deal with partitioner when defaultParallelism is explict set.

Author: jiangxb1987

core/src/main/scala/org/apache/spark/Partitioner.scala

@@ -43,17 +43,19 @@ object Partitioner {
   /**
    * Choose a partitioner to use for a cogroup-like operation between a number of RDDs.
    *
-   * If any of the RDDs already has a partitioner, and the number of partitions of the
-   * partitioner is either greater than or is less than and within a single order of
-   * magnitude of the max number of upstream partitions, choose that one.
+   * If spark.default.parallelism is set, we'll use the value of SparkContext defaultParallelism
+   * as the default partitions number, otherwise we'll use the max number of upstream partitions.
    *
-   * Otherwise, we use a default HashPartitioner. For the number of partitions, if
-   * spark.default.parallelism is set, then we'll use the value from SparkContext
-   * defaultParallelism, otherwise we'll use the max number of upstream partitions.
+   * If any of the RDDs already has a partitioner, and the partitioner is an eligible one (with a
+   * partitions number that is not less than the max number of upstream partitions by an order of
+   * magnitude), or the number of partitions is larger than the default one, we'll choose the
+   * exsiting partitioner.
    *
-   * Unless spark.default.parallelism is set, the number of partitions will be the
-   * same as the number of partitions in the largest upstream RDD, as this should
-   * be least likely to cause out-of-memory errors.
+   * Otherwise, we'll use a new HashPartitioner with the default partitions number.
+   *
+   * Unless spark.default.parallelism is set, the number of partitions will be the same as the
+   * number of partitions in the largest upstream RDD, as this should be least likely to cause
+   * out-of-memory errors.
    *
    * We use two method parameters (rdd, others) to enforce callers passing at least 1 RDD.
    */
@@ -67,31 +69,32 @@ object Partitioner {
       None
     }
 
-    if (isEligiblePartitioner(hasMaxPartitioner, rdds)) {
+    val defaultNumPartitions = if (rdd.context.conf.contains("spark.default.parallelism")) {
+      rdd.context.defaultParallelism
+    } else {
+      rdds.map(_.partitions.length).max
+    }
+
+    // If the existing max partitioner is an eligible one, or its partitions number is larger
+    // than the default number of partitions, use the existing partitioner.
+    if (hasMaxPartitioner.nonEmpty && (isEligiblePartitioner(hasMaxPartitioner.get, rdds) ||
+        defaultNumPartitions < hasMaxPartitioner.get.getNumPartitions)) {
       hasMaxPartitioner.get.partitioner.get
     } else {
-      if (rdd.context.conf.contains("spark.default.parallelism")) {
-        new HashPartitioner(rdd.context.defaultParallelism)
-      } else {
-        new HashPartitioner(rdds.map(_.partitions.length).max)
-      }
+      new HashPartitioner(defaultNumPartitions)
     }
   }
 
   /**
-   * Returns true if the number of partitions of the RDD is either greater
-   * than or is less than and within a single order of magnitude of the
-   * max number of upstream partitions;
-   * otherwise, returns false
+   * Returns true if the number of partitions of the RDD is either greater than or is less than and
+   * within a single order of magnitude of the max number of upstream partitions, otherwise returns
+   * false.
    */
   private def isEligiblePartitioner(
-     hasMaxPartitioner: Option[RDD[_]],
+     hasMaxPartitioner: RDD[_],
      rdds: Seq[RDD[_]]): Boolean = {
-    if (hasMaxPartitioner.isEmpty) {
-      return false
-    }
     val maxPartitions = rdds.map(_.partitions.length).max
-    log10(maxPartitions) - log10(hasMaxPartitioner.get.getNumPartitions) < 1
+    log10(maxPartitions) - log10(hasMaxPartitioner.getNumPartitions) < 1
   }
 }
 

core/src/test/scala/org/apache/spark/PartitioningSuite.scala

@@ -284,7 +284,38 @@ class PartitioningSuite extends SparkFunSuite with SharedSparkContext with Priva
     assert(partitioner3.numPartitions == rdd3.getNumPartitions)
     assert(partitioner4.numPartitions == rdd3.getNumPartitions)
     assert(partitioner5.numPartitions == rdd4.getNumPartitions)
+  }
 
