Github user jfenc91 commented on a diff in the pull request: https://github.com/apache/storm/pull/1679#discussion_r79659246 --- Diff: external/storm-kafka-client/src/main/java/org/apache/storm/kafka/spout/KafkaSpout.java --- @@ -266,26 +266,32 @@ private void doSeekRetriableTopicPartitions() { if (offsetAndMeta != null) { kafkaConsumer.seek(rtp, offsetAndMeta.offset() + 1); // seek to the next offset that is ready to commit in next commit cycle } else { - kafkaConsumer.seekToEnd(toArrayList(rtp)); // Seek to last committed offset + kafkaConsumer.seek(rtp, acked.get(rtp).committedOffset + 1); // Seek to last committed offset --- End diff -- @srdo Yes, in the case that the failed tuple is not the first message in the batch since the last committed offset things worked fine how they were previously. I agree. Honestly, the idea of auto commit mode seems to go against the philosophy of storm's processing guarantees. Putting the offsets that need to be retried in memory isn't enough in the case of restarts. I propose we address auto commit mode in a separate PR. Sound alright?
--- If your project is set up for it, you can reply to this email and have your reply appear on GitHub as well. If your project does not have this feature enabled and wishes so, or if the feature is enabled but not working, please contact infrastructure at infrastruct...@apache.org or file a JIRA ticket with INFRA. ---