KAFKA-13030 FindCoordinators batching causes slow poll when requestin…

[chia7712]

issue: https://issues.apache.org/jira/browse/KAFKA-13030

related: #10743

https://github.com/apache/kafka/blob/trunk/clients/src/main/java/org/apache/kafka/clients/consumer/internals/AbstractCoordinator.java#L888

            if (e instanceof NoBatchedFindCoordinatorsException) {
                batchFindCoordinator = false;
                clearFindCoordinatorFuture();
                lookupCoordinator();
                return;
            } 

The current request future is NOT updated so it can't be completed until timeout. It causes a slow poll when user poll data from older broker the first time.

Broken Behavior

  var consumer = new KafkaConsumer<>();
  var records = consumer.poll(Duration.ofSeconds(30));
  // there are producers which are pushing data to topic.
  // [BEFORE PR-10743] the return data from fist poll could NOT be empty
  // [AFTER PR-10743] the return data from first poll is always empty. To make matters worst, it returns only if duration is expired (if users set a large duration ...)
  assert records.count() > 0;

Committer Checklist (excluded from commit message)

• Verify design and implementation
• Verify test coverage and CI build status
• Verify documentation (including upgrade notes)

[showuon]

@chia7712 , nice find! Could we add tests for it?

[chia7712]

nice find! Could we add tests for it?

done

[mimaison]

Thanks @chia7712. This if statement is being removed in #10965 so I think it should address this issue

[chia7712]

This if statement is being removed in #10965 so I think it should address this issue

you are right. close this as duplicate :)