You are viewing a plain text version of this content. The canonical link for it is here.
Posted to issues@spark.apache.org by "Yuanjian Li (Jira)" <ji...@apache.org> on 2019/09/26 09:51:00 UTC
[jira] [Resolved] (SPARK-28845) Enable
spark.sql.execution.sortBeforeRepartition only for retried stages
[ https://issues.apache.org/jira/browse/SPARK-28845?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]
Yuanjian Li resolved SPARK-28845.
---------------------------------
Resolution: Won't Do
After further investigation, I found the objective of performing sort only for the retried indeterminate stage is unable to achieve. That will break our assumption for the `outputDeterministicLevel` for each RDD, which should be defined when the job submitted. While here we expected the output deterministic level depends on the stage attempt number.
As the SPARK-25341 completed, the current behavior depends on the config `spark.sql.execution.sortBeforeRepartition`. Spark will sort before repartition, and rerun the failed tasks when setting the config to true. On the contrary, the whole stage will rerun.
> Enable spark.sql.execution.sortBeforeRepartition only for retried stages
> ------------------------------------------------------------------------
>
> Key: SPARK-28845
> URL: https://issues.apache.org/jira/browse/SPARK-28845
> Project: Spark
> Issue Type: Improvement
> Components: Spark Core, SQL
> Affects Versions: 3.0.0
> Reporter: Yuanjian Li
> Priority: Major
>
> For fixing the correctness bug of SPARK-28699, we disable radix sortĀ for the scenario of repartition in Spark SQL. This will cause a performance regression.
> So for limiting the performance overhead, we'll do the optimizing work by only enable sort for the repartition operation while stage retries happening. This work depends on SPARK-25341.
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
---------------------------------------------------------------------
To unsubscribe, e-mail: issues-unsubscribe@spark.apache.org
For additional commands, e-mail: issues-help@spark.apache.org