Chapter 12. Managing AMQ Streams

This chapter covers tasks to maintain a deployment of AMQ Streams.

12.1. Working with custom resources

You can use oc commands to retrieve information and perform other operations on AMQ Streams custom resources.

Using oc with the status subresource of a custom resource allows you to get the information about the resource.

12.1.1. Performing oc operations on custom resources

Use oc commands, such as get, describe, edit, or delete, to perform operations on resource types. For example, oc get kafkatopics retrieves a list of all Kafka topics and oc get kafkas retrieves all deployed Kafka clusters.

When referencing resource types, you can use both singular and plural names: oc get kafkas gets the same results as oc get kafka.

You can also use the short name of the resource. Learning short names can save you time when managing AMQ Streams. The short name for Kafka is k, so you can also run oc get k to list all Kafka clusters.

oc get k

NAME         DESIRED KAFKA REPLICAS   DESIRED ZK REPLICAS
my-cluster   3                        3

Table 12.1. Long and short names for each AMQ Streams resource

AMQ Streams resourceLong nameShort name

Kafka

kafka

k

Kafka Topic

kafkatopic

kt

Kafka User

kafkauser

ku

Kafka Connect

kafkaconnect

kc

Kafka Connect S2I

kafkaconnects2i

kcs2i

Kafka Connector

kafkaconnector

kctr

Kafka Mirror Maker

kafkamirrormaker

kmm

Kafka Mirror Maker 2

kafkamirrormaker2

kmm2

Kafka Bridge

kafkabridge

kb

Kafka Rebalance

kafkarebalance

kr

12.1.1.1. Resource categories

Categories of custom resources can also be used in oc commands.

All AMQ Streams custom resources belong to the category strimzi, so you can use strimzi to get all the AMQ Streams resources with one command.

For example, running oc get strimzi lists all AMQ Streams custom resources in a given namespace.

oc get strimzi

NAME                                   DESIRED KAFKA REPLICAS DESIRED ZK REPLICAS
kafka.kafka.strimzi.io/my-cluster      3                      3

NAME                                   PARTITIONS REPLICATION FACTOR
kafkatopic.kafka.strimzi.io/kafka-apps 3          3

NAME                                   AUTHENTICATION AUTHORIZATION
kafkauser.kafka.strimzi.io/my-user     tls            simple

The oc get strimzi -o name command returns all resource types and resource names. The -o name option fetches the output in the type/name format

oc get strimzi -o name

kafka.kafka.strimzi.io/my-cluster
kafkatopic.kafka.strimzi.io/kafka-apps
kafkauser.kafka.strimzi.io/my-user

You can combine this strimzi command with other commands. For example, you can pass it into a oc delete command to delete all resources in a single command.

oc delete $(oc get strimzi -o name)

kafka.kafka.strimzi.io "my-cluster" deleted
kafkatopic.kafka.strimzi.io "kafka-apps" deleted
kafkauser.kafka.strimzi.io "my-user" deleted

Deleting all resources in a single operation might be useful, for example, when you are testing new AMQ Streams features.

12.1.1.2. Querying the status of sub-resources

There are other values you can pass to the -o option. For example, by using -o yaml you get the output in YAML format. Usng -o json will return it as JSON.

You can see all the options in oc get --help.

One of the most useful options is the JSONPath support, which allows you to pass JSONPath expressions to query the Kubernetes API. A JSONPath expression can extract or navigate specific parts of any resource.

For example, you can use the JSONPath expression {.status.listeners[?(@.type=="tls")].bootstrapServers} to get the bootstrap address from the status of the Kafka custom resource and use it in your Kafka clients.

Here, the command finds the bootstrapServers value of the tls listeners.

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

my-cluster-kafka-bootstrap.myproject.svc:9093

By changing the type condition to @.type=="external" or @.type=="plain" you can also get the address of the other Kafka listeners.

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

192.168.1.247:9094

You can use jsonpath to extract any other property or group of properties from any custom resource.

12.1.2. AMQ Streams custom resource status information

Several resources have a status property, as described in the following table.