+  test("defaultPartitioner when defaultParallelism is set") {
+    assert(!sc.conf.contains("spark.default.parallelism"))
+    try {
+      sc.conf.set("spark.default.parallelism", "4")
+
+      val rdd1 = sc.parallelize((1 to 1000).map(x => (x, x)), 150)
+      val rdd2 = sc.parallelize(Array((1, 2), (2, 3), (2, 4), (3, 4)))
+        .partitionBy(new HashPartitioner(10))
+      val rdd3 = sc.parallelize(Array((1, 6), (7, 8), (3, 10), (5, 12), (13, 14)))
+        .partitionBy(new HashPartitioner(100))
+      val rdd4 = sc.parallelize(Array((1, 2), (2, 3), (2, 4), (3, 4)))
+        .partitionBy(new HashPartitioner(9))
+      val rdd5 = sc.parallelize((1 to 10).map(x => (x, x)), 11)
+
+      val partitioner1 = Partitioner.defaultPartitioner(rdd1, rdd2)
+      val partitioner2 = Partitioner.defaultPartitioner(rdd2, rdd3)
+      val partitioner3 = Partitioner.defaultPartitioner(rdd3, rdd1)
+      val partitioner4 = Partitioner.defaultPartitioner(rdd1, rdd2, rdd3)
+      val partitioner5 = Partitioner.defaultPartitioner(rdd4, rdd5)
+      val partitioner6 = Partitioner.defaultPartitioner(rdd5, rdd5)
+
+      assert(partitioner1.numPartitions == rdd2.getNumPartitions)
+      assert(partitioner2.numPartitions == rdd3.getNumPartitions)
+      assert(partitioner3.numPartitions == rdd3.getNumPartitions)
+      assert(partitioner4.numPartitions == rdd3.getNumPartitions)
+      assert(partitioner5.numPartitions == rdd4.getNumPartitions)
+      assert(partitioner6.numPartitions == sc.defaultParallelism)
+    } finally {
+      sc.conf.remove("spark.default.parallelism")
+    }
   }
 }
 

core/src/test/scala/org/apache/spark/rdd/PairRDDFunctionsSuite.scala

@@ -322,8 +322,7 @@ class PairRDDFunctionsSuite extends SparkFunSuite with SharedSparkContext {
   }
 
   // See SPARK-22465
-  test("cogroup between multiple RDD" +
-    " with number of partitions similar in order of magnitude") {
+  test("cogroup between multiple RDD with number of partitions similar in order of magnitude") {
     val rdd1 = sc.parallelize((1 to 1000).map(x => (x, x)), 20)
     val rdd2 = sc
       .parallelize(Array((1, 1), (1, 2), (2, 1), (3, 1)))
@@ -332,6 +331,48 @@ class PairRDDFunctionsSuite extends SparkFunSuite with SharedSparkContext {
     assert(joined.getNumPartitions == rdd2.getNumPartitions)
   }
 
+  test("cogroup between multiple RDD when defaultParallelism is set without proper partitioner") {
+    assert(!sc.conf.contains("spark.default.parallelism"))
+    try {
+      sc.conf.set("spark.default.parallelism", "4")
+      val rdd1 = sc.parallelize((1 to 1000).map(x => (x, x)), 20)
+      val rdd2 = sc.parallelize(Array((1, 1), (1, 2), (2, 1), (3, 1)), 10)
+      val joined = rdd1.cogroup(rdd2)
+      assert(joined.getNumPartitions == sc.defaultParallelism)
+    } finally {
+      sc.conf.remove("spark.default.parallelism")
+    }
+  }
+
+  test("cogroup between multiple RDD when defaultParallelism is set with proper partitioner") {
+    assert(!sc.conf.contains("spark.default.parallelism"))
+    try {
+      sc.conf.set("spark.default.parallelism", "4")
+      val rdd1 = sc.parallelize((1 to 1000).map(x => (x, x)), 20)
+      val rdd2 = sc.parallelize(Array((1, 1), (1, 2), (2, 1), (3, 1)))
+        .partitionBy(new HashPartitioner(10))
+      val joined = rdd1.cogroup(rdd2)
+      assert(joined.getNumPartitions == rdd2.getNumPartitions)
+    } finally {
+      sc.conf.remove("spark.default.parallelism")
+    }
+  }
+
+  test("cogroup between multiple RDD when defaultParallelism is set with huge number of " +
+    "partitions from upstream RDDs") {
+    assert(!sc.conf.contains("spark.default.parallelism"))
+    try {
+      sc.conf.set("spark.default.parallelism", "4")
+      val rdd1 = sc.parallelize((1 to 1000).map(x => (x, x)), 1000)
+      val rdd2 = sc.parallelize(Array((1, 1), (1, 2), (2, 1), (3, 1)))
+        .partitionBy(new HashPartitioner(10))
+      val joined = rdd1.cogroup(rdd2)
+      assert(joined.getNumPartitions == rdd2.getNumPartitions)
+    } finally {
+      sc.conf.remove("spark.default.parallelism")
+    }
+  }
+
   test("rightOuterJoin") {
     val rdd1 = sc.parallelize(Array((1, 1), (1, 2), (2, 1), (3, 1)))
     val rdd2 = sc.parallelize(Array((1, 'x'), (2, 'y'), (2, 'z'), (4, 'w')))