cloud-fan commented on issue #24892: [SPARK-25341][Core] Support rolling back a shuffle map stage and re-generate the shuffle files URL: https://github.com/apache/spark/pull/24892#issuecomment-521499643 @vanzin I think your concern is valid. Seems the shuffle writing policy is contradictory to itself: if a partition has multiple shuffle write tasks, 1) if they are on the same node (e.g. a stage becomes zombie and rerun), first write wins. 2) if they are on different nodes (speculative tasks), last write wins. I think we should stick with "first write wins". As you said we should only update the map status if the task is the first one that completes.
---------------------------------------------------------------- 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]