Table 12.2. Custom resource status properties

AMQ Streams resourceSchema referencePublishes status information on…​

Kafka

Section 13.2.76, “KafkaStatus schema reference”

The Kafka cluster.

KafkaConnect

Section 13.2.102, “KafkaConnectStatus schema reference”

The Kafka Connect cluster, if deployed.

KafkaConnectS2I

Section 13.2.106, “KafkaConnectS2IStatus schema reference”

The Kafka Connect cluster with Source-to-Image support, if deployed.

KafkaConnector

Section 13.2.141, “KafkaConnectorStatus schema reference”

KafkaConnector resources, if deployed.

KafkaMirrorMaker

Section 13.2.129, “KafkaMirrorMakerStatus schema reference”

The Kafka MirrorMaker tool, if deployed.

KafkaTopic

Section 13.2.109, “KafkaTopicStatus schema reference”

Kafka topics in your Kafka cluster.

KafkaUser

Section 13.2.122, “KafkaUserStatus schema reference”

Kafka users in your Kafka cluster.

KafkaBridge

Section 13.2.138, “KafkaBridgeStatus schema reference”

The AMQ Streams Kafka Bridge, if deployed.

The status property of a resource provides information on the resource’s:

  • Current state, in the status.conditions property
  • Last observed generation, in the status.observedGeneration property

The status property also provides resource-specific information. For example:

  • KafkaStatus provides information on listener addresses, and the id of the Kafka cluster.
  • KafkaConnectStatus provides the REST API endpoint for Kafka Connect connectors.
  • KafkaUserStatus provides the user name of the Kafka user and the Secret in which their credentials are stored.
  • KafkaBridgeStatus provides the HTTP address at which external client applications can access the Bridge service.

A resource’s current state is useful for tracking progress related to the resource achieving its desired state, as defined by the spec property. The status conditions provide the time and reason the state of the resource changed and details of events preventing or delaying the operator from realizing the resource’s desired state.

The last observed generation is the generation of the resource that was last reconciled by the Cluster Operator. If the value of observedGeneration is different from the value of metadata.generation, the operator has not yet processed the latest update to the resource. If these values are the same, the status information reflects the most recent changes to the resource.

AMQ Streams creates and maintains the status of custom resources, periodically evaluating the current state of the custom resource and updating its status accordingly. When performing an update on a custom resource using oc edit, for example, its status is not editable. Moreover, changing the status would not affect the configuration of the Kafka cluster.

Here we see the status property specified for a Kafka custom resource.

Kafka custom resource with status

apiVersion: kafka.strimzi.io/v1beta2
kind: Kafka
metadata:
spec:
  # ...
status:
  conditions: 1
  - lastTransitionTime: 2021-07-23T23:46:57+0000
    status: "True"
    type: Ready 2
  observedGeneration: 4 3
  listeners: 4
  - addresses:
    - host: my-cluster-kafka-bootstrap.myproject.svc
      port: 9092
    type: plain
  - addresses:
    - host: my-cluster-kafka-bootstrap.myproject.svc
      port: 9093
    certificates:
    - |
      -----BEGIN CERTIFICATE-----
      ...
      -----END CERTIFICATE-----
    type: tls
  - addresses:
    - host: 172.29.49.180
      port: 9094
    certificates:
    - |
      -----BEGIN CERTIFICATE-----
      ...
      -----END CERTIFICATE-----
    type: external
  clusterId: CLUSTER-ID 5
# ...

1
Status conditions describe criteria related to the status that cannot be deduced from the existing resource information, or are specific to the instance of a resource.
2
The Ready condition indicates whether the Cluster Operator currently considers the Kafka cluster able to handle traffic.
3
The observedGeneration indicates the generation of the Kafka custom resource that was last reconciled by the Cluster Operator.
4
The listeners describe the current Kafka bootstrap addresses by type.
5
The Kafka cluster id.
Important

The address in the custom resource status for external listeners with type nodeport is currently not supported.

