zjncs opened a new pull request, #11063:
URL: https://github.com/apache/rocketmq/pull/11063

   ### Motivation
   
   In `ConsumeMessagePopConcurrentlyService.checkNeedAckOrDelay`, the delay 
level for the next retry is picked by scanning the pop delay table top-down:
   
   ```java
   int delayLevel = delayLevelTable.length - 1;
   for (; delayLevel >= 0; delayLevel--) {
       if (msgDelaytime >= delayLevelTable[delayLevel] * 1000) {
           delayLevel++;
           break;
       }
   }
   
   changePopInvisibleTime(msgExt, consumerGroup, delayLevel);
   ```
   
   If the message is younger than `delayLevelTable[0]` seconds (`msgDelaytime < 
delayLevelTable[0] * 1000`) the loop never breaks and exits with `delayLevel == 
-1`. `changePopInvisibleTime` only guards `0 == delayLevel`, so `-1` falls 
through to:
   
   ```java
   int delaySecond = delayLevel >= delayLevelTable.length ? 
delayLevelTable[delayLevelTable.length - 1] : delayLevelTable[delayLevel];
   ```
   
   which evaluates `delayLevelTable[-1]` **before** the try block and throws 
`ArrayIndexOutOfBoundsException`. The exception propagates through 
`processConsumeResult` into `ConsumeRequest.run()`, which has no try/catch: the 
consume task dies, the remaining messages of the batch are neither acked nor 
scheduled for retry, and the post-consume hooks/stats are skipped.
   
   This is reachable whenever a redelivered message is consumed while younger 
than the first delay level (10s by default) — e.g. with a small 
`maxReconsumeTimes` (setting 0 makes every first failed consume of a fresh 
message hit this path) or with producer/consumer clock skew shrinking 
`msgDelaytime`.
   
   ### Modifications
   
   Clamp `delayLevel` to 0 after the loop. Level 0 is already handled by 
`changePopInvisibleTime` by falling back to `msg.getReconsumeTimes()`, which 
yields a sensible first-level delay.
   
   ### Verification
   
   Fail-before (new test 
`testProcessConsumeResultWithMaxReconsumeReachedAndFreshMessage`, run against 
the unpatched code — `RECONSUME_LATER` with `maxReconsumeTimes=0`, pop delay 
table `{10, 30, 60}` and a message 1s old):
   
   ```
   Tests run: 2, Failures: 0, Errors: 1 -- 
ConsumeMessagePopConcurrentlyServiceTest
   java.lang.ArrayIndexOutOfBoundsException: Index -1 out of bounds for length 3
   ```
   
   Pass-after:
   
   ```
   Tests run: 11, Failures: 0, Errors: 0, Skipped: 0 -- 
ConsumeMessagePopConcurrentlyServiceTest
   ```
   


-- 
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