SEZ9 commented on issue #9940: URL: https://github.com/apache/seatunnel/issues/9940#issuecomment-5724001633
@SEPURI-SAI-KRISHNA thanks for the correction and for keeping the detailed write-up on #12332 rather than here. On the mechanism: agreed that if `examineConsumeStats` resolves its route only from the group's `%RETRY%` topic, then the `TOPIC_NOT_EXIST` (17) that `RocketMqAdminUtil.currentOffsets` swallows into an empty map is about the retry topic, not the data topic. That is a meaningful narrowing, and your three questions are the right ones to separate this from the two causes already recorded in this thread: 1. `No topic route info` naming `%RETRY%` + consumer group instead of the data topic 2. data topic route healthy at the same moment 3. broker still showing committed offsets for `track_report_group` I also appreciate you withdrawing the "only has to be unlucky once" point. With `setPartitionStartOffset()` invoked once from `run()` and not from `discoverySplits()`, the exposure is a single lookup at start, which matches your revised description. What I would still like from you: - On #12349, please make sure the change distinguishes "retry-topic route temporarily unavailable" from "group has genuinely committed nothing" rather than just retrying the lookup, and add a test that exercises the response-code-17-on-`%RETRY%` path so we do not regress back to treating it as an empty offset map. - Since this is still established from code and response codes rather than a captured rewind, please note that explicitly in the #12332 description so anyone triaging knows the evidence bar. - If the original reporter comes back with the three data points above, let's record the outcome here so this issue's conclusion stays accurate. Otherwise the earlier status in this thread stands: checkpoint-restore duplicates are addressed by #10778, a true cold start with no committed offset is expected RocketMQ behavior, and your third path is tracked in #12332 / #12349. <!-- streview-comment:1122 --> -- 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]