Note

The Kafka bootstrap addresses listed in the status do not signify that those endpoints or the Kafka cluster is in a ready state.

Accessing status information

You can access status information for a resource from the command line. For more information, see Section 12.1.3, “Finding the status of a custom resource”.

12.1.3. Finding the status of a custom resource

This procedure describes how to find the status of a custom resource.

Prerequisites

  • An OpenShift cluster.
  • The Cluster Operator is running.

Procedure

  • Specify the custom resource and use the -o jsonpath option to apply a standard JSONPath expression to select the status property:

    oc get kafka <kafka_resource_name> -o jsonpath='{.status}'

    This expression returns all the status information for the specified custom resource. You can use dot notation, such as status.listeners or status.observedGeneration, to fine-tune the status information you wish to see.

Additional resources

12.2. Pausing reconciliation of custom resources

Sometimes it is useful to pause the reconciliation of custom resources managed by AMQ Streams Operators, so that you can perform fixes or make updates. If reconciliations are paused, any changes made to custom resources are ignored by the Operators until the pause ends.

If you want to pause reconciliation of a custom resource, set the strimzi.io/pause-reconciliation annotation to true in its configuration. This instructs the appropriate Operator to pause reconciliation of the custom resource. For example, you can apply the annotation to the KafkaConnect resource so that reconciliation by the Cluster Operator is paused.

You can also create a custom resource with the pause annotation enabled. The custom resource is created, but it is ignored.

Important

It is not currently possible to pause reconciliation of KafkaTopic resources.

Prerequisites

  • The AMQ Streams Operator that manages the custom resource is running.

Procedure

  1. Annotate the custom resource in OpenShift, setting pause-reconciliation to true:

    oc annotate KIND-OF-CUSTOM-RESOURCE NAME-OF-CUSTOM-RESOURCE strimzi.io/pause-reconciliation="true"

    For example, for the KafkaConnect custom resource:

    oc annotate KafkaConnect my-connect strimzi.io/pause-reconciliation="true"
  2. Check that the status conditions of the custom resource show a change to ReconciliationPaused:

    oc describe KIND-OF-CUSTOM-RESOURCE NAME-OF-CUSTOM-RESOURCE

    The type condition changes to ReconciliationPaused at the lastTransitionTime.

    Example custom resource with a paused reconciliation condition type

    apiVersion: kafka.strimzi.io/v1beta2
    kind: KafkaConnect
    metadata:
      annotations:
        strimzi.io/pause-reconciliation: "true"
        strimzi.io/use-connector-resources: "true"
      creationTimestamp: 2021-03-12T10:47:11Z
      #...
    spec:
      # ...
    status:
      conditions:
      - lastTransitionTime: 2021-03-12T10:47:41.689249Z
        status: "True"
        type: ReconciliationPaused

Resuming from pause

  • To resume reconciliation, you can set the annotation to false, or remove the annotation.

12.3. Manually starting rolling updates of Kafka and ZooKeeper clusters

AMQ Streams supports the use of annotations on StatefulSet and Pod resources to manually trigger a rolling update of Kafka and ZooKeeper clusters through the Cluster Operator. Rolling updates restart the pods of the resource with new ones.

Manually performing a rolling update on a specific pod or set of pods from the same StatefulSet is usually only required in exceptional circumstances. However, rather than deleting the pods directly, if you perform the rolling update through the Cluster Operator you ensure that:

  • The manual deletion of the pod does not conflict with simultaneous Cluster Operator operations, such as deleting other pods in parallel.
  • The Cluster Operator logic handles the Kafka configuration specifications, such as the number of in-sync replicas.

12.3.1. Prerequisites

To perform a manual rolling update, you need a running Cluster Operator and Kafka cluster.

See the Deploying and Upgrading AMQ Streams on OpenShift guide for instructions on running a:

12.3.2. Performing a rolling update using a StatefulSet annotation

This procedure describes how to manually trigger a rolling update of an existing Kafka cluster or ZooKeeper cluster using an OpenShift StatefulSet annotation.

