Github user zsxwing commented on a diff in the pull request:
https://github.com/apache/spark/pull/3505#discussion_r22774569
--- Diff: core/src/main/scala/org/apache/spark/rdd/SkewedJoinRDD.scala ---
@@ -0,0 +1,345 @@
+/*
+ * Licensed to the Apache Software Foundation (ASF) under one or more
+ * contributor license agreements. See the NOTICE file distributed with
+ * this work for additional information regarding copyright ownership.
+ * The ASF licenses this file to You under the Apache License, Version 2.0
+ * (the "License"); you may not use this file except in compliance with
+ * the License. You may obtain a copy of the License at
+ *
+ * http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+package org.apache.spark.rdd
+
+import java.io.{ObjectOutputStream, IOException}
+
+import scala.reflect.ClassTag
+
+import org.apache.spark._
+import org.apache.spark.serializer.Serializer
+import org.apache.spark.shuffle.ShuffleHandle
+import org.apache.spark.util.Utils
+import org.apache.spark.util.collection._
+
+private[spark] sealed trait JoinType[K, L, R, PAIR <: Product2[_, _]]
extends Serializable {
+
+ def flatten(i: Iterator[(K, (Iterable[Chunk[L]], Iterable[Chunk[R]]))]):
Iterator[(K, PAIR)]
+
+}
+
+private[spark] object JoinType {
+
+ def inner[K, L, R] = new JoinType[K, L, R, (L, R)] {
+
+ override def flatten(i: Iterator[(K, (Iterable[Chunk[L]],
Iterable[Chunk[R]]))]) =
+ i flatMap {
+ case (key, pair) => {
+ if (pair._1.size < pair._2.size) {
+ yieldPair(pair._1, pair._2, key, (p1: L, p2: R) => (p1, p2))
+ } else {
+ yieldPair(pair._2, pair._1, key, (p2: R, p1: L) => (p1, p2))
+ }
+ }
+ }
+ }
+
+ def leftOuter[K, L, R] = new JoinType[K, L, R, (L, Option[R])] {
+
+ override def flatten(i: Iterator[(K, (Iterable[Chunk[L]],
Iterable[Chunk[R]]))]) =
+ i flatMap {
+ case (key, pair) => {
+ if (pair._2.size == 0) {
+ for (chunk <- pair._1.iterator;
+ v <- chunk
+ ) yield (key, (v, None)): (K, (L, Option[R]))
+ }
+ else if (pair._1.size < pair._2.size) {
+ yieldPair(pair._1, pair._2, key, (p1: L, p2: R) => (p1,
Some(p2)))
+ } else {
+ yieldPair(pair._2, pair._1, key, (p2: R, p1: L) => (p1,
Some(p2)))
+ }
+ }
+ }
+ }
+
+ def rightOuter[K, L, R] = new JoinType[K, L, R, (Option[L], R)] {
+
+ override def flatten(i: Iterator[(K, (Iterable[Chunk[L]],
Iterable[Chunk[R]]))]) =
+ i flatMap {
+ case (key, pair) => {
+ if (pair._1.size == 0) {
+ for (chunk <- pair._2.iterator;
+ v <- chunk
+ ) yield (key, (None, v)): (K, (Option[L], R))
+ }
+ else if (pair._1.size < pair._2.size) {
+ yieldPair(pair._1, pair._2, key, (p1: L, p2: R) => (Some(p1),
p2))
+ } else {
+ yieldPair(pair._2, pair._1, key, (p2: R, p1: L) => (Some(p1),
p2))
+ }
+ }
+ }
+ }
+
+ def fullOuter[K, L, R] = new JoinType[K, L, R, (Option[L], Option[R])] {
+
+ override def flatten(i: Iterator[(K, (Iterable[Chunk[L]],
Iterable[Chunk[R]]))]) =
+ i flatMap {
+ case (key, pair) => {
+ if (pair._1.size == 0) {
+ for (chunk <- pair._2.iterator;
+ v <- chunk
+ ) yield (key, (None, Some(v))): (K, (Option[L], Option[R]))
+ }
+ else if (pair._2.size == 0) {
+ for (chunk <- pair._1.iterator;
+ v <- chunk
+ ) yield (key, (Some(v), None)): (K, (Option[L], Option[R]))
