Loading...

Type: Story
Resolution: Done
Priority: Major
Fix Version/s: None
Affects Version/s: 2.1.0.GA
Component/s: None
Labels:
None

Blocked:
False
Blocked Reason:

Hide

None

Show
None
Ready:
False
GSS Priority:

SFDC Cases Counter:
SFDC Cases Open:
SFDC Cases Links:

[Problem] How to configure producers/consumers in MirrorMaker2 connectors?
- In particular, the producer and consumer configurations in MirrorSourceConnector are important for tuning MirrorMaker2. However, currently there is no clear description of it in our documentation:
  - What configuration methods are provided for producers/consumers?
  - Which Kafka CR yaml configurations affect which producer/consumer configurations?
- In fact, even the official upstream Kafka documentation on MirrorMaker2 is quite unclear on how to configure producers/consumers in MirrorMakers.
- And for example, “my-target-cluster.config:producer.buffer.memory” is used in our documentation(1), however, there is no mention explaining that a producer.* or consumer.* is available in Kafka CR yaml.
  - Note that "my-target-cluster.config:producer.buffer.memory" affects not only MirrorSourceConnector’s producer but also MirrorCheckpointConnector’s and MirrorHeartbeatConnector’s one. (I don't think it's a good method to configure it.)

[RFE 1] The documentation include how to configure the MirrorMaker2 connector producers/consumers

List of producers/consumers within each in connectors
How to change the configurations for each producer/consumer

For example, list of producers/consumers and configurations as below, (as long as each of these configurations is used, it is configured only for the corresponding producer/consumer and does not affect others):

- MirrorSourceConnector
    - producer (mirrors.sourceConnector.config: producer.override.*) #created in Kafka Connect framework
    - consumer (mirrors.sourceConnector.config: consumer.* or mirrors.sourceConnector.config: source.consumer.* ) #created in MirrorMaker2
    - producer for offset commit (mirrors.sourceConnector.config: producer.*) #created in MirrorMaker2
- MirrorHeartbeatConnector
    - producer (mirrors.heartbeatConnector.config: producer.override.*) #created in Kafka Connect framework
- MirrorCheckpointConnector
    - producer (mirrors.checkpointConnector.config: producer.override.*) #created in Kafka Connect framework
    - consumer (mirrors.checkpointConnector.config: consumer.*) #created in MirrorMaker2
hoge

[RFE 2] Add configuration items for producers/consumers in each connector definition in Kafka CR yaml

As you can see from the above list of producers/consumers in {}RFE 1{}, at first glance at the configuration, it is not clear which producer/consumer is affected.
- some have "override" and others have not. These cases depend on whether the producers/consumers were created with the Kafka Connect framework or with MirrorMaker2.

Therefore, it could be helpful to the user to add explicit producer and consumer configuration items. For example:

apiVersion: kafka.strimzi.io/v1beta2                                                                                                                                                                                                                                                                                                                                kind: KafkaMirrorMaker2
metadata:
  name: my-mirror-maker-2
spec:
...
  mirrors:
...
    sourceConnector:
       config:
           ...
      sourceProducer:
        config:
           ...
      sourceConsumer:
        config:
           ...
      offsetProducer:
        config:
           ...
...

In this case, it is not enough to just add the documentation, because it involves changes to the Kafka CRD.

As another option, there is also a way to provide a prefix for specific producer/consumer in mirrors.xxxConnector.config without changing the Kafka CRD as below:

apiVersion: kafka.strimzi.io/v1beta2                                                                                                                                                                                                                                                                                                                                kind: KafkaMirrorMaker2
metadata:
  name: my-mirror-maker-2
spec:
...
  mirrors:
...
    sourceConnector:
      config:
        sourceProducer.xxx.xxx
        sourceConsumer.xxx.xxx
        offsetProducer.xxx.xxx

- I have stated this as a RFE to consider in development team, however, I am neutral on this RFE to change this Kafka CR yaml. There are also some concerns:
  - The producer/consumer of each connector is not documented in the official community Kafka documentation.
  - New producer/consumer may be added/removed in the future. The challenge is whether we can continue to maintain them.
  - For a radical solution, the upstream Kafka community may need to list producers/consumers in MirrorMaker2 connectors and provide explicit ways to configure them.
    - In fact, even the upstream Kafka community documentation on MirrorMaker2 is quite unclear on how to configure producers/consumers.
    - And the scope of the configuration's impact is completely unimaginable.
    - I believe I can create KIP/JIRA/PR if needed for the upstream Kafka community.

[as far as I've researched]
- MirrorMaker2 has multiple Kafka Connectors, with multiple producers/consumers within each, some of them belonging to the Kafka Connect framework, and others specific to MirrorMaker2.
- The configuration method differs between MirrorMaker-specific producers/consumers and Kafka Connect framework producers/consumers.
  - producers/consumers created in Kafka Connect framework use following configurations
    - mirrors.xxxConnector.config: {producer or consumer}.override.*
    - {source_or_target}
      .config: {producer_or_consumer}.* is overridden

- producers/consumers created in MirrorMaker2 use following configurations
  - mirrors.xxxConnector.config: {source_or_target}.{producer_or_consumer}
  - mirrors.xxxConnector.config: {producer_or_consumer}.* is overridden

list of producers/consumers and configurations

As far as I have actually tried(attached file: kafka-mirror-maker-2.yaml) and confirmed them using the debugger, the following configurations were used for each producer/consumer.

