Github user mateiz commented on a diff in the pull request:
https://github.com/apache/spark/pull/1562#discussion_r15426437
--- Diff: core/src/test/scala/org/apache/spark/PartitioningSuite.scala ---
@@ -102,6 +100,34 @@ class PartitioningSuite extends FunSuite with
SharedSparkContext with PrivateMet
partitioner.getPartition(Row(100))
}
+ test("RangePartitioner should run only one job if data is roughly
balanced") {
+ val rdd = sc.makeRDD(0 until 20, 20).flatMap { i =>
+ val random = new java.util.Random(i)
+ Iterator.fill(5000 * i)((random.nextDouble() + i, i))
+ }.cache()
+ for (numPartitions <- Seq(10, 20, 40)) {
+ val partitioner = new RangePartitioner(numPartitions, rdd)
+ assert(partitioner.numPartitions === numPartitions)
+ assert(partitioner.singlePass === true)
+ val counts = rdd.keys.map(key =>
partitioner.getPartition(key)).countByValue().values
+ assert(counts.max < 2.0 * counts.min)
+ }
+ }
+
+ test("RangePartitioner should work well on unbalanced data") {
+ val rdd = sc.makeRDD(0 until 20, 20).flatMap { i =>
+ val random = new java.util.Random(i)
+ Iterator.fill(20 * i * i * i)((random.nextDouble() + i, i))
+ }.cache()
+ for (numPartitions <- Seq(2, 4, 8)) {
+ val partitioner = new RangePartitioner(numPartitions, rdd)
+ assert(partitioner.numPartitions === numPartitions)
+ assert(partitioner.singlePass === false)
+ val counts = rdd.keys.map(key =>
partitioner.getPartition(key)).countByValue().values
+ assert(counts.max < 2.0 * counts.min)
+ }
+ }
+
--- End diff --
Can you add some tests where the whole RDD has 0 elements, and some tests
where individual partitions have 0 elements and others have more? That's where
divide by zero errors can happen.
---
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 [email protected] or file a JIRA ticket
with INFRA.
---
