[
https://issues.apache.org/jira/browse/KAFKA-8586?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16871481#comment-16871481
]
ASF GitHub Bot commented on KAFKA-8586:
---------------------------------------
C0urante commented on pull request #6993: KAFKA-8586: Fail source tasks when
producers fail to send records
URL: https://github.com/apache/kafka/pull/6993
[Jira](https://issues.apache.org/jira/browse/KAFKA-8586)
Previously, if the producer for a source task failed to send a record with a
non-retriable error, the record would be silently skipped over. The source task
would be allowed to commit offsets for the skipped record, and its status would
remain at `RUNNING`.
The changes here cause source tasks to transition to the `FAILED` state if
their producers fail to send a record with a non-retriable error, and they also
change the logic for offset commits to wait for confirmation that records have
made it to Kafka before their offsets can be committed.
Tested by running Connect unit tests locally.
### Committer Checklist (excluded from commit message)
- [ ] Verify design and implementation
- [ ] Verify test coverage and CI build status
- [ ] Verify documentation (including upgrade notes)
----------------------------------------------------------------
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
For queries about this service, please contact Infrastructure at:
us...@infra.apache.org
> Source task producers silently fail to send records
> ---------------------------------------------------
>
> Key: KAFKA-8586
> URL: https://issues.apache.org/jira/browse/KAFKA-8586
> Project: Kafka
> Issue Type: Bug
> Components: KafkaConnect
> Affects Versions: 2.3.0
> Reporter: Chris Egerton
> Assignee: Chris Egerton
> Priority: Major
>
> The Connect framework marks source records as successfully sent when they are
> dispatched to the producer, instead of when they are actually sent to Kafka.
> [This is assumed to be good
> enough|https://github.com/apache/kafka/blob/3e9d1c1411c5268de382f9dfcc95bdf66d0063a0/connect/runtime/src/main/java/org/apache/kafka/connect/runtime/WorkerSourceTask.java#L324-L331]
> since the Connect framework sets up its producer to use infinite retries on
> retriable errors, but in the case of an authorization or authentication
> failure with a secured Kafka broker, the errors aren't retriable and cause
> the producer to invoke its send callback with an exception and then give up
> on sending the message. This is a problem since the callback currently used
> by the WorkerSourceTask class when it invokes Producer.send(...) logs the
> exception and does nothing else. This leads to data loss since the source
> offsets for those failed records are committed, and the status of the task is
> never affected so users may not even know that something is wrong unless they
> check the worker log files or notice that data isn't flowing into Kafka.
> Until and unless someone does notice that something's wrong, the task will
> continue processing records and committing offsets, even though nothing is
> making it into Kafka.
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
