete message. The consuming client application receives the chunks, which are assembled before deserialization. Complete messages are delivered to the rest of the consuming application in order according to the offset of the first or last chunk for each set of chunked messages. Successful delivery of the complete message is checked against offset metadata to avoid duplicates during a rebalance.

Image of inline messaging flow
Figure 7. Inline messaging flow

Inline messaging has a performance overhead on the consumer side because of the buffering required, particularly when handling a series of large messages in parallel. The chunks of large messages can become interleaved, so that it is not always possible to commit when all the chunks of a message have been consumed if the chunks of another large message in the buffer are incomplete. For this reason, the buffering is usually supported by persisting message chunks or by implementing commit logic.

Configuration to handle larger messages

If larger messages cannot be avoided, and to avoid blocks at any point of the message flow, you can increase message limits. To do this, configure message.max.bytes at the topic level to set the maximum record batch size for individual topics. If you set message.max.bytes at the broker level, larger messages are allowed for all topics.

The broker will reject any message that is greater than the limit set with message.max.bytes. The buffer size for the producers (max.request.size) and consumers (message.max.bytes) must be able to accommodate the larger messages.

15.3.10. Controlling the log flush of message data

Generally, the recommendation is to not set explicit flush thresholds and let the operating system perform background flush using its default settings. Partition replication provides greater data durability than writes to any single disk, as a failed broker can recover from its in-sync replicas.

Log flush properties control the periodic writes of cached message data to disk. The scheduler specifies the frequency of checks on the log cache in milliseconds:

# ...
log.flush.scheduler.interval.ms=2000
# ...

You can control the frequency of the flush based on the maximum amount of time that a message is kept in-memory and the maximum number of messages in the log before writing to disk:

# ...
log.flush.interval.ms=50000
log.flush.interval.messages=100000
# ...

The wait between flushes includes the time to make the check and the specified interval before the flush is carried out. Increasing the frequency of flushes can affect throughput.

If you are using application flush management, setting lower flush thresholds might be appropriate if you are using faster disks.

15.3.11. Partition rebalancing for availability

Partitions can be replicated across brokers for fault tolerance. For a given partition, one broker is elected leader and handles all produce requests (writes to the log). Partition followers on other brokers replicate the partition data of the partition leader for data reliability in the event of the leader failing.

Followers do not normally serve clients, though rack configuration allows a consumer to consume messages from the closest replica when a Kafka cluster spans multiple datacenters. Followers operate only to replicate messages from the partition leader and allow recovery should the leader fail. Recovery requires an in-sync follower. Followers stay in sync by sending fetch requests to the leader, which returns messages to the follower in order. The follower is considered to be in sync if it has caught up with the most recently committed message on the leader. The leader checks this by looking at the last offset requested by the follower. An out-of-sync follower is usually not eligible as a leader should the current leader fail, unless unclean leader election is allowed.

You can adjust the lag time before a follower is considered out of sync:

# ...
replica.lag.time.max.ms=30000
# ...

Lag time puts an upper limit on the time to replicate a message to all in-sync replicas and how long a producer has to wait for an acknowledgment. If a follower fails to make a fetch request and catch up with the latest message within the specified lag time, it is removed from in-sync replicas. You can reduce the lag time to detect failed replicas sooner, but by doing so you might increase the number of followers that fall out of sync needlessly. The right lag time value depends on both network latency and broker disk bandwidth.

When a leader partition is no longer available, one of the in-sync replicas is chosen as the new leader. The first broker in a partition’s list of replicas is known as the preferred leader. By default, Kafka is enabled for automatic partition leader rebalancing based on a periodic check of leader distribution. That is, Kafka checks to see if the preferred leader is the current leader. A rebalance ensures that leaders are evenly distributed across brokers and brokers are not overloaded.

You can use Cruise Control for Strimzi to figure out replica assignments to brokers that balance load evenly across the cluster. Its calculation takes into account the differing load experienced by leaders and followers. A failed leader affects the balance of a Kafka cluster because the remaining brokers get the extra work of leading additional partitions.

For the assignment found by Cruise Control to actually be balanced it is necessary that partitions are lead by the preferred leader. Kafka can automatically ensure that the preferred leader is being used (where possible), changing the current leader if necessary. This ensures that the cluster remains in the balanced state found by Cruise Control.

You can control the frequency, in seconds, of the rebalance check and the maximum percentage of imbalance allowed for a broker before a rebalance is triggered.

#...
auto.leader.rebalance.enable=true
leader.imbalance.check.interval.seconds=300
leader.imbalance.per.broker.percentage=10
#...

The percentage leader imbalance for a broker is the ratio between the current number of partitions for which the broker is the current leader and the number of partitions for which it is the preferred leader. You can set the percentage to zero to ensure that preferred leaders are always elected, assuming they are in sync.

If the checks for rebalances need more control, you can disable automated rebalances. You can then choose when to trigger a rebalance using the kafka-leader-election.sh command line tool.

Note
 The Grafana dashboards provided with Strimzi show metrics for under-replicated partitions and partitions that do not have an active leader.

