[
https://issues.apache.org/jira/browse/KAFKA-8021?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16781850#comment-16781850
]
Abhishek Mendhekar commented on KAFKA-8021:
-------------------------------------------
[~sriharsha] the issue is in 0.11 and higher version (including latest trunk).
Since this is a very small race I am trying to reproduce this in an integration
test but I was able to verify it via unit tests.
> KafkaProducer.flush() can show unexpected behavior when a batch is split
> ------------------------------------------------------------------------
>
> Key: KAFKA-8021
> URL: https://issues.apache.org/jira/browse/KAFKA-8021
> Project: Kafka
> Issue Type: Bug
> Components: clients, producer
> Affects Versions: 0.11.0.0
> Reporter: Abhishek Mendhekar
> Assignee: Abhishek Mendhekar
> Priority: Major
>
> KafkaProducer.flush() marks the flush in progress and then waits for all
> incomplete batches to be completed (waits on the producer batch futures to
> finish).
> The behavior is seen when a batch is split due to MESSAGE_TOO_LARGE exception.
> The large batch is split into smaller batches (2 or more) but
> ProducerBatch.split() marks the large batch future as complete before adding
> the new batches in the incomplete list of batches. At this time if the
> KafkaProducer.flush() is called then it'll make a copy of existing incomplete
> list of batches and waits for them to complete while ignoring the large batch
> that was split into smaller batches.
>
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
