Justine Olshan created KAFKA-18464:
--------------------------------------
Summary: Empty Abort Transaction can fence producer incorrectly
with Transactions V2
Key: KAFKA-18464
URL: https://issues.apache.org/jira/browse/KAFKA-18464
Project: Kafka
Issue Type: Bug
Affects Versions: 4.0.0
Reporter: Justine Olshan
Assignee: Justine Olshan
Consider the following scenario using transactions V2.
Producer writes a transaction and commits.
Producer starts the next transaction and tries to add a partition but gets an
non-retriable error. Because the application doesn't know whether the partition
was added/records written the producer aborts. In the case where the partition
was not added, and the previous transaction was still in PrepareCommit state,
we self-fence.
This is because the epoch of the current state == incoming request epoch and
the PrepareCommit state currently only allows for retries where the request
epoch is -1 the current epoch.
Instead of hitting this, we should just return the concurrent transactions
error when we have pending state and do the epoch check second. In this
scenario, we will complete commit before proceeding to the empty abort.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