Procedure

  1. Find the name of the StatefulSet that controls the Kafka or ZooKeeper pods you want to manually update.

    For example, if your Kafka cluster is named my-cluster, the corresponding StatefulSet names are my-cluster-kafka and my-cluster-zookeeper.

  2. Annotate the StatefulSet resource in OpenShift.

    Use oc annotate:

    oc annotate statefulset cluster-name-kafka strimzi.io/manual-rolling-update=true
    
    oc annotate statefulset cluster-name-zookeeper strimzi.io/manual-rolling-update=true
  3. Wait for the next reconciliation to occur (every two minutes by default). A rolling update of all pods within the annotated StatefulSet is triggered, as long as the annotation was detected by the reconciliation process. When the rolling update of all the pods is complete, the annotation is removed from the StatefulSet.

12.3.3. Performing a rolling update using a Pod annotation

This procedure describes how to manually trigger a rolling update of an existing Kafka cluster or ZooKeeper cluster using an OpenShift Pod annotation. When multiple pods from the same StatefulSet are annotated, consecutive rolling updates are performed within the same reconciliation run.

Procedure

  1. Find the name of the Kafka or ZooKeeper Pod you want to manually update.

    For example, if your Kafka cluster is named my-cluster, the corresponding Pod names are my-cluster-kafka-index and my-cluster-zookeeper-index. The index starts at zero and ends at the total number of replicas.

  2. Annotate the Pod resource in OpenShift.

    Use oc annotate:

    oc annotate pod cluster-name-kafka-index strimzi.io/manual-rolling-update=true
    
    oc annotate pod cluster-name-zookeeper-index strimzi.io/manual-rolling-update=true
  3. Wait for the next reconciliation to occur (every two minutes by default). A rolling update of the annotated Pod is triggered, as long as the annotation was detected by the reconciliation process. When the rolling update of a pod is complete, the annotation is removed from the Pod.

12.4. Discovering services using labels and annotations

Service discovery makes it easier for client applications running in the same OpenShift cluster as AMQ Streams to interact with a Kafka cluster.

A service discovery label and annotation is generated for services used to access the Kafka cluster:

  • Internal Kafka bootstrap service
  • HTTP Bridge service

The label helps to make the service discoverable, and the annotation provides connection details that a client application can use to make the connection.

The service discovery label, strimzi.io/discovery, is set as true for the Service resources. The service discovery annotation has the same key, providing connection details in JSON format for each service.

Example internal Kafka bootstrap service

apiVersion: v1
kind: Service
metadata:
  annotations:
    strimzi.io/discovery: |-
      [ {
        "port" : 9092,
        "tls" : false,
        "protocol" : "kafka",
        "auth" : "scram-sha-512"
      }, {
        "port" : 9093,
        "tls" : true,
        "protocol" : "kafka",
        "auth" : "tls"
      } ]
  labels:
    strimzi.io/cluster: my-cluster
    strimzi.io/discovery: "true"
    strimzi.io/kind: Kafka
    strimzi.io/name: my-cluster-kafka-bootstrap
  name: my-cluster-kafka-bootstrap
spec:
  #...

Example HTTP Bridge service

apiVersion: v1
kind: Service
metadata:
  annotations:
    strimzi.io/discovery: |-
      [ {
        "port" : 8080,
        "tls" : false,
        "auth" : "none",
        "protocol" : "http"
      } ]
  labels:
    strimzi.io/cluster: my-bridge
    strimzi.io/discovery: "true"
    strimzi.io/kind: KafkaBridge
    strimzi.io/name: my-bridge-bridge-service

12.4.1. Returning connection details on services

You can find the services by specifying the discovery label when fetching services from the command line or a corresponding API call.

oc get service -l strimzi.io/discovery=true

The connection details are returned when retrieving the service discovery label.

12.5. Recovering a cluster from persistent volumes

You can recover a Kafka cluster from persistent volumes (PVs) if they are still present.

