Details
-
Bug
-
Status: Open
-
Major
-
Resolution: Unresolved
-
2.0.2-alpha, 0.23.4
-
None
-
None
Description
When the ApplicationMaster starts up and generates (and potentially recovers) AMInfo records for the history file, those events should be flushed to disk in case the AM crashes badly. Failure to do so means the AMInfos could be lost and the previous AM attempts lost upon recovery. See discussion in MAPREDUCE-4729.
Attachments
Issue Links
- relates to
-
MAPREDUCE-4729 job history UI not showing all job attempts
- Closed