zhangshenghang opened a new issue, #12383:
URL: https://github.com/apache/seatunnel/issues/12383

   ### Search before asking
   
   - [x] I searched in the [issues](https://github.com/apache/seatunnel/issues) 
and found nothing similar.
   
   ### What happened
   
   `RocketMqIT` failed 43 invocations (14 failures + 29 errors) in a single CI 
run — every container leg of `testSinkRocketMq` (and siblings) errored at the 
same place: after the test consumer successfully polled messages and called 
`commitSync()`, the broker-side consumer offset for `MessageQueue 
[topic=test_topic, brokerName=broker-a, queueId=0]` was **never visible** — not 
just lagging, but absent from `RocketMqAdminUtil.currentOffsets` — for the full 
60s window.
   
   ### Evidence (fork CI run on branch `fix-12353-cancel-state-priority`, 
2026-09-17, jobs `rocketmq-connector-it (8/11, ubuntu-latest)`)
   
   ```
   ConditionTimeoutException: Consume offset should be visible for MessageQueue
   [topic=test_topic, brokerName=broker-a, queueId=0] ==> expected: not <null> 
within 1 minutes.
       at RocketMqIT.waitConsumedOffsetsSynced(RocketMqIT.java:577)
       at RocketMqIT.getRocketMqConsumerData(RocketMqIT.java:558)
   ```
   
   - Message consumption itself worked (test logs show `Consumer test_topic 
data total 10..70`).
   - `updateConsumeOffsetToBroker` + `commitSync()` were called (the while loop 
completed — `data total` is logged after it).
   - The admin-side offset query then never returned an entry for broker-a 
queue 0 within 60s.
   
   Note this window was already widened from 30s → 60s for a previous 
occurrence, so "slow broker" alone does not explain an offset that never 
appears at all.
   
   ### Expected behavior
   
   After `commitSync()` returns, `queryConsumerOffset`/admin offset lookups for 
the consumed queues should observe the committed offset (within the retry 
window).
   
   ### Suspect areas
   
   - Offset commit vs. consumer group rebalance: the test consumer shuts down 
right after `commitSync()`; if a rebalance/instance deregistration races the 
commit, the broker may drop or never persist the offset for that queue.
   - `RocketMqAdminUtil.currentOffsets` silently omitting queues whose 
`queryConsumerOffset` returns `OFFSET_NOT_FOUND` or throws, which turns a 
broker-side miss into a permanent map miss instead of a retryable lag.
   - Topic route propagation timing for queues on `broker-a` when multiple 
queues/brokers exist.
   
   ### Are you willing to submit PR?
   
   - [ ] Yes, I am willing to investigate and submit a PR, but would first like 
maintainer input on the suspect areas above.


-- 
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]

Reply via email to