You might want to do this, for example, after:

  • A namespace was deleted unintentionally
  • A whole OpenShift cluster is lost, but the PVs remain in the infrastructure

12.5.1. Recovery from namespace deletion

Recovery from namespace deletion is possible because of the relationship between persistent volumes and namespaces. A PersistentVolume (PV) is a storage resource that lives outside of a namespace. A PV is mounted into a Kafka pod using a PersistentVolumeClaim (PVC), which lives inside a namespace.

The reclaim policy for a PV tells a cluster how to act when a namespace is deleted. If the reclaim policy is set as:

  • Delete (default), PVs are deleted when PVCs are deleted within a namespace
  • Retain, PVs are not deleted when a namespace is deleted

To ensure that you can recover from a PV if a namespace is deleted unintentionally, the policy must be reset from Delete to Retain in the PV specification using the persistentVolumeReclaimPolicy property:

apiVersion: v1
kind: PersistentVolume
# ...
spec:
  # ...
  persistentVolumeReclaimPolicy: Retain

Alternatively, PVs can inherit the reclaim policy of an associated storage class. Storage classes are used for dynamic volume allocation.

By configuring the reclaimPolicy property for the storage class, PVs that use the storage class are created with the appropriate reclaim policy. The storage class is configured for the PV using the storageClassName property.

apiVersion: v1
kind: StorageClass
metadata:
  name: gp2-retain
parameters:
  # ...
# ...
reclaimPolicy: Retain
apiVersion: v1
kind: PersistentVolume
# ...
spec:
  # ...
  storageClassName: gp2-retain
Note

If you are using Retain as the reclaim policy, but you want to delete an entire cluster, you need to delete the PVs manually. Otherwise they will not be deleted, and may cause unnecessary expenditure on resources.

12.5.2. Recovery from loss of an OpenShift cluster

When a cluster is lost, you can use the data from disks/volumes to recover the cluster if they were preserved within the infrastructure. The recovery procedure is the same as with namespace deletion, assuming PVs can be recovered and they were created manually.

12.5.3. Recovering a deleted cluster from persistent volumes

This procedure describes how to recover a deleted cluster from persistent volumes (PVs).

In this situation, the Topic Operator identifies that topics exist in Kafka, but the KafkaTopic resources do not exist.

When you get to the step to recreate your cluster, you have two options:

  1. Use Option 1 when you can recover all KafkaTopic resources.

    The KafkaTopic resources must therefore be recovered before the cluster is started so that the corresponding topics are not deleted by the Topic Operator.

  2. Use Option 2 when you are unable to recover all KafkaTopic resources.

    In this case, you deploy your cluster without the Topic Operator, delete the Topic Operator topic store metadata, and then redeploy the Kafka cluster with the Topic Operator so it can recreate the KafkaTopic resources from the corresponding topics.

Note

If the Topic Operator is not deployed, you only need to recover the PersistentVolumeClaim (PVC) resources.

Before you begin

In this procedure, it is essential that PVs are mounted into the correct PVC to avoid data corruption. A volumeName is specified for the PVC and this must match the name of the PV.

For more information, see:

Note

The procedure does not include recovery of KafkaUser resources, which must be recreated manually. If passwords and certificates need to be retained, secrets must be recreated before creating the KafkaUser resources.

Procedure

  1. Check information on the PVs in the cluster:

    oc get pv

    Information is presented for PVs with data.

    Example output showing columns important to this procedure:

    NAME                                         RECLAIMPOLICY CLAIM
    pvc-5e9c5c7f-3317-11ea-a650-06e1eadd9a4c ... Retain ...    myproject/data-my-cluster-zookeeper-1
    pvc-5e9cc72d-3317-11ea-97b0-0aef8816c7ea ... Retain ...    myproject/data-my-cluster-zookeeper-0
    pvc-5ead43d1-3317-11ea-97b0-0aef8816c7ea ... Retain ...    myproject/data-my-cluster-zookeeper-2
    pvc-7e1f67f9-3317-11ea-a650-06e1eadd9a4c ... Retain ...    myproject/data-0-my-cluster-kafka-0
    pvc-7e21042e-3317-11ea-9786-02deaf9aa87e ... Retain ...    myproject/data-0-my-cluster-kafka-1
    pvc-7e226978-3317-11ea-97b0-0aef8816c7ea ... Retain ...    myproject/data-0-my-cluster-kafka-2
    • NAME shows the name of each PV.
    • RECLAIM POLICY shows that PVs are retained.
    • CLAIM shows the link to the original PVCs.
  2. Recreate the original namespace:

    oc create namespace myproject
  3. Recreate the original PVC resource specifications, linking the PVCs to the appropriate PV:

