Details
-
Bug
-
Status: Resolved
-
Major
-
Resolution: Fixed
-
None
Description
When an executor is removed, tasks on other executors trying to fetch data from the lost executor run into the INPUT_READ_FAILURE. Because parent task(s) that produced the blocks lost along with the executor are guaranteed to be retried via the EXECUTOR_REMOVED event, we can simply retry just the tasks that ran into INPUT_READ_FAILURE.