Hi,
while testing Kafka 3.9 (KRaft), we encountered Unclean Leader Elections (ULE) during a partition reassignment. We found an existing issue that describes the problem: https://issues.apache.org/jira/browse/KAFKA-19148 (https://issues.apache.org/jira/browse/KAFKA-19148) As suggested in the issue, the ULE appears to be a false positive and is fixed in Kafka 4.1 (which we have verified). However, when reproducing the problem using the README attached to KAFKA- 19148, we observed message loss (not only reordering) while producing to the cluster with acks=1 (--producer-property acks=1) during the reassignment. My question: Is this message loss expected behavior when producing with acks=1 during partition reassignment? Notes: - The KAFKA-19148 issue has not received any updates for a long time (and is primarily about ULE). - I understand that acks=1 increases throughput but is not safe if the leader is removed immediately or unexpectedly. - We reproduced the message loss on older Kafka versions as well (3.9 KRaft, 3.9 Zookeeper, 2.3, 1.1). Thank you, Pavel -- Pavel Čermák Infrastructure Teamleader Level 1 tel: +420 724 443 700 [email protected](mailto:[email protected]) www.seznam.cz(https://www.seznam.cz/) Seznam.cz, a.s., Radlická 3294/10, 150 00 Praha 5 (https://mapy.com/s/eznam#szn-eml-no-preview)
