[
https://issues.apache.org/jira/browse/BEAM-12631?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17478262#comment-17478262
]
Kenneth Knowles commented on BEAM-12631:
----------------------------------------
We tend to prefer passing in functions for just the parts where it is safe to
provide custom behavior; this is generally better than inheritance + override.
In this case, I think the runner is expected to choose a number of splits that
work best. [~chamikara] may have some ideas about the design intention.
> Allow Inheritance from KafkaUnboundedSource
> -------------------------------------------
>
> Key: BEAM-12631
> URL: https://issues.apache.org/jira/browse/BEAM-12631
> Project: Beam
> Issue Type: New Feature
> Components: io-java-kafka
> Reporter: HongXiang Ren
> Priority: P3
> Original Estimate: 4h
> Remaining Estimate: 4h
>
> Reason:
> We have dataflow to call one internal API (the API call take 60ms). So we
> want to start more threads to call this API. Kafka IO thread is control by
> split by default. And split is control by total partitions and system
> suggestion number. And system suggestion number is 4 * max workers. which is
> too low for us.
> Solution:
> Use facture to allow user inheritance KafkaUnboundedSource. and override the
> split function like
>
> {code:java}
> // code placeholder
> @Override
> public List<KafkaUnboundedSource<K, V>> split(int desiredNumSplits,
> PipelineOptions options)
> throws Exception {
> int split = this.getMaxSplits();
> if(split <= 0){
> return super.split(desiredNumSplits, options);
> }else{
> return super.split(split, options);
> }
> }
> {code}
> this may also useful for Override other feature of the KafkaUnboundedSource
> without copy all the io code.
--
This message was sent by Atlassian Jira
(v8.20.1#820001)
