[SPARK-25341] Support rolling back a shuffle map stage and re-generate the shuffle files

• Type: Improvement
• Status: Resolved
• Priority: Major
• Resolution: Fixed
• Affects Version/s: 3.0.0
• Fix Version/s: 3.0.0
• Component/s: Spark Core
• Assignee: Yuanjian Li
• Reporter: Wenchen Fan
• Created: 05/Sep/18 17:26
• Updated: 30/Jun/20 08:05
• Resolved: 23/Sep/19 08:17

Description

This is a follow up of https://issues.apache.org/jira/browse/SPARK-23243

To completely fix that problem, Spark needs to be able to rollback a shuffle map stage and rerun all the map tasks.

According to https://github.com/apache/spark/pull/9214 , Spark doesn't support it currently, as in shuffle writing "first write wins".

Since overwriting shuffle files is hard, we can extend the shuffle id to include a "shuffle generation number". Then the reduce task can specify which generation of shuffle it wants to read. https://github.com/apache/spark/pull/6648 seems in the right direction.

Issue Links

• blocks: SPARK-28845 Enable spark.sql.execution.sortBeforeRepartition only for retried stages
• causes: SPARK-32124 [SHS] Failed to parse FetchFailed TaskEndReason from event log produce by Spark 2.4
• is related to: SPARK-25342 Support rolling back a result stage
• relates to: SPARK-23243 Shuffle+Repartition on an RDD could lead to incorrect answers
• links to: GitHub Pull Request #24110
• links to: GitHub Pull Request #24892
• links to: GitHub Pull Request #25620