[SPARK-23243] Shuffle+Repartition on an RDD could lead to incorrect answers

• Type: Bug
• Status: Resolved
• Priority: Blocker
• Resolution: Fixed
• Affects Version/s: 1.6.0, 2.0.0, 2.1.0, 2.2.0, 2.3.0
• Fix Version/s: 2.2.3, 2.3.2, 2.4.0
• Component/s: Spark Core
• Labels: correctness
• Target Version/s: 2.4.0
• Assignee: Wenchen Fan
• Reporter: Xingbo Jiang
• Created: 26/Jan/18 23:00
• Updated: 11/Sep/19 19:05
• Resolved: 05/Sep/18 22:37

Description

The RDD repartition also uses the round-robin way to distribute data, this can also cause incorrect answers on RDD workload the similar way as in https://issues.apache.org/jira/browse/SPARK-23207

The approach that fixes DataFrame.repartition() doesn't apply on the RDD repartition issue, as discussed in https://github.com/apache/spark/pull/20393#issuecomment-360912451

We track for alternative solutions for this issue in this task.

Issue Links

• is duplicated by: SPARK-25156 Same query returns different result
• is related to: SPARK-28699 Cache an indeterminate RDD could lead to incorrect result while stage rerun
• is related to: SPARK-25342 Support rolling back a result stage
• is related to: SPARK-25341 Support rolling back a shuffle map stage and re-generate the shuffle files
• relates to: SPARK-23207 Shuffle+Repartition on an DataFrame could lead to incorrect answers
• relates to: SPARK-29042 Sampling-based RDD with unordered input should be INDETERMINATE
• links to: [Github] Pull Request #20414 (jiangxb1987)
• links to: [Github] Pull Request #21698 (jiangxb1987)
• links to: [Github] Pull Request #22112 (cloud-fan)
• links to: [Github] Pull Request #22354 (cloud-fan)
• links to: [Github] Pull Request #22382 (bersprockets)
• links to: GitHub Pull Request #25755