15.3.12. Unclean leader election

Leader election to an in-sync replica is considered clean because it guarantees no loss of data. And this is what happens by default. But what if there is no in-sync replica to take on leadership? Perhaps the ISR (in-sync replica) only contained the leader when the leader’s disk died. If a minimum number of in-sync replicas is not set, and there are no followers in sync with the partition leader when its hard drive fails irrevocably, data is already lost. Not only that, but a new leader cannot be elected because there are no in-sync followers.

You can configure how Kafka handles leader failure:

# ...
unclean.leader.election.enable=false
# ...

Unclean leader election is disabled by default, which means that out-of-sync replicas cannot become leaders. With clean leader election, if no other broker was in the ISR when the old leader was lost, Kafka waits until that leader is back online before messages can be written or read. Unclean leader election means out-of-sync replicas can become leaders, but you risk losing messages. The choice you make depends on whether your requirements favor availability or durability.

You can override the default configuration for specific topics at the topic level. If you cannot afford the risk of data loss, then leave the default configuration.

15.3.13. Avoiding unnecessary consumer group rebalances

For consumers joining a new consumer group, you can add a delay so that unnecessary rebalances to the broker are avoided:

# ...
group.initial.rebalance.delay.ms=3000
# ...

The delay is the amount of time that the coordinator waits for members to join. The longer the delay, the more likely it is that all the members will join in time and avoid a rebalance. But the delay also prevents the group from consuming until the period has ended.

15.4. Kafka producer configuration tuning

Use a basic producer configuration with optional properties that are tailored to specific use cases.

Adjusting your configuration to maximize throughput might increase latency or vice versa. You will need to experiment and tune your producer configuration to get the balance you need.

15.4.1. Basic producer configuration

Connection and serializer properties are required for every producer. Generally, it is good practice to add a client id for tracking, and use compression on the producer to reduce batch sizes in requests.

In a basic producer configuration:

  • The order of messages in a partition is not guaranteed.

  • The acknowledgment of messages reaching the broker does not guarantee durability.

Basic producer configuration properties
# ...
bootstrap.servers=localhost:9092 (1)
key.serializer=org.apache.kafka.common.serialization.StringSerializer (2)
value.serializer=org.apache.kafka.common.serialization.StringSerializer (3)
client.id=my-client (4)
compression.type=gzip (5)
# ...

  1. (Required) Tells the producer to connect to a Kafka cluster using a host:port bootstrap server address for a Kafka broker. The producer uses the address to discover and connect to all brokers in the cluster. Use a comma-separated list to specify two or three addresses in case a server is down, but it’s not necessary to provide a list of all the brokers in the cluster.

  2. (Required) Serializer to transform the key of each message to bytes prior to them being sent to a broker.

  3. (Required) Serializer to transform the value of each message to bytes prior to them being sent to a broker.

  4. (Optional) The logical name for the client, which is used in logs and metrics to identify the source of a request.

  5. (Optional) The codec for compressing messages, which are sent and might be stored in compressed format and then decompressed when reaching a consumer. Compression is useful for improving throughput and reducing the load on storage, but might not be suitable for low latency applications where the cost of compression or decompression could be prohibitive.

15.4.2. Data durability

Message delivery acknowledgments minimize the likelihood that messages are lost. Acknowledgments are enabled by default with the acks property set at acks=all.

Acknowledging message delivery
# ...
acks=all (1)
# ...

  1. acks=all forces a leader replica to replicate messages to a certain number of followers before acknowledging that the message request was successfully received.

The acks=all setting offers the strongest guarantee of delivery, but it will increase the latency between the producer sending a message and receiving acknowledgment. If you don’t require such strong guarantees, a setting of acks=0 or acks=1 provides either no delivery guarantees or only acknowledgment that the leader replica has written the record to its log.

With acks=all, the leader waits for all in-sync replicas to acknowledge message delivery. A topic’s min.insync.replicas configuration sets the minimum required number of in-sync replica acknowledgements. The number of acknowledgements include that of the leader and followers.

A typical starting point is to use the following configuration:

  • Producer configuration:

    • acks=all (default)

  • Broker configuration for topic replication:

    • default.replication.factor=3 (default = 1)

    • min.insync.replicas=2 (default = 1)

When you create a topic, you can override the default replication factor. You can also override min.insync.replicas at the topic level in the topic configuration.

Strimzi uses this configuration in the example configuration files for multi-node deployment of Kafka.

The following table describes how this configuration operates depending on the availability of followers that replicate the leader replica.

Table 10. Follower availability
Number of followers available and in-sync Acknowledgements Producer can send messages?

2

The leader waits for 2 follower acknowledgements

Yes

1

The leader waits for 1 follower acknowledgement

Yes

0

The leader raises an exception

No

A topic replication factor of 3 creates one leader replica and two followers. In this configuration, the producer can continue if a single follower is unavailable. Some delay can occur whilst removing a failed broker from the in-sync replicas or a creating a new leader. If the second follower is also unavailable, message delivery will not be successful. Instead of acknowledging successful message delivery, the leader sends an error (not enough replicas) to the producer. The producer raises an equivalent exception. With retries configuration, the producer can resend the failed message request.

