ump45nose opened a new pull request, #2698: URL: https://github.com/apache/datafusion-ballista/pull/2698
# Which issue does this PR close? Closes #2691. # Rationale for this change #2689 refunds vcore reservations when `launch_tasks` cannot get executor metadata, but the tasks in that batch were already recorded as `Running` at bind time. They were never launched, never reported, and never reset, so the job hung until the executor was lost. Because the refund sets `if_revive`, a repeated miss also binds more partitions onto the same stranded path. `pop_next_task`'s contract already says a bound-but-unlaunched task must have its status reset so it can be scheduled elsewhere; this fills that gap for the metadata-miss branch. # What changes are included in this PR? - Add `ExecutionGraph::reset_unlaunched_task` (static + AQE) that returns a still-`Running` task_info slot to pending via `reset_task_info`. - On the metadata-lookup failure path in `SchedulerState::launch_tasks`, call `TaskManager::reset_unlaunched_tasks` before refunding reservations. - Extend `launch_tasks_refunds_vcores_when_executor_metadata_missing` to assert pending task count is restored after the miss. # Are there any user-facing changes? No. --- This PR was prepared with AI assistance; I reviewed the change and ran the checks above. -- 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]
