[
https://issues.apache.org/jira/browse/KAFKA-10186?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17182086#comment-17182086
]
Matthias J. Sax commented on KAFKA-10186:
-----------------------------------------
Thanks [~nym3r0s] – working on the code in parallel is not a requirement and
the KIP itself should not contain implementation details – it could still be
useful to do both in parallel. It's up to you what you find more useful.
There is a typo in the KIP:
{quote}he is a valid state to be in - the
{quote}
And yes, the KIP should propose a concrete solution before you start the
discussion, but it's sufficient to cover the the new exception class we use in
the KIP, something like `public class TransactionAbortedException extends
ApiException` (just as an initial though).
> Aborting transaction with pending data should throw non-fatal exception
> -----------------------------------------------------------------------
>
> Key: KAFKA-10186
> URL: https://issues.apache.org/jira/browse/KAFKA-10186
> Project: Kafka
> Issue Type: Improvement
> Components: producer
> Reporter: Sophie Blee-Goldman
> Assignee: Gokul Srinivas
> Priority: Major
> Labels: needs-kip, newbie, newbie++
>
> Currently if you try to abort a transaction with any pending (non-flushed)
> data, the send exception is set to
> {code:java}
> KafkaException("Failing batch since transaction was aborted"){code}
> This exception type is generally considered fatal, but this is a valid state
> to be in -- the point of throwing the exception is to alert that the records
> will not be sent, not that you are in an unrecoverable error state.
> We should throw a different (possibly new) type of exception here to
> distinguish from fatal and recoverable errors.
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