Note
 If the system fails, there is a risk of unsent data in the buffer being lost.

15.4.3. Ordered delivery

Idempotent producers avoid duplicates as messages are delivered exactly once. IDs and sequence numbers are assigned to messages to ensure the order of delivery, even in the event of failure. If you are using acks=all for data consistency, using idempotency makes sense for ordered delivery. Idempotency is enabled for producers by default. With idempotency enabled, you can set the number of concurrent in-flight requests to a maximum of 5 for message ordering to be preserved.

Ordered delivery with idempotency
# ...
enable.idempotence=true (1)
max.in.flight.requests.per.connection=5 (2)
acks=all (3)
retries=2147483647 (4)
# ...

  1. Set to true to enable the idempotent producer.

  2. With idempotent delivery the number of in-flight requests may be greater than 1 while still providing the message ordering guarantee. The default is 5 in-flight requests.

  3. Set acks to all.

  4. Set the number of attempts to resend a failed message request.

If you choose not to use acks=all and disable idempotency because of the performance cost, set the number of in-flight (unacknowledged) requests to 1 to preserve ordering. Otherwise, a situation is possible where Message-A fails only to succeed after Message-B was already written to the broker.

Ordered delivery without idempotency
# ...
enable.idempotence=false (1)
max.in.flight.requests.per.connection=1 (2)
retries=2147483647
# ...

  1. Set to false to disable the idempotent producer.

  2. Set the number of in-flight requests to exactly 1.

15.4.4. Reliability guarantees

Idempotence is useful for exactly once writes to a single partition. Transactions, when used with idempotence, allow exactly once writes across multiple partitions.

Transactions guarantee that messages using the same transactional ID are produced once, and either all are successfully written to the respective logs or none of them are.

# ...
enable.idempotence=true
max.in.flight.requests.per.connection=5
acks=all
retries=2147483647
transactional.id=UNIQUE-ID (1)
transaction.timeout.ms=900000 (2)
# ...

  1. Specify a unique transactional ID.

  2. Set the maximum allowed time for transactions in milliseconds before a timeout error is returned. The default is 900000 or 15 minutes.

The choice of transactional.id is important in order that the transactional guarantee is maintained. Each transactional id should be used for a unique set of topic partitions. For example, this can be achieved using an external mapping of topic partition names to transactional ids, or by computing the transactional id from the topic partition names using a function that avoids collisions.

15.4.5. Optimizing producers for throughput and latency

Usually, the requirement of a system is to satisfy a particular throughput target for a proportion of messages within a given latency. For example, targeting 500,000 messages per second with 95% of messages being acknowledged within 2 seconds.

It’s likely that the messaging semantics (message ordering and durability) of your producer are defined by the requirements for your application. For instance, it’s possible that you don’t have the option of using acks=0 or acks=1 without breaking some important property or guarantee provided by your application.

Broker restarts have a significant impact on high percentile statistics. For example, over a long period the 99th percentile latency is dominated by behavior around broker restarts. This is worth considering when designing benchmarks or comparing performance numbers from benchmarking with performance numbers seen in production.

Depending on your objective, Kafka offers a number of configuration parameters and techniques for tuning producer performance for throughput and latency.

Message batching (linger.ms and batch.size)

Message batching delays sending messages in the hope that more messages destined for the same broker will be sent, allowing them to be batched into a single produce request. Batching is a compromise between higher latency in return for higher throughput. Time-based batching is configured using linger.ms, and size-based batching is configured using batch.size.

Compression (compression.type)

Message compression adds latency in the producer (CPU time spent compressing the messages), but makes requests (and potentially disk writes) smaller, which can increase throughput. Whether compression is worthwhile, and the best compression to use, will depend on the messages being sent. Compression happens on the thread which calls KafkaProducer.send(), so if the latency of this method matters for your application you should consider using more threads.

Pipelining (max.in.flight.requests.per.connection)

Pipelining means sending more requests before the response to a previous request has been received. In general more pipelining means better throughput, up to a threshold at which other effects, such as worse batching, start to counteract the effect on throughput.

Lowering latency

When your application calls KafkaProducer.send() the messages are:

  • Processed by any interceptors

  • Serialized

  • Assigned to a partition

  • Compressed

  • Added to a batch of messages in a per-partition queue

At which point the send() method returns. So the time send() is blocked is determined by:

  • The time spent in the interceptors, serializers and partitioner

  • The compression algorithm used

  • The time spent waiting for a buffer to use for compression

Batches will remain in the queue until one of the following occurs:

  • The batch is full (according to batch.size)

  • The delay introduced by linger.ms has passed

  • The sender is about to send message batches for other partitions to the same broker, and it is possible to add this batch too

  • The producer is being flushed or closed

Look at the configuration for batching and buffering to mitigate the impact of send() blocking on latency.

