SEZ9 commented on issue #13320:
URL: https://github.com/apache/seatunnel/issues/13320#issuecomment-7022882417

   @imRYiUK thanks for picking this up and for the clear summary.
   
   On your question: constants are fine for this fix, so please don't add 
settings for the retry count and interval. A bounded 3 attempts, 1s apart, 
keeps the change small and inside `PhysicalVertex`. If real-case logs later 
show that window is too short, we can revisit making them configurable.
   
   The behaviour you describe matches the plan you posted:
   - A failed `CheckTaskGroupIsExecutingOperation` invocation is retried 
instead of being treated as "not executing".
   - A departed worker or a definitive `false` still fails the vertex 
immediately.
   - If every attempt fails, the existing restore fallback is unchanged.
   - An interrupt stops the retries and keeps the flag.
   
   The four tests in `PhysicalVertexTest` cover the cases from your plan.
   
   What is left:
   - **Real-case validation:** agreed that this is still the gap. The earlier 
confirmation on `89f91a89af15f33d97a942d8b65f4fc1b0c80bf5` also used a test 
double for the remote task, so neither the bug nor the fix has been shown 
against a real master failover, partition or GC pause. If you can run one, 
please post the engine logs around the probe and the resulting vertex state.
   - **PR description:** please state the mock-only caveat there as well, so 
reviewers see it.
   - **Code review:** I haven't gone through the code yet, and will do that on 
the PR itself.
   
   <!-- streview-comment:2453 -->


-- 
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