[ https://issues.apache.org/jira/browse/SPARK-6104?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14342869#comment-14342869 ]
Reynold Xin commented on SPARK-6104: ------------------------------------ Yes, large aggregation with auto spilling to disk is on the roadmap for Spark SQL. I'm going to close the ticket because this ticket is not actionable. I will have another one tracking this. cc [~marmbrus] > spark SQL shuffle OOM > --------------------- > > Key: SPARK-6104 > URL: https://issues.apache.org/jira/browse/SPARK-6104 > Project: Spark > Issue Type: Improvement > Affects Versions: 1.2.0 > Reporter: Hong Shen > > Currently, spark shuffle can use ExternalAppendOnlyMap to combine data that > have fetched, but it depend on ShuffleDependency's aggregator, while spark > SQL's shuffle have not define aggregator, so it will easily cause OOM in > large scale of data. > In our cluster, we have used spark SQL, but it often appear OOM, it's hard to > reset spark.default.parallelism every day. I think we should add aggregator > for spark SQL shuffle. -- This message was sent by Atlassian JIRA (v6.3.4#6332) --------------------------------------------------------------------- To unsubscribe, e-mail: issues-unsubscr...@spark.apache.org For additional commands, e-mail: issues-h...@spark.apache.org