Spark快速上手(4)Spark核心编程-Spark分区器(Partitioner)@(RDD-K_V) - ToB企服应用市场:ToB评测及商务社交产业平台

class RangePartitioner[K : Ordering : ClassTag, V](
partitions: Int,
rdd: RDD[_ <: Product2[K, V]],
private var ascending: Boolean = true,
val samplePointsPerPartitionHint: Int = 20)
extends Partitioner {
// A constructor declared in order to maintain backward compatibility for Java, when we add the
// 4th constructor parameter samplePointsPerPartitionHint. See SPARK-22160.
// This is added to make sure from a bytecode point of view, there is still a 3-arg ctor.
def this(partitions: Int, rdd: RDD[_ <: Product2[K, V]], ascending: Boolean) = {
this(partitions, rdd, ascending, samplePointsPerPartitionHint = 20)
}
// We allow partitions = 0, which happens when sorting an empty RDD under the default settings.
require(partitions >= 0, s"Number of partitions cannot be negative but found $partitions.")
require(samplePointsPerPartitionHint > 0,
s"Sample points per partition must be greater than 0 but found $samplePointsPerPartitionHint")
// 获取RDD中key类型数据的排序器
private var ordering = implicitly[Ordering[K]]
// An array of upper bounds for the first (partitions - 1) partitions
private var rangeBounds: Array[K] = {
if (partitions <= 1) {
// 如果给定的分区数是一个的情况下，直接返回一个空的集合，表示数据不进行分区
Array.empty
} else {
// This is the sample size we need to have roughly balanced output partitions, capped at 1M.
// Cast to double to avoid overflowing ints or longs
// 给定总的数据抽样大小，最多1M的数据量（10^6），最少20倍的RDD分区数量，也就是每个RDD分区至少抽取20条数据
val sampleSize = math.min(samplePointsPerPartitionHint.toDouble * partitions, 1e6)
// Assume the input partitions are roughly balanced and over-sample a little bit.
// 计算每个分区抽样的数据量大小，假设输入数据每个分区分布的比较均匀
// 对于超大数据集（分区数量超过5万的）乘以3会让数据稍微增大一点，对于分区数低于5万的数据集，每个分区抽取数据量为60条也不算多
val sampleSizePerPartition = math.ceil(3.0 * sampleSize / rdd.partitions.length).toInt
// 从RDD中抽取数据，返回值：（总RDD数据量，Array[分区id, 当前分区的数据量, 当前分区抽取的数据]）
val (numItems, sketched) = RangePartitioner.sketch(rdd.map(_._1), sampleSizePerPartition)
if (numItems == 0L) {
// 如果总的数据量为0（RDD为空），那么直接返回一个空的数组
Array.empty
} else {
// If a partition contains much more than the average number of items, we re-sample from it
// to ensure that enough items are collected from that partition.
// 计算总样本数量和总记录数的占比，占比最大为1.0
val fraction = math.min(sampleSize / math.max(numItems, 1L), 1.0)
// 保存样本数据的集合buffer
val candidates = ArrayBuffer.empty[(K, Float)]
// 保存数据分布不均衡的分区id（数据量超过fraction比率的分区）
val imbalancedPartitions = mutable.Set.empty[Int]
// 计算抽取出来的样本数据
sketched.foreach { case (idx, n, sample) =>
if (fraction * n > sampleSizePerPartition) {
// 如果fraction乘以当前分区中的数据量大于之前计算的每个分区的抽样数据大小，那么表示当前分区抽取的数据太少了，该分区数据分布不均衡，需要重新抽取
imbalancedPartitions += idx
} else {
// 当前分区不属于数据分布不均衡的分区，计算占比权重，并添加到candidates集合中
// The weight is 1 over the sampling probability.
val weight = (n.toDouble / sample.length).toFloat
for (key <- sample) {
candidates += ((key, weight))
}
}
}
// 对数据分布不均衡的RDD分区，重新进行数据抽样
if (imbalancedPartitions.nonEmpty) {
// Re-sample imbalanced partitions with the desired sampling probability.
// 获取数据分布不均衡的RDD分区，并构成RDD
val imbalanced = new PartitionPruningRDD(rdd.map(_._1), imbalancedPartitions.contains)
// 随机种子
val seed = byteswap32(-rdd.id - 1)
// 利用RDD的sample抽样函数API进行数据抽样
val reSampled = imbalanced.sample(withReplacement = false, fraction, seed).collect()
val weight = (1.0 / fraction).toFloat
candidates ++= reSampled.map(x => (x, weight))
}
// 将最终的抽样数据计算出rangeBounds
RangePartitioner.determineBounds(candidates, math.min(partitions, candidates.size))
}
}
}
// 下一个RDD的分区数量是rangeBounds数组中元素数量+1个
def numPartitions: Int = rangeBounds.length + 1
// 二分查找器，内部使用Java中的Arrays提供的二分查找方法
private var binarySearch: ((Array[K], K) => Int) = CollectionsUtils.makeBinarySearch[K]
// 根据RDD的key值返回对应的分区id，从0开始
def getPartition(key: Any): Int = {
// 强制转换key类型为RDD中原本的数据类型
val k = key.asInstanceOf[K]
var partition = 0
if (rangeBounds.length <= 128) {
// If we have less than 128 partitions naive search
// 如果分区数据小于等于128个，那么直接本地循环寻找当前k所属的分区下标
while (partition < rangeBounds.length && ordering.gt(k, rangeBounds(partition))) {
partition += 1
}
} else {
// Determine which binary search method to use only once.
// 如果分区数量大于128个，那么使用二分查找方法寻找对应k所属的下标
// 但是如果k在rangeBounds中没有出现，实质上返回的是一个负数（范围）或者是一个超过rangeBounds大小的数（最后一个分区，比所有的数据都大）
partition = binarySearch(rangeBounds, k)
// binarySearch either returns the match location or -[insertion point]-1
if (partition < 0) {
partition = -partition-1
}
if (partition > rangeBounds.length) {
partition = rangeBounds.length
}
}
// 根据数据排序是升序还是降序进行数据的排列，默认为升序
if (ascending) {
partition
} else {
rangeBounds.length - partition
}
}
override def equals(other: Any): Boolean = other match {
case r: RangePartitioner[_, _] =>
r.rangeBounds.sameElements(rangeBounds) && r.ascending == ascending
case _ =>
false
}
override def hashCode(): Int = {
val prime = 31
var result = 1
var i = 0
while (i < rangeBounds.length) {
result = prime * result + rangeBounds(i).hashCode
i += 1
}
result = prime * result + ascending.hashCode
result
}
@throws(classOf[IOException])
private def writeObject(out: ObjectOutputStream): Unit = Utils.tryOrIOException {
val sfactory = SparkEnv.get.serializer
sfactory match {
case js: JavaSerializer => out.defaultWriteObject()
case _ =>
out.writeBoolean(ascending)
out.writeObject(ordering)
out.writeObject(binarySearch)
val ser = sfactory.newInstance()
Utils.serializeViaNestedStream(out, ser) { stream =>
stream.writeObject(scala.reflect.classTag[Array[K]])
stream.writeObject(rangeBounds)
}
}
}
@throws(classOf[IOException])
private def readObject(in: ObjectInputStream): Unit = Utils.tryOrIOException {
val sfactory = SparkEnv.get.serializer
sfactory match {
case js: JavaSerializer => in.defaultReadObject()
case _ =>
ascending = in.readBoolean()
ordering = in.readObject().asInstanceOf[Ordering[K]]
binarySearch = in.readObject().asInstanceOf[(Array[K], K) => Int]
val ser = sfactory.newInstance()
Utils.deserializeViaNestedStream(in, ser) { ds =>
implicit val classTag = ds.readObject[ClassTag[Array[K]]]()
rangeBounds = ds.readObject[Array[K]]()
}
}
}
}

