DanielLeens commented on issue #13317:
URL: https://github.com/apache/seatunnel/issues/13317#issuecomment-6983715635

   I reproduced acceptance of an older terminal event and release of the 
**current** slot through actual production methods at 
`89f91a89af15f33d97a942d8b65f4fc1b0c80bf5` (Java 11.0.27).
   
   Steps:
   1. Construct a PhysicalVertex in `FAILED`, and retain a 
`TaskExecutionState(location, FAILED, "old attempt terminal event")` for that 
generation.
   2. Invoke real `reset()` (`FAILED → CREATED`), then advance the vertex to 
`RUNNING` to represent a new attempt at the same `TaskGroupLocation`.
   3. Configure the JobMaster's owned-slot map with a current slot for that 
location.
   4. Deliver the saved old report through real 
`JobMaster.updateTaskExecutionState()` and real `releaseTaskGroupResource()`.
   5. Assert that the new vertex becomes `FAILED` and the mocked 
ResourceManager receives a release request containing the **current slot**, not 
a separately identified old slot.
   
   [Executable probe and exact build/run 
commands](https://github.com/apache/seatunnel/issues/13316#issuecomment-6983695881):
 see `staleEvent()` in `LifecycleRepros.java`.
   
   Scope: in-process event injection with actual 
state-transition/report-routing/release methods. I did **not** simulate a 
worker rejoin, real delayed network delivery, full redeployment, or actual slot 
reassignment. This confirms what happens **if** an old terminal report arrives 
after the state reset; it does not measure how often that delivery ordering 
occurs in a 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