    - MirrorSourceConnector
        - producer in WorkerSourceTask #created in Kafka Connect framework
            - sourceConnector.config: producer.override.buffer.memory(=7770303)
                - my-target-cluster.config: producer.buffer.memory(=7770202) is overridden
        - consumer in MirrorSourceTask:93 #created in MirrorMaker2
            - (sourceConnector.config: source.consumer.fetch.max.bytes(=55500306))
                - always prefix is “source”. However, the prefix can be omitted, as it is sufficiently specific.
            - sourceConnector.config: consumer.fetch.max.bytes(=55500302) 
        - offsetProducer in MirrorSourceTask:94 as offsetProducer : #created in MirrorMaker2
            - (sourceConnector.config: source.producer.buffer.memory(=7770306))
                - “source” prefix depend on where OffsetSyncStore is created on source or target cluster. However, the prefix can be omitted, as it is sufficiently specific.
            - sourceConnector.config:producer.buffer.memory(=7770302)
    - MirrorHeartbeatConnector
        - producer in WorkerSourceTask  #created in Kafka Connect framework
            - heartbeatConnector.config: producer.override.buffer.memory(=7770403)
                - my-target-cluster.config: producer.buffer.memory(=7770202) is overridden
    - MirrorCheckpointConnector
        - producer in WorkerSourceTask #created in Kafka Connect framework
            - checkpointConnector.config: producer.override.buffer.memory(=7770503)
                - my-target-cluster.config: producer.buffer.memory(=7770202) is overridden
        - consumer in OffsetSyncStore in MirrorCheckpointTask:94  #created in MirrorMaker2 
            - (checkpointConnector.config: source.consumer.fetch.max.bytes(=55500506))
                - “source” prefix depend on where OffsetSyncStore is created on source or target cluster. However, the prefix can be omitted, as it is sufficiently specific.
            - checkpointConnector.config: consumer.fetch.max.bytes:55500502
    - Kafka Connect
        - KafkaBasedLog for each backing stores (KafkaOffsetBackingStore, KafkaStatusBackingStore and KafkaConfigBackingStore)
            - producer #value=  my-target-cluster.config:buffer.memory=7770201
            - consumer #value= my-target-cluster.config:fetch.max.bytes=55500201

Investigating AMQ Streams on RHEL

I used the attached mm2deployment_rhel.yaml for testing.
The following result is with only my-source-cluster->my-target-cluster enabled and "/opt/kafka/bin/connect-mirror-maker.sh --clusters my-target-cluster".
However, if my-source-cluster->my-target-cluster & my-target-cluster->my-source-cluster are enabled, the cluster with inverted source and target are also used as valid producer/consumer configurations.

- MirrorSourceConnector 
    - consumer to retrieve topic messages from the source cluster
        - connect-mirror-maker.properties: my-source-cluster.consumer.fetch.max.bytes 55500110(MirrorSourceTask:93)
    - producer to send topic messages to the target cluster
        - connect-mirror-maker.properties: my-target-cluster.producer.buffer.memory 7770111(buildWorkerTak:601) 
    - producer to write to the offset sync topic, which maps the source and target offsets for replicated topic partitions.
        - connect-mirror-maker.properties: my-source-cluster.producer.buffer.memory 7770110(MirrorSourceTask:94)
- MirrorHeartbeatConnector
    - producer to emit heartbeats 
        - connect-mirror-maker.properties: my-target-cluster.producer.buffer.memory 7770111(buildWorkerTask:601) 
- MirrorCheckpointConnector
    - producer to emits consumer offset checkpoints
        - connect-mirror-maker.properties: my-target-cluster.producer.buffer.memory 7770111(buildWorkerTask:601)
    - consumer to load the offset sync topic 
        - connect-mirror-maker.properties: my-source-cluster.consumer.fetch.max.bytes 55500110(MirrorCheckpointTask:94)

(1) "producer.buffer.memory" is used in our documentation
https://access.redhat.com/documentation/en-us/red_hat_amq_streams/2.1/html-single/configuring_amq_streams_on_openshift/index#con-mirrormaker-high-volume-messages-str

- - Sort By Name
  - Sort By Date
  - Ascending
  - Descending
  - Thumbnails
  - List
  - Download All

ENTMQST-4116_kafka-mirror-maker-2.yaml
7 kB
2022/07/01 1:39 PM
mm2deployment_rhel.yaml
3 kB
2022/07/26 12:14 PM

Details

Description

Attachments

Attachments

Easy Agile Planning Poker

Activity

People

Dates