Github user marmbrus commented on a diff in the pull request: https://github.com/apache/spark/pull/1134#discussion_r14678203 --- Diff: sql/core/src/main/scala/org/apache/spark/sql/execution/joins.scala --- @@ -400,3 +401,73 @@ case class BroadcastNestedLoopJoin( streamedPlusMatches.flatMap(_._1), sqlContext.sparkContext.makeRDD(rightOuterMatches)) } } + +/** + * :: DeveloperApi :: + * In some case ,data skew happens.SkewJoin sample the table rdd to find the largest key, + * then make the largest key rows as a table rdd.The left rdd will be made leftSkewedtable + * rdd without the largest key and the maxkeyskewedtable rdd with the largest key. + * Then,join the two table with the righttable. + * Finally,union the two result rdd. + */ +@DeveloperApi +case class SkewJoinCartesianProduct( + left: SparkPlan, + right: SparkPlan, + condition: Option[Expression])(@transient sc: SparkContext) extends BinaryNode { + override def output = left.output ++ right.output + + @transient lazy val boundCondition = + InterpretedPredicate( + condition + .map(c => BindReferences.bindReference(c, left.output ++ right.output)) + .getOrElse(Literal(true))) + + def execute() = { + + val skewedTable = left.execute() + //This will later write as configuration + val sample = skewedTable.sample(false, 0.3, 9).collect() + val sortedSample = sample.sortWith((row1, row2) => row1.hashCode() > row2.hashCode()) + var max = 0 + var num = sample.size - 1 + var temp = 0 + var maxrowKey = sortedSample(0) + //find the largest key + if (sortedSample.size > 1) { + for (i <- 1 to num) { + if (sortedSample(i - 1) == sortedSample(i)) { + temp += 1 + } + else { + if (temp > max) { + max = temp + maxrowKey = sortedSample(i - 1) + } + temp = 0 + } + } + } + // I also Send a Pull Request as RDD API span fun to offer the function + //split a RDD as two through span function + // .https://github.com/apache/spark/pull/1306 + // val (maxKeySkewedTable, mainSkewedTable) = skewedTable.span(row => { + // skewSideKeyGenerator(row).toString().equals(maxrowKey.toString()) + // }) + val maxKeySkewedTable = skewedTable.filter(row => { + row.toString().equals(maxrowKey.toString()) --- End diff -- Why are you comparing the strings of the rows instead of the rows themselves?
--- If your project is set up for it, you can reply to this email and have your reply appear on GitHub as well. If your project does not have this feature enabled and wishes so, or if the feature is enabled but not working, please contact infrastructure at infrastruct...@apache.org or file a JIRA ticket with INFRA. ---