Details
-
Bug
-
Status: Open
-
Minor
-
Resolution: Unresolved
-
None
-
None
-
None
-
None
Description
Sometimes, mappers restart themselves for some reason (there are several reasons, for example, because it has lost access to the data it generated which a reducer may later need.). However, not all reducers will depend on that mapper. Accordingly, jobs can get into a state in which all reducers have completed but some mappers are still running. This is a bug--when the reducers have completed, the job should be marked as completed immediately and any trailing map tasks killed.