# ...
linger.ms=100 (1)
batch.size=16384 (2)
buffer.memory=33554432 (3)
# ...

  1. The linger property adds a delay in milliseconds so that larger batches of messages are accumulated and sent in a request. The default is 0'.

  2. If a maximum batch.size in bytes is used, a request is sent when the maximum is reached, or messages have been queued for longer than linger.ms (whichever comes sooner). Adding the delay allows batches to accumulate messages up to the batch size.

  3. The buffer size must be at least as big as the batch size, and be able to accommodate buffering, compression and in-flight requests.

Increasing throughput

Improve throughput of your message requests by adjusting the maximum time to wait before a message is delivered and completes a send request.

You can also direct messages to a specified partition by writing a custom partitioner to replace the default.

# ...
delivery.timeout.ms=120000 (1)
partitioner.class=my-custom-partitioner (2)

# ...

  1. The maximum time in milliseconds to wait for a complete send request. You can set the value to MAX_LONG to delegate to Kafka an indefinite number of retries. The default is 120000 or 2 minutes.

  2. Specify the class name of the custom partitioner.

15.5. Kafka consumer configuration tuning

Use a basic consumer configuration with optional properties that are tailored to specific use cases.

When tuning your consumers your primary concern will be ensuring that they cope efficiently with the amount of data ingested. As with the producer tuning, be prepared to make incremental changes until the consumers operate as expected.

15.5.1. Basic consumer configuration

Connection and deserializer properties are required for every consumer. Generally, it is good practice to add a client id for tracking.

In a consumer configuration, irrespective of any subsequent configuration:

  • The consumer fetches from a given offset and consumes the messages in order, unless the offset is changed to skip or re-read messages.

  • The broker does not know if the consumer processed the responses, even when committing offsets to Kafka, because the offsets might be sent to a different broker in the cluster.

Basic consumer configuration properties
# ...
bootstrap.servers=localhost:9092 (1)
key.deserializer=org.apache.kafka.common.serialization.StringDeserializer  (2)
value.deserializer=org.apache.kafka.common.serialization.StringDeserializer  (3)
client.id=my-client (4)
group.id=my-group-id (5)
# ...

  1. (Required) Tells the consumer to connect to a Kafka cluster using a host:port bootstrap server address for a Kafka broker. The consumer uses the address to discover and connect to all brokers in the cluster. Use a comma-separated list to specify two or three addresses in case a server is down, but it is not necessary to provide a list of all the brokers in the cluster. If you are using a loadbalancer service to expose the Kafka cluster, you only need the address for the service because the availability is handled by the loadbalancer.

  2. (Required) Deserializer to transform the bytes fetched from the Kafka broker into message keys.

  3. (Required) Deserializer to transform the bytes fetched from the Kafka broker into message values.

  4. (Optional) The logical name for the client, which is used in logs and metrics to identify the source of a request. The id can also be used to throttle consumers based on processing time quotas.

  5. (Conditional) A group id is required for a consumer to be able to join a consumer group.

15.5.2. Scaling data consumption using consumer groups

Consumer groups share a typically large data stream generated by one or multiple producers from a given topic. Consumers are grouped using a group.id property, allowing messages to be spread across the members. One of the consumers in the group is elected leader and decides how the partitions are assigned to the consumers in the group. Each partition can only be assigned to a single consumer.

If you do not already have as many consumers as partitions, you can scale data consumption by adding more consumer instances with the same group.id. Adding more consumers to a group than there are partitions will not help throughput, but it does mean that there are consumers on standby should one stop functioning. If you can meet throughput goals with fewer consumers, you save on resources.

Consumers within the same consumer group send offset commits and heartbeats to the same broker. So the greater the number of consumers in the group, the higher the request load on the broker.

# ...
group.id=my-group-id (1)
# ...

  1. Add a consumer to a consumer group using a group id.

15.5.3. Message ordering guarantees

Kafka brokers receive fetch requests from consumers that ask the broker to send messages from a list of topics, partitions and offset positions.

A consumer observes messages in a single partition in the same order that they were committed to the broker, which means that Kafka only provides ordering guarantees for messages in a single partition. Conversely, if a consumer is consuming messages from multiple partitions, the order of messages in different partitions as observed by the consumer does not necessarily reflect the order in which they were sent.

If you want a strict ordering of messages from one topic, use one partition per consumer.

15.5.4. Optimizing consumers for throughput and latency

Control the number of messages returned when your client application calls KafkaConsumer.poll().

Use the fetch.max.wait.ms and fetch.min.bytes properties to increase the minimum amount of data fetched by the consumer from the Kafka broker. Time-based batching is configured using fetch.max.wait.ms, and size-based batching is configured using fetch.min.bytes.

If CPU utilization in the consumer or broker is high, it might be because there are too many requests from the consumer. You can adjust fetch.max.wait.ms and fetch.min.bytes properties higher so that there are fewer requests and messages are delivered in bigger batches. By adjusting higher, throughput is improved with some cost to latency. You can also adjust higher if the amount of data being produced is low.

