[ https://issues.apache.org/jira/browse/SPARK-9599?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14653723#comment-14653723 ]
Sean Owen edited comment on SPARK-9599 at 8/4/15 2:33 PM: ---------------------------------------------------------- For example, in the case of groupByKey, how would anything know a key mapped to many values before performing a shuffle anyway? EDIT: err, I mean a distributed count operation, which isn't trivial but not a full shuffle I suppose. Interesting, not sure if there are subtle reasons this is hard to work around or not, because now the very partitioning is a function of all values in the parent. was (Author: srowen): For example, in the case of groupByKey, how would anything know a key mapped to many values before performing a shuffle anyway? > Dynamic partitioning based on key-distribution > ---------------------------------------------- > > Key: SPARK-9599 > URL: https://issues.apache.org/jira/browse/SPARK-9599 > Project: Spark > Issue Type: Improvement > Components: Shuffle, Spark Core > Affects Versions: 1.4.1 > Reporter: Zoltán Zvara > > When - for example - using {{groupByKey}} operator with default > {{HashPartitioner}}, there might be a case when heavy keys get partitioned > into the same bucket, later raising an OOM error at the result partition. A > domain-based partitioner might not be able to help, when the outstanding > key-distribution changes from time to time (for example while dealing with > data streams). > Spark should identify these situations and change the partitioning > accordingly when a partitioning would raise an OOM later. -- 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