slachiewicz opened a new pull request, #314: URL: https://github.com/apache/flink-connector-kafka/pull/314
With POOLING a committed id is recycled under a bumped epoch. Recovering from the earlier checkpoint re-commits it and is fenced; the writer's LISTING abort skips the id as precommitted, so the newer open transaction lingers until `transaction.timeout.ms` and pins the last stable offset. On a fenced commit with reused ids the committer now bumps the epoch of that id through a short-lived producer, which aborts the orphan. INCREMENTING is unchanged. Root cause of one of the two failure modes in FLINK-40585; the other is a harness race fixed in #313. Verified: new `KafkaCommitterTest` case fails on `main` and passes here; `KafkaSinkITCase#rescaleListing` 10/10 and `ExactlyOnceKafkaWriterITCase` 13/13 locally. *This change was created with AI assistance.* -- 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. To unsubscribe, e-mail: [email protected] For queries about this service, please contact Infrastructure at: [email protected]