For example, if you set fetch.max.wait.ms to 500ms and fetch.min.bytes to 16384 bytes, when Kafka receives a fetch request from the consumer it will respond when the first of either threshold is reached.

Conversely, you can adjust the fetch.max.wait.ms and fetch.min.bytes properties lower to improve end-to-end latency.

# ...
fetch.max.wait.ms=500 (1)
fetch.min.bytes=16384 (2)
# ...

  1. The maximum time in milliseconds the broker will wait before completing fetch requests. The default is 500 milliseconds.

  2. If a minimum batch size in bytes is used, a request is sent when the minimum is reached, or messages have been queued for longer than fetch.max.wait.ms (whichever comes sooner). Adding the delay allows batches to accumulate messages up to the batch size.

Lowering latency by increasing the fetch request size

Use the fetch.max.bytes and max.partition.fetch.bytes properties to increase the maximum amount of data fetched by the consumer from the Kafka broker.

The fetch.max.bytes property sets a maximum limit in bytes on the amount of data fetched from the broker at one time.

The max.partition.fetch.bytes sets a maximum limit in bytes on how much data is returned for each partition, which must always be larger than the number of bytes set in the broker or topic configuration for max.message.bytes.

The maximum amount of memory a client can consume is calculated approximately as:

NUMBER-OF-BROKERS * fetch.max.bytes and NUMBER-OF-PARTITIONS * max.partition.fetch.bytes

If memory usage can accommodate it, you can increase the values of these two properties. By allowing more data in each request, latency is improved as there are fewer fetch requests.

# ...
fetch.max.bytes=52428800 (1)
max.partition.fetch.bytes=1048576 (2)
# ...

  1. The maximum amount of data in bytes returned for a fetch request.

  2. The maximum amount of data in bytes returned for each partition.

15.5.5. Avoiding data loss or duplication when committing offsets

The Kafka auto-commit mechanism allows a consumer to commit the offsets of messages automatically. If enabled, the consumer will commit offsets received from polling the broker at 5000ms intervals.

The auto-commit mechanism is convenient, but it introduces a risk of data loss and duplication. If a consumer has fetched and transformed a number of messages, but the system crashes with processed messages in the consumer buffer when performing an auto-commit, that data is lost. If the system crashes after processing the messages, but before performing the auto-commit, the data is duplicated on another consumer instance after rebalancing.

Auto-committing can avoid data loss only when all messages are processed before the next poll to the broker, or the consumer closes.

To minimize the likelihood of data loss or duplication, you can set enable.auto.commit to false and develop your client application to have more control over committing offsets. Or you can use auto.commit.interval.ms to decrease the intervals between commits.

# ...
enable.auto.commit=false (1)
# ...

  1. Auto commit is set to false to provide more control over committing offsets.

By setting to enable.auto.commit to false, you can commit offsets after all processing has been performed and the message has been consumed. For example, you can set up your application to call the Kafka commitSync and commitAsync commit APIs.

The commitSync API commits the offsets in a message batch returned from polling. You call the API when you are finished processing all the messages in the batch. If you use the commitSync API, the application will not poll for new messages until the last offset in the batch is committed. If this negatively affects throughput, you can commit less frequently, or you can use the commitAsync API. The commitAsync API does not wait for the broker to respond to a commit request, but risks creating more duplicates when rebalancing. A common approach is to combine both commit APIs in an application, with the commitSync API used just before shutting the consumer down or rebalancing to make sure the final commit is successful.

Controlling transactional messages

Consider using transactional ids and enabling idempotence (enable.idempotence=true) on the producer side to guarantee exactly-once delivery. On the consumer side, you can then use the isolation.level property to control how transactional messages are read by the consumer.

The isolation.level property has two valid values:

  • read_committed

  • read_uncommitted (default)

Use read_committed to ensure that only transactional messages that have been committed are read by the consumer. However, this will cause an increase in end-to-end latency, because the consumer will not be able to return a message until the brokers have written the transaction markers that record the result of the transaction (committed or aborted).

# ...
enable.auto.commit=false
isolation.level=read_committed (1)
# ...

  1. Set to read_committed so that only committed messages are read by the consumer.

15.5.6. Recovering from failure to avoid data loss

Use the session.timeout.ms and heartbeat.interval.ms properties to configure the time taken to check and recover from consumer failure within a consumer group.

The session.timeout.ms property specifies the maximum amount of time in milliseconds a consumer within a consumer group can be out of contact with a broker before being considered inactive and a rebalancing is triggered between the active consumers in the group. When the group rebalances, the partitions are reassigned to the members of the group.

The heartbeat.interval.ms property specifies the interval in milliseconds between heartbeat checks to the consumer group coordinator to indicate that the consumer is active and connected. The heartbeat interval must be lower, usually by a third, than the session timeout interval.

If you set the session.timeout.ms property lower, failing consumers are detected earlier, and rebalancing can take place quicker. However, take care not to set the timeout so low that the broker fails to receive a heartbeat in time and triggers an unnecessary rebalance.

