Chapter 6. Handling high volumes of messages

If your Streams for Apache Kafka 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) and sink connectors.

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, 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:

For Kafka Connect sink connectors, the data streaming pipeline to the target external data source is as follows:

For MirrorMaker 2, the data mirroring pipeline to the target Kafka cluster is as follows:

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:

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.

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
  # ...

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
    # ...

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
    # ...

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
    }
}'

apiVersion: kafka.strimzi.io/v1beta2
kind: KafkaMirrorMaker2
metadata:
  name: my-mirror-maker2
spec:
  version: 3.7.0
  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