DanielLeens commented on issue #13325:
URL: https://github.com/apache/seatunnel/issues/13325#issuecomment-6983659842

   I reproduced monitor blocking through the actual 
`SubPlan.updatePipelineState → stateProcess → PhysicalVertex.cancel` chain on 
`89f91a89af15f33d97a942d8b65f4fc1b0c80bf5` (Java 11.0.27).
   
   Steps:
   1. Construct a real SubPlan and PhysicalVertex, with the worker present in 
membership and the execution probe returning true.
   2. Make the `CancelTaskOperation` invocation future fail with 
`ExecutionException`, while the worker remains present and the task future 
remains incomplete.
   3. On one thread call `SubPlan.updatePipelineState(CANCELING)`. Wait for two 
cancel attempts through the real retry loop.
   4. Start another thread that tries to enter `synchronized(subPlan)`. It is 
`BLOCKED`, while the cancel-state thread is alive and task completion is still 
pending.
   5. Control: change only membership lookup to report the worker absent. The 
cancellation thread exits, the monitor waiter proceeds, and the vertex becomes 
`CANCELED`.
   
   Observed:
   ```text
   attempts=2, pipeline-monitor-waiter=BLOCKED, cancel thread alive=true
   declaring worker absent releases monitors; vertex=CANCELED
   ```
   
   [Executable Java probe and exact build/run 
commands](https://github.com/apache/seatunnel/issues/13314#issuecomment-6983602245):
 see `cancelMonitors()`.
   
   This is a bounded, deterministic invocation-failure test. It demonstrates 
the monitor-holding retry behavior, not a measured minutes-long GC pause or an 
experimentally infinite loop in a live cluster.
   


-- 
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