Decreasing the heartbeat interval reduces the chance of accidental rebalancing, but more frequent heartbeats increases the overhead on broker resources.

15.5.7. Managing offset policy

Use the auto.offset.reset property to control how a consumer behaves when no offsets have been committed, or a committed offset is no longer valid or deleted.

Suppose you deploy a consumer application for the first time, and it reads messages from an existing topic. Because this is the first time the group.id is used, the __consumer_offsets topic does not contain any offset information for this application. The new application can start processing all existing messages from the start of the log or only new messages. The default reset value is latest, which starts at the end of the partition, and consequently means some messages are missed. To avoid data loss, but increase the amount of processing, set auto.offset.reset to earliest to start at the beginning of the partition.

Also consider using the earliest option to avoid messages being lost when the offsets retention period (offsets.retention.minutes) configured for a broker has ended. If a consumer group or standalone consumer is inactive and commits no offsets during the retention period, previously committed offsets are deleted from __consumer_offsets.

# ...
heartbeat.interval.ms=3000 (1)
session.timeout.ms=45000 (2)
auto.offset.reset=earliest (3)
# ...

  1. Adjust the heartbeat interval lower according to anticipated rebalances.

  2. If no heartbeats are received by the Kafka broker before the timeout duration expires, the consumer is removed from the consumer group and a rebalance is initiated. If the broker configuration has a group.min.session.timeout.ms and group.max.session.timeout.ms, the session timeout value must be within that range.

  3. Set to earliest to return to the start of a partition and avoid data loss if offsets were not committed.

If the amount of data returned in a single fetch request is large, a timeout might occur before the consumer has processed it. In this case, you can lower max.partition.fetch.bytes or increase session.timeout.ms.

15.5.8. Minimizing the impact of rebalances

The rebalancing of a partition between active consumers in a group is the time it takes for:

  • Consumers to commit their offsets

  • The new consumer group to be formed

  • The group leader to assign partitions to group members

  • The consumers in the group to receive their assignments and start fetching

Clearly, the process increases the downtime of a service, particularly when it happens repeatedly during a rolling restart of a consumer group cluster.

In this situation, you can use the concept of static membership to reduce the number of rebalances. Rebalancing assigns topic partitions evenly among consumer group members. Static membership uses persistence so that a consumer instance is recognized during a restart after a session timeout.

The consumer group coordinator can identify a new consumer instance using a unique id that is specified using the group.instance.id property. During a restart, the consumer is assigned a new member id, but as a static member it continues with the same instance id, and the same assignment of topic partitions is made.

If the consumer application does not make a call to poll at least every max.poll.interval.ms milliseconds, the consumer is considered to be failed, causing a rebalance. If the application cannot process all the records returned from poll in time, you can avoid a rebalance by using the max.poll.interval.ms property to specify the interval in milliseconds between polls for new messages from a consumer. Or you can use the max.poll.records property to set a maximum limit on the number of records returned from the consumer buffer, allowing your application to process fewer records within the max.poll.interval.ms limit.

# ...
group.instance.id=UNIQUE-ID (1)
max.poll.interval.ms=300000 (2)
max.poll.records=500 (3)
# ...

  1. The unique instance id ensures that a new consumer instance receives the same assignment of topic partitions.

  2. Set the interval to check the consumer is continuing to process messages.

  3. Sets the number of processed records returned from the consumer.

15.6. Handling high volumes of messages

If your Strimzi deployment needs to handle a high volume of messages, you can use configuration options to optimize for throughput and latency.

Producer and consumer configuration can help control the size and frequency of requests to Kafka brokers. For more information on the configuration options, see the following:

You can also use the same configuration options with the producers and consumers used by the Kafka Connect runtime source connectors (including MirrorMaker 2.0) and sink connectors.

Source connectors

  • Producers from the Kafka Connect runtime send messages to the Kafka cluster.

  • For MirrorMaker 2.0, since the source system is Kafka, consumers retrieve messages from a source Kafka cluster.

Sink connectors

  • Consumers from the Kafka Connect runtime retrieve messages from the Kafka cluster.

For consumers, you might increase the amount of data fetched in a single fetch request to reduce latency. You increase the fetch request size using the fetch.max.bytes and max.partition.fetch.bytes properties. You can also set a maximum limit on the number of messages returned from the consumer buffer using the max.poll.records property.

For MirrorMaker 2.0, configure the fetch.max.bytes, max.partition.fetch.bytes, and max.poll.records values at the source connector level (consumer.*), as they relate to the specific consumer that fetches messages from the source.

For producers, you might increase the size of the message batches sent in a single produce request. You increase the batch size using the batch.size property. A larger batch size reduces the number of outstanding messages ready to be sent and the size of the backlog in the message queue. Messages being sent to the same partition are batched together. A produce request is sent to the target cluster when the batch size is reached. By increasing the batch size, produce requests are delayed and more messages are added to the batch and sent to brokers at the same time. This can improve throughput when you have just a few topic partitions that handle large numbers of messages.

Consider the number and size of the records that the producer handles for a suitable producer batch size.