复制代码

/**
* Determines the bounds for range partitioning from candidates with weights indicating how many
* items each represents. Usually this is 1 over the probability used to sample this candidate.
*
* @param candidates unordered candidates with weights
* @param partitions number of partitions
* @return selected bounds
*/
def determineBounds[K : Ordering : ClassTag](
candidates: ArrayBuffer[(K, Float)],
partitions: Int): Array[K] = {
val ordering = implicitly[Ordering[K]]
// 按照数据进行排序，默认升序排序
val ordered = candidates.sortBy(_._1)
// 获取总的样本数据大小
val numCandidates = ordered.size
// 计算总的权重大小
val sumWeights = ordered.map(_._2.toDouble).sum
// 计算步长
val step = sumWeights / partitions
var cumWeight = 0.0
var target = step
val bounds = ArrayBuffer.empty[K]
var i = 0
var j = 0
var previousBound = Option.empty[K]
while ((i < numCandidates) && (j < partitions - 1)) {
// 获取排序后的第i个数据及权重
val (key, weight) = ordered(i)
// 累计权重
cumWeight += weight
if (cumWeight >= target) {
// Skip duplicate values.
// 权重已经达到一个步长的范围，计算出一个分区id的值
if (previousBound.isEmpty || ordering.gt(key, previousBound.get)) {// 上一个边界值为空，或者当前边界值key数据大于上一个边界的值，那么当前key有效，进行计算
// 添加当前key到边界集合中
bounds += key
// 累计target步长界限
target += step
// 分区数量加1
j += 1
// 上一个边界的值重置为当前边界的值
previousBound = Some(key)
}
}
i += 1
}
// 返回结果
bounds.toArray
}

复制代码

// CustomPartitioner
import org.apache.spark.Partitioner
/**
* @param numPartition 分区数量
*/
class CustomPartitioner(numPartition: Int) extends Partitioner{
// 返回分区的总数
override def numPartitions: Int = numPartition
// 根据传入的 key 返回分区的索引
override def getPartition(key: Any): Int = {
key.toString.toInt % numPartition
}
}
// CustomPartitionerDemo
import com.work.util.SparkUtil
import org.apache.spark.SparkContext
import org.apache.spark.rdd.RDD
object CustomPartitionerDemo {
def main(args: Array[String]): Unit = {
val sc: SparkContext = SparkUtil.getSparkContext()
println("=================== 原始数据 =====================")
// zipWithIndex 该函数将 RDD 中的元素和这个元素在 RDD 中的 ID（索引号）组合成键值对
val data: RDD[(Int, Long)] = sc.parallelize(0 to 10, 1).zipWithIndex()
println(data.collect().toBuffer)
println("=================== 分区和数据组合成 Map =====================")
val func: (Int, Iterator[(Int, Long)]) => Iterator[String] = (index: Int, iter: Iterator[(Int, Long)]) => {
iter.map(x => "[partID:" + index + ", value:" + x + "]")
}
val array: Array[String] = data.mapPartitionsWithIndex(func).collect()
for (i <- array) {
println(i)
}
println("=================== 自定义5个分区和数据组合成 Map =====================")
val rdd1: RDD[(Int, Long)] = data.partitionBy(new CustomPartitioner(5))
val array1: Array[String] = rdd1.mapPartitionsWithIndex(func).collect()
for (i <- array1) {
println(i)
}
}
}

复制代码