    For example:

    apiVersion: v1
    kind: PersistentVolumeClaim
    metadata:
      name: data-0-my-cluster-kafka-0
    spec:
      accessModes:
      - ReadWriteOnce
      resources:
        requests:
          storage: 100Gi
      storageClassName: gp2-retain
      volumeMode: Filesystem
      volumeName: pvc-7e1f67f9-3317-11ea-a650-06e1eadd9a4c
  4. Edit the PV specifications to delete the claimRef properties that bound the original PVC.

    For example:

    apiVersion: v1
    kind: PersistentVolume
    metadata:
      annotations:
        kubernetes.io/createdby: aws-ebs-dynamic-provisioner
        pv.kubernetes.io/bound-by-controller: "yes"
        pv.kubernetes.io/provisioned-by: kubernetes.io/aws-ebs
      creationTimestamp: "<date>"
      finalizers:
      - kubernetes.io/pv-protection
      labels:
        failure-domain.beta.kubernetes.io/region: eu-west-1
        failure-domain.beta.kubernetes.io/zone: eu-west-1c
      name: pvc-7e226978-3317-11ea-97b0-0aef8816c7ea
      resourceVersion: "39431"
      selfLink: /api/v1/persistentvolumes/pvc-7e226978-3317-11ea-97b0-0aef8816c7ea
      uid: 7efe6b0d-3317-11ea-a650-06e1eadd9a4c
    spec:
      accessModes:
      - ReadWriteOnce
      awsElasticBlockStore:
        fsType: xfs
        volumeID: aws://eu-west-1c/vol-09db3141656d1c258
      capacity:
        storage: 100Gi
      claimRef:
        apiVersion: v1
        kind: PersistentVolumeClaim
        name: data-0-my-cluster-kafka-2
        namespace: myproject
        resourceVersion: "39113"
        uid: 54be1c60-3319-11ea-97b0-0aef8816c7ea
      nodeAffinity:
        required:
          nodeSelectorTerms:
          - matchExpressions:
            - key: failure-domain.beta.kubernetes.io/zone
              operator: In
              values:
              - eu-west-1c
            - key: failure-domain.beta.kubernetes.io/region
              operator: In
              values:
              - eu-west-1
      persistentVolumeReclaimPolicy: Retain
      storageClassName: gp2-retain
      volumeMode: Filesystem

    In the example, the following properties are deleted:

    claimRef:
      apiVersion: v1
      kind: PersistentVolumeClaim
      name: data-0-my-cluster-kafka-2
      namespace: myproject
      resourceVersion: "39113"
      uid: 54be1c60-3319-11ea-97b0-0aef8816c7ea
  5. Deploy the Cluster Operator.

    oc create -f install/cluster-operator -n my-project
  6. Recreate your cluster.

    Follow the steps depending on whether or not you have all the KafkaTopic resources needed to recreate your cluster.

    Option 1: If you have all the KafkaTopic resources that existed before you lost your cluster, including internal topics such as committed offsets from __consumer_offsets:

    1. Recreate all KafkaTopic resources.

      It is essential that you recreate the resources before deploying the cluster, or the Topic Operator will delete the topics.

    2. Deploy the Kafka cluster.

      For example:

      oc apply -f kafka.yaml

    Option 2: If you do not have all the KafkaTopic resources that existed before you lost your cluster:

    1. Deploy the Kafka cluster, as with the first option, but without the Topic Operator by removing the topicOperator property from the Kafka resource before deploying.