Use linger.ms to add a wait time in milliseconds to delay produce requests when producer load decreases. The delay means that more records can be added to batches if they are under the maximum batch size.

Configure the batch.size and linger.ms values at the source connector level (producer.override.*), as they relate to the specific producer that sends messages to the target Kafka cluster.

For Kafka Connect source connectors, the data streaming pipeline to the target Kafka cluster is as follows:

Data streaming pipeline for Kafka Connect source connector

external data source → (Kafka Connect tasks) source message queue → producer buffer → target Kafka topic

For Kafka Connect sink connectors, the data streaming pipeline to the target external data source is as follows:

Data streaming pipeline for Kafka Connect sink connector

source Kafka topic → (Kafka Connect tasks) sink message queue → consumer buffer → external data source

For MirrorMaker 2.0, the data mirroring pipeline to the target Kafka cluster is as follows:

Data mirroring pipeline for MirrorMaker 2.0

source Kafka topic → (Kafka Connect tasks) source message queue → producer buffer → target Kafka topic

The producer sends messages in its buffer to topics in the target Kafka cluster. While this is happening, Kafka Connect tasks continue to poll the data source to add messages to the source message queue.

The size of the producer buffer for the source connector is set using the producer.override.buffer.memory property. Tasks wait for a specified timeout period (offset.flush.timeout.ms) before the buffer is flushed. This should be enough time for the sent messages to be acknowledged by the brokers and offset data committed. The source task does not wait for the producer to empty the message queue before committing offsets, except during shutdown.

If the producer is unable to keep up with the throughput of messages in the source message queue, buffering is blocked until there is space available in the buffer within a time period bounded by max.block.ms. Any unacknowledged messages still in the buffer are sent during this period. New messages are not added to the buffer until these messages are acknowledged and flushed.

You can try the following configuration changes to keep the underlying source message queue of outstanding messages at a manageable size:

  • Increasing the default value in milliseconds of the offset.flush.timeout.ms

  • Ensuring that there are enough CPU and memory resources

  • Increasing the number of tasks that run in parallel by doing the following:

    • Increasing the number of tasks that run in parallel using the tasksMax property

    • Increasing the number of worker nodes that run tasks using the replicas property

Consider the number of tasks that can run in parallel according to the available CPU and memory resources and number of worker nodes. You might need to keep adjusting the configuration values until they have the desired effect.

15.6.1. Configuring Kafka Connect for high-volume messages

Kafka Connect fetches data from the source external data system and hands it to the Kafka Connect runtime producers so that it’s replicated to the target cluster.

The following example shows configuration for Kafka Connect using the KafkaConnect custom resource.

Example Kafka Connect configuration for handling high volumes of messages
apiVersion: kafka.strimzi.io/v1beta2
kind: KafkaConnect
metadata:
  name: my-connect-cluster
  annotations:
    strimzi.io/use-connector-resources: "true"
spec:
  replicas: 3
  config:
    offset.flush.timeout.ms: 10000
    # ...
  resources:
    requests:
      cpu: "1"
      memory: 2Gi
    limits:
      cpu: "2"
      memory: 2Gi
  # ...

Producer configuration is added for the source connector, which is managed using the KafkaConnector custom resource.

Example source connector configuration for handling high volumes of messages
apiVersion: kafka.strimzi.io/v1beta2
kind: KafkaConnector
metadata:
  name: my-source-connector
  labels:
    strimzi.io/cluster: my-connect-cluster
spec:
  class: org.apache.kafka.connect.file.FileStreamSourceConnector
  tasksMax: 2
  config:
    producer.override.batch.size: 327680
    producer.override.linger.ms: 100
    # ...
Note
 FileStreamSourceConnector and FileStreamSinkConnector are provided as example connectors. For information on deploying them as KafkaConnector resources, see Deploying example KafkaConnector resources.

Consumer configuration is added for the sink connector.

Example sink connector configuration for handling high volumes of messages
apiVersion: kafka.strimzi.io/v1beta2
kind: KafkaConnector
metadata:
  name: my-sink-connector
  labels:
    strimzi.io/cluster: my-connect-cluster
spec:
  class: org.apache.kafka.connect.file.FileStreamSinkConnector
  tasksMax: 2
  config:
    consumer.fetch.max.bytes: 52428800
    consumer.max.partition.fetch.bytes: 1048576
    consumer.max.poll.records: 500
    # ...

If you are using the Kafka Connect API instead of the KafkaConnector custom resource to manage your connectors, you can add the connector configuration as a JSON object.

Example curl request to add source connector configuration for handling high volumes of messages
curl -X POST \
  http://my-connect-cluster-connect-api:8083/connectors \
  -H 'Content-Type: application/json' \
  -d '{ "name": "my-source-connector",
    "config":
    {
      "connector.class":"org.apache.kafka.connect.file.FileStreamSourceConnector",
      "file": "/opt/kafka/LICENSE",
      "topic":"my-topic",
      "tasksMax": "4",
      "type": "source"
      "producer.override.batch.size": 327680
      "producer.override.linger.ms": 100
    }
}'

15.6.2. Configuring MirrorMaker 2.0 for high-volume messages

MirrorMaker 2.0 fetches data from the source cluster and hands it to the Kafka Connect runtime producers so that it’s replicated to the target cluster.

