advancedxy 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_r319179534
##########
File path:
core/src/main/scala/org/apache/spark/shuffle/sort/SortShuffleManager.scala
##########
@@ -79,9 +81,9 @@ private[spark] class SortShuffleManager(conf: SparkConf)
extends ShuffleManager
}
/**
- * A mapping from shuffle ids to the number of mappers producing output for
those shuffles.
+ * A mapping from shuffle ids to the task ids of mappers producing output
for those shuffles.
*/
- private[this] val numMapsForShuffle = new ConcurrentHashMap[Int, Int]()
+ private[this] val taskIdMapsForShuffle = new ConcurrentHashMap[Int,
ArrayBuffer[Long]]()
Review comment:
Well, @xuanyuanking should already be aware of that 100K Mappers is not that
rare for large production jobs. That would be ~10MB for single one map stage.
Maybe we should removes old shuffleId's data just like the scheduler removes
old stages. However I do believe it's fine for now. Let's revise this when it
actually hits.
----------------------------------------------------------------
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:
[email protected]
With regards,
Apache Git Services
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
