chaoyoung commented on issue #16415:
URL: https://github.com/apache/dubbo/issues/16415#issuecomment-6757391758

   Independently reproduced the same callback-executor ownership gap with 
released Dubbo 3.3.7, Netty 4.2.19.Final and OpenJDK 25. No custom transport 
SPI was loaded.
   
   The useful distinguishing sequence is a synchronous unary deadline **before 
response headers**:
   
   1. A real DeadlineFuture expires; AsyncRpcResult closes the per-call 
ThreadlessExecutor.
   2. UnaryClientCallListener.onStart has not run, so its timeout-cancellation 
listener has not yet been registered. The HTTP/2 stream can remain active.
   3. Response HEADERS and DATA are exchanged through paired Netty HTTP/2 
codecs and multiplexed child channels.
   4. The real response handler submits DATA to SerializingExecutor, but the 
shutdown executor drops the task. The captured DATA buffer remains at refCnt=1, 
including after channel/decoder cleanup. The fixture manually releases it only 
after recording those observations.
   
   This corroborates the root cause described by #16418; I am not opening a 
competing fix PR. A regression that exercises an actual pre-header deadline and 
late codec-delivered DATA would complement the existing shutdown unit test. The 
paired-codec fixture still uses a mocked upper client-call listener; this is 
not a production-load measurement or a full TCP RPC benchmark.
   
   There is also a useful negative control: once local RST has actually been 
processed and exchanged, subsequent peer DATA writes fail and produce no client 
DATA callbacks. Directly injecting DATA after cancellation should not be 
presented as proof that it can arrive over a reset wire stream.
   
   I have not applied #16418 to this fixture, so this comment does not claim 
that the PR has been verified to cover all submission races or all cancellation 
behavior.
   
   Investigation and this summary were AI-assisted.
   


-- 
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]


---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]

Reply via email to