chaoyoung opened a new pull request, #16583:
URL: https://github.com/apache/dubbo/pull/16583

   ## What is the purpose of the change?
   
   A Triple HTTP/2 write batch can end on a closed stream. Flushing that final 
child channel leaves an earlier live sibling's response buffered and its write 
future pending. Flush through the channel that owns `HttpWriteQueueHandler` at 
the existing batch boundaries instead.
   
   Add an optional batch flush callback to `HttpWriteQueue`, retaining the 
original constructor's fallback behavior. The handler supplies 
`ctx.channel()::flush`, which traverses the owning channel's full outbound 
pipeline, including codecs and flush consolidation. Command execution, 
ordering, chunk sizes and future propagation remain unchanged.
   
   Related: [#11929](https://github.com/apache/dubbo/issues/11929), the closed 
parent-channel flush proposal. The regression here specifically covers a batch 
whose final command targets a closed HTTP/2 stream.
   
   ## Validation
   
   - Before the fix, a paired-codec regression passed with a live final stream 
and failed with a closed final stream (2 tests, 1 failure, 0 errors). The 
failure was missing connection outbound response bytes, observed before any 
extra connection flush or peer pumping.
   - After the fix, 4 paired-codec cases pass: live/closed final stream, each 
with/without `FlushConsolidationHandler`. They assert outbound bytes, live 
write completion, independent final-write success/failure and one connection 
flush per small batch.
   - 5 queue tests cover 1, 2, 129, 130 and 259 commands, including scheduling, 
order, completion and the existing chunk boundaries.
   - The complete `dubbo-remoting-http12` test suite passes: 40 tests, 0 
failures, 0 errors, 2 existing disabled `XmlSafetyTest` cases. Spotless check 
passes in the same reactor invocation.
   
   ```sh
   mvn -pl dubbo-remoting/dubbo-remoting-http12 -am \
     '-Dtest=org/apache/dubbo/remoting/http12/**' \
     -Dsurefire.failIfNoSpecifiedTests=false test
   ```
   
   Dependencies are built in the reactor; no local artifact installation is 
required. Verification used OpenJDK 25.0.4.1 and Maven 3.9.16. This is an 
`EmbeddedChannel` transport regression, not a socket-level/full-RPC test or a 
measurement of production frequency. HTTP/3 and WebSocket integration are not 
exercised. The full repository test suite has not been run.
   
   Fixes #16582
   
   AI assistance was used to develop the change, tests, and this description.
   


-- 
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