Details
-
Improvement
-
Status: Resolved
-
Major
-
Resolution: Fixed
-
None
Description
Currently, Celeborn does not handle barrier stages in spark.
Barrier stages have an all-or-none property: so any task failure in the stage results in reexecution of all tasks : whether this is due to failure of current stage execution, or due to re-execution of a parent (barrier) stage due to downstream fetch failures.
Other bugs which are related: CELEBORN-1496, CELEBORN-1498
Attachments
Issue Links
- links to