      If you include the Topic Operator in the deployment, the Topic Operator will delete all the topics.

    2. Delete the internal topic store topics from the Kafka cluster:

      oc run kafka-admin -ti --image=registry.redhat.io/amq7/amq-streams-kafka-27-rhel7:1.7.0 --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

      The command must correspond to the type of listener and authentication used to access the Kafka cluster.

    3. Enable the Topic Operator by redeploying the Kafka cluster with the topicOperator property to recreate the KafkaTopic resources.

      For example:

      apiVersion: kafka.strimzi.io/v1beta2
      kind: Kafka
      metadata:
        name: my-cluster
      spec:
        #...
        entityOperator:
          topicOperator: {} 1
          #...
    1
    Here we show the default configuration, which has no additional properties. You specify the required configuration using the properties described in Section 13.2.67, “EntityTopicOperatorSpec schema reference”.
  7. Verify the recovery by listing the KafkaTopic resources:

    oc get KafkaTopic

12.6. Tuning client configuration

Use configuration properties to optimize the performance of Kafka 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. For example, for producers you can tune latency and throughput of messages so that clients can respond to data in real time. Or you can change the configuration to provide stronger message durability guarantees.

You might start by analyzing client metrics to gauge where to make your initial configurations, then make incremental changes and further comparisons until you have the configuration you need.

12.6.1. 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.

12.6.1.1. 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.
# ...
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.

12.6.1.2. Data durability

You can apply greater data durability, to minimize the likelihood that messages are lost, using message delivery acknowledgments.

# ...
acks=all 1
# ...
1
Specifying acks=all forces a partition leader to replicate messages to a certain number of followers before acknowledging that the message request was successfully received. Because of the additional checks, acks=all increases the latency between the producer sending a message and receiving acknowledgment.

The number of brokers which need to have appended the messages to their logs before the acknowledgment is sent to the producer is determined by the topic’s min.insync.replicas configuration. A typical starting point is to have a topic replication factor of 3, with two in-sync replicas on other brokers. In this configuration, the producer can continue unaffected if a single broker is unavailable. If a second broker becomes unavailable, the producer won’t receive acknowledgments and won’t be able to produce more messages.

Topic configuration to support acks=all

# ...
min.insync.replicas=2 1
# ...

1
Use 2 in-sync replicas. The default is 1.
Note

If the system fails, there is a risk of unsent data in the buffer being lost.

12.6.1.3. 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, enabling idempotency makes sense for ordered delivery.

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 are not using acks=all and 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.

12.6.1.4. 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.

12.6.1.5. Optimizing 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.

12.6.2. 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.

12.6.2.1. 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.
# ...
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.

Consumer groups are used to share a typically large data stream generated by multiple producers from a given topic. Consumers are grouped using a group.id, allowing messages to be spread across the members.

12.6.2.2. 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 with the same group.id property are in the same group. 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.

12.6.2.3. 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.

12.6.2.4. Optimizing 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.

12.6.2.5. 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.

12.6.2.5.1. 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.

12.6.2.6. 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.

12.6.2.7. 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=10000 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.

12.6.2.8. 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.

12.7. Uninstalling AMQ Streams

This procedure describes how to uninstall AMQ Streams and remove resources related to the deployment.

Prerequisites

In order to perform this procedure, identify resources created specifically for a deployment and referenced from the AMQ Streams 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, KafkaConnectS2I, KafkaMirrorMaker, or KafkaBridge configuration.

Procedure

  1. Delete the Cluster Operator Deployment, related CustomResourceDefinitions, and RBAC resources:

    oc delete -f install/cluster-operator
    Warning

    Deleting CustomResourceDefinitions results in the garbage collection of the corresponding custom resources (Kafka, KafkaConnect, KafkaConnectS2I, KafkaMirrorMaker, or KafkaBridge) and the resources dependent on them (Deployments, StatefulSets, and other dependent resources).

  2. Delete the resources you identified in the prerequisites.

12.8. Frequently asked questions