+ }
+ else if (pair._1.size < pair._2.size) {
+ yieldPair(pair._1, pair._2, key, (p1: L, p2: R) => (Some(p1),
Some(p2)))
+ } else {
+ yieldPair(pair._2, pair._1, key, (p2: R, p1: L) => (Some(p1),
Some(p2)))
+ }
+ }
+ }
+ }
+
+ private def yieldPair[K, OUT, IN, PAIR <: Product2[_, _]](
+ outer: Iterable[Chunk[OUT]], inner: Iterable[Chunk[IN]], key: K,
toPair: (OUT, IN) => PAIR) =
+ for (
+ outerChunk <- outer.iterator;
+ innerChunk <- inner.iterator;
+ outerValue <- outerChunk;
+ innerValue <- innerChunk
+ ) yield (key, toPair(outerValue, innerValue))
+}
+
+private[spark] sealed trait JoinSplitDep extends Serializable
+
+private[spark] case class NarrowJoinSplitDep(
+ rdd: RDD[_],
+ splitIndex: Int,
+ var split: Partition) extends JoinSplitDep {
+
+ @throws(classOf[IOException])
+ private def writeObject(oos: ObjectOutputStream): Unit =
Utils.tryOrIOException {
+ // Update the reference to parent split at the time of task
serialization
+ split = rdd.partitions(splitIndex)
+ oos.defaultWriteObject()
+ }
+}
+
+private[spark] case class ShuffleJoinSplitDep(handle: ShuffleHandle)
extends JoinSplitDep
+
+private[spark] class JoinPartition(idx: Int, val left: JoinSplitDep, val
right: JoinSplitDep)
+ extends Partition with Serializable {
+ override val index: Int = idx
+
+ override def hashCode(): Int = idx
+}
+
+private[spark] class BufferWrapper private(val isChunkBuffer: Boolean,
buffer: Iterable[Any])
+ extends Serializable {
+
+ def this(buffer: CompactBuffer[_]) {
+ this(false, buffer)
+ }
+
+ def this(buffer: ChunkBuffer[_]) {
+ this(true, buffer)
+ }
+
+ def getChunkBuffer[T: ClassTag]: ChunkBuffer[T] =
buffer.asInstanceOf[ChunkBuffer[T]]
+
+ def getCompactBuffer[T: ClassTag]: CompactBuffer[T] =
buffer.asInstanceOf[CompactBuffer[T]]
+
+ def asChunkIterable[T: ClassTag]: Iterable[Chunk[T]] = {
+ if (isChunkBuffer) {
+ getChunkBuffer[T]
+ }
+ else {
+ val buffer = getCompactBuffer[T]
+ if (buffer.isEmpty) {
+ Iterable[Chunk[T]]()
+ }
+ else {
+ Iterable(new Chunk[T](buffer))
+ }
+ }
+ }
+}
+
+private[spark] class SkewedJoinRDD[K, L, R, PAIR <: Product2[_, _]](
+ left: RDD[(K, L)], right: RDD[(K, R)], part: Partitioner, joinType:
JoinType[K, L, R, PAIR])
+ (implicit kt: ClassTag[K], lt: ClassTag[L], rt: ClassTag[R], ord:
Ordering[K])
--- End diff --
I think `skewedGroupby` also requires the key is ordering. Otherwise, using
a hash comparator, because the hash conflicts, the may require loading all
values of a key into memory in some special situation.
E.g., assume k1 and k2 have the same hashcode, if using a hash comparator,
the data in two external files may be (assume value is 1):
file1: (k1, 1), (k2, 1), (otherKey, 1), ...
file2: (k2, 1), (k2, 1), (k2, 1), ... ,.... (k1, 1), (otherKey, 1), ...
In file2, k1 is behind all `k2` (We use a hash comparator, so it's
possible). Becuase we scan file1 at first, we will process `k1` before `k2`.
But when processing k1 in `file2`, we need to load all `k2`'s values in
`file2`, it may cause OOM.
If the key is ordering, we can avoid this extreme case.
---
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.
---
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
