cloud-fan commented on a change in pull request #25620: [SPARK-25341][Core]
Support rolling back a shuffle map stage and re-generate the shuffle files
URL: https://github.com/apache/spark/pull/25620#discussion_r319121193
##########
File path: core/src/main/scala/org/apache/spark/scheduler/DAGScheduler.scala
##########
@@ -1607,11 +1616,18 @@ private[spark] class DAGScheduler(
case mapStage: ShuffleMapStage =>
val numMissingPartitions =
mapStage.findMissingPartitions().length
if (numMissingPartitions < mapStage.numTasks) {
- // TODO: support to rollback shuffle files.
- // Currently the shuffle writing is "first write wins",
so we can't re-run a
- // shuffle map stage and overwrite existing shuffle
files. We have to finish
- // SPARK-8029 first.
- abortStage(mapStage, generateErrorMessage(mapStage),
None)
+ if
(sc.getConf.get(config.SHUFFLE_USE_OLD_FETCH_PROTOCOL)) {
Review comment:
what about general shuffles? It looks to me that the shuffle reader needs to
keep both the legacy int map id and the new long map id. When
`SHUFFLE_USE_OLD_FETCH_PROTOCOL` is true, construct the new shuffle request
with the new long map id. Otherwise, construct the old shuffle request with the
legacy int map id.
----------------------------------------------------------------
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
For queries about this service, please contact Infrastructure at:
us...@infra.apache.org
With regards,
Apache Git Services
---------------------------------------------------------------------
To unsubscribe, e-mail: reviews-unsubscr...@spark.apache.org
For additional commands, e-mail: reviews-h...@spark.apache.org
