Github user cloud-fan commented on the issue: https://github.com/apache/spark/pull/22112 Update, according to the discussion in https://github.com/apache/spark/pull/9214 , the current behavior of shuffle writing is: "first write wins". We can't simply change it to "last write wins", as we may have concurrent read-write issues. To switch to "last write wins", we have to finish https://github.com/apache/spark/pull/6648 first. Since it's not realistic to complete https://github.com/apache/spark/pull/6648 before Spark 2.4, in this PR I fail the job directly if we hit a fetch failure and the preceding map stage is indeterminate. The error message asks users to do checkpoint to avoid this issue.
--- --------------------------------------------------------------------- To unsubscribe, e-mail: reviews-unsubscr...@spark.apache.org For additional commands, e-mail: reviews-h...@spark.apache.org