The following example shows the configuration for MirrorMaker 2.0 using the KafkaMirrorMaker2 custom resource.

Example MirrorMaker 2.0 configuration for handling high volumes of messages
apiVersion: kafka.strimzi.io/v1beta2
kind: KafkaMirrorMaker2
metadata:
  name: my-mirror-maker2
spec:
  version: 3.3.2
  replicas: 1
  connectCluster: "my-cluster-target"
  clusters:
  - alias: "my-cluster-source"
    bootstrapServers: my-cluster-source-kafka-bootstrap:9092
  - alias: "my-cluster-target"
    config:
      offset.flush.timeout.ms: 10000
    bootstrapServers: my-cluster-target-kafka-bootstrap:9092
  mirrors:
  - sourceCluster: "my-cluster-source"
    targetCluster: "my-cluster-target"
    sourceConnector:
      tasksMax: 2
      config:
        producer.override.batch.size: 327680
        producer.override.linger.ms: 100
        consumer.fetch.max.bytes: 52428800
        consumer.max.partition.fetch.bytes: 1048576
        consumer.max.poll.records: 500
    # ...
  resources:
    requests:
      cpu: "1"
      memory: Gi
    limits:
      cpu: "2"
      memory: 4Gi

15.6.3. Checking the MirrorMaker 2.0 message flow

If you are using Prometheus and Grafana to monitor your deployment, you can check the MirrorMaker 2.0 message flow.

The example MirrorMaker 2.0 Grafana dashboards provided with Strimzi show the following metrics related to the flush pipeline.

  • The number of messages in Kafka Connect’s outstanding messages queue

  • The available bytes of the producer buffer

  • The offset commit timeout in milliseconds

You can use these metrics to gauge whether or not you need to tune your configuration based on the volume of messages.

Additional resources

16. Uninstalling Strimzi

You can uninstall Strimzi using the CLI or by unsubscribing from OperatorHub.io.

Use the same approach you used to install Strimzi.

When you uninstall Strimzi, you will need to identify resources created specifically for a deployment and referenced from the Strimzi resource.

Such resources include:

  • Secrets (Custom CAs and certificates, Kafka Connect secrets, and other Kafka secrets)

  • Logging ConfigMaps (of type external)

These are resources referenced by Kafka, KafkaConnect, KafkaMirrorMaker, or KafkaBridge configuration.

Warning
 Deleting CustomResourceDefinitions results in the garbage collection of the corresponding custom resources (Kafka, KafkaConnect, KafkaMirrorMaker, or KafkaBridge) and the resources dependent on them (Deployments, StatefulSets, and other dependent resources).

16.1. Uninstalling Strimzi using the CLI

This procedure describes how to use the kubectl command-line tool to uninstall Strimzi and remove resources related to the deployment.

Prerequisites

  • Access to a Kubernetes cluster using an account with cluster-admin or strimzi-admin permissions.

  • You have identified the resources to be deleted.

    You can use the following kubectl CLI command to find resources and also verify that they have been removed when you have uninstalled Strimzi.

    Command to find resources related to a Strimzi deployment
    kubectl get <resource_type> --all-namespaces | grep <kafka_cluster_name>

    Replace <resource_type> with the type of the resource you are checking, such as secret or configmap.

Procedure

  1. Delete the Cluster Operator Deployment, related CustomResourceDefinitions, and RBAC resources.

    Specify the installation files used to deploy the Cluster Operator.

    kubectl delete -f install/cluster-operator

  2. Delete the resources you identified in the prerequisites.

    kubectl delete <resource_type> <resource_name> -n <namespace>

    Replace <resource_type> with the type of resource you are deleting and <resource_name> with the name of the resource.

    Example to delete a secret
    kubectl delete secret my-cluster-clients-ca -n my-project

16.2. Uninstalling Strimzi from OperatorHub.io

This procedure describes how to uninstall Strimzi from OperatorHub.io and remove resources related to the deployment.

You perform the steps using the kubectl command-line tool.

Prerequisites

  • Access to a Kubernetes cluster using an account with cluster-admin or strimzi-admin permissions.

  • You have identified the resources to be deleted.

    You can use the following kubectl CLI command to find resources and also verify that they have been removed when you have uninstalled Strimzi.

    Command to find resources related to a Strimzi deployment
    kubectl get <resource_type> --all-namespaces | grep <kafka_cluster_name>

    Replace <resource_type> with the type of the resource you are checking, such as secret or configmap.

Procedure

  1. Delete the Strimzi subscription.

    kubectl delete subscription strimzi-cluster-operator -n <namespace>

  2. Delete the cluster service version (CSV).

    kubectl delete csv strimzi-cluster-operator.<version>  -n <namespace>

  3. Remove related CRDs.

    kubectl get crd -l app=strimzi -o name | xargs kubectl delete