Skip to content

[SPARK-21782][Core] Repartition creates skews when numPartitions is a power of 2 #18990

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Closed
megaserg wants to merge 2 commits into from
Closed

[SPARK-21782][Core] Repartition creates skews when numPartitions is a power of 2 #18990

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
2cb7550
[SPARK-21782][Core] Repartition creates skews when numPartitions is a…
megaserg Aug 18, 2017
bee7fca
Scramble seed bits instead of using non-deterministic ctor
megaserg Aug 18, 2017
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
3 changes: 2 additions & 1 deletion core/src/main/scala/org/apache/spark/rdd/RDD.scala
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -24,6 +24,7 @@ import scala.collection.mutable.ArrayBuffer
import scala.io.Codec
import scala.language.implicitConversions
import scala.reflect.{classTag, ClassTag}
import scala.util.hashing

import com.clearspring.analytics.stream.cardinality.HyperLogLogPlus
import org.apache.hadoop.io.{BytesWritable, NullWritable, Text}
Expand Down Expand Up @@ -448,7 +449,7 @@ abstract class RDD[T: ClassTag](
if (shuffle) {
/** Distributes elements evenly across output partitions, starting from a random partition. */
val distributePartition = (index: Int, items: Iterator[T]) => {
var position = (new Random(index)).nextInt(numPartitions)
var position = (new Random(hashing.byteswap32(index))).nextInt(numPartitions)
items.map { t =>
// Note that the hash code of the key will just be the key itself. The HashPartitioner
// will mod it with the number of total partitions.
Expand Down
6 changes: 4 additions & 2 deletions core/src/test/scala/org/apache/spark/rdd/RDDSuite.scala
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -347,16 +347,18 @@ class RDDSuite extends SparkFunSuite with SharedSparkContext {
val partitions = repartitioned.glom().collect()
// assert all elements are present
assert(repartitioned.collect().sortWith(_ > _).toSeq === input.toSeq.sortWith(_ > _).toSeq)
// assert no bucket is overloaded
// assert no bucket is overloaded or empty
for (partition <- partitions) {
val avg = input.size / finalPartitions
val maxPossible = avg + initialPartitions
assert(partition.length <= maxPossible)
assert(partition.length <= maxPossible)
assert(!partition.isEmpty)
}
}

testSplitPartitions(Array.fill(100)(1), 10, 20)
testSplitPartitions(Array.fill(10000)(1) ++ Array.fill(10000)(2), 20, 100)
testSplitPartitions(Array.fill(1000)(1), 250, 128)
}

test("coalesced RDDs") {
Expand Down