Details
-
Improvement
-
Status: Resolved
-
Major
-
Resolution: Won't Fix
-
3.3.0
-
None
-
None
Description
Related to MAPREDUCE-7282 a variant solution
- we add a new method for committers to declare whether they can recover from a task attempt commit failure
- default = true; v2 and (external) EMR spark committers can return false
execution engine - MR, Spark, can look at this after a task attempt fails to commit and decide what to do
recoverable: execute/commit another task attempt
non-recoverable, one of (Configured)
- warn and continue
- abort the job
with the job abort option, users would be confident that if a failure happened during the commit phase, they'd know about it and choose how to recover.
I'd use a long/unusual name, so that in, say, Spark, reflection to could be used to find and call the method & so compile against older releases
Attachments
Issue Links
- is related to
-
SPARK-33019 Use spark.hadoop.mapreduce.fileoutputcommitter.algorithm.version=1 by default
- Resolved
- is superceded by
-
MAPREDUCE-7341 Add a task-manifest output committer for Azure and GCS
- Resolved
- relates to
-
MAPREDUCE-7282 MR v2 commit algorithm should be deprecated and not the default
- Resolved