martin-g opened a new pull request, #2242:
URL: https://github.com/apache/datafusion-ballista/pull/2242

   # Which issue does this PR close?
   
   Closes #2226
   
   # Rationale for this change
   
   Emit `ExecutorLost` event after failed `state.launch_tasks()` call.
   
   # What changes are included in this PR?
   
   `launch_tasks` removes failed executors via
   `SchedulerState::remove_executor`, which resets affected stages but does NOT 
post an `ExecutorLost` event. The grace-period timer that fails jobs on an 
empty cluster is only armed inside the `ExecutorLost` event handler. If 
`launch_tasks` is the path that removes the last executor (e.g. AQE plans a new 
stage right as executors die), no `ExecutorLost` event is ever posted, the 
grace timer never fires, and the job hangs forever.
   
   Post `ExecutorLost` for each failed executor so the event handler's 
empty-cluster check runs. Calling
   `task_manager.executor_lost()` twice for the same executor is safe 
(idempotent: the second call finds nothing to reset).
   
   
   # Are there any user-facing changes?
   
   No


-- 
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