RockteMQ-AI commented on issue #10911:
URL: https://github.com/apache/rocketmq/issues/10911#issuecomment-5262139665

   **Issue Evaluation**
   
   Category: `bug` | Status: **Confirmed — Needs Investigation**
   
   Thank you for the exceptionally detailed report. The evidence you have 
gathered is compelling:
   
   - **BODYCRC match** confirms the corruption occurred before persistence (not 
a disk or read-tool issue)
   - **Cross-connection frame embedding** (PULL_MESSAGE frames inside a 
SEND_MESSAGE body) points to a problem in the Broker's Netty receive/decode 
pipeline
   - **Temporal correlation** with `TooLongFrameException` bursts across 
independent connections strengthens the decode-path hypothesis
   
   **Key areas to investigate in the codebase:**
   
   1. `RemotingDecoder` / `NettyDecoder` — whether `cumulator` 
(ByteToMessageDecoder) can leak or share `ByteBuf` references across 
connections when `serverPooledByteBufAllocatorEnable=true`
   2. `NettyRemotingServer` channel pipeline — whether pooled direct buffers 
are correctly isolated per-channel under high concurrency with mixed SEND/PULL 
traffic
   3. `Broker2Client` response path — whether callback frames from internal 
broker-to-broker or broker-to-client paths can bleed into the user-data write 
path
   
   **Severity:** High — silent data corruption in persisted messages is a data 
integrity issue.
   **Reproducibility:** Not yet reproduced on unmodified 4.9.8, which makes 
this harder to triage. The distribution-specific audit you mentioned is 
important.
   
   This warrants a deep code review of the remoting layer, particularly the 
interaction between pooled ByteBuf allocators and the frame decoder under 
concurrent heterogeneous traffic.
   
   ---
   *Automated evaluation by RockteMQ-AI*
   


-- 
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