Spark --RDD算子

1. RDD算子分类

1.1 Transformation

Transformation(转换):根据数据集创建一个新的 数据集，计算后返回一个新的RDD。例如，一个RDD进行map操作后，生成了新的RDD。
RDD中的所有转换都是延迟加载的，也就是说，他们并不会直接计算结果。相反的，他们只是记住这些应用到基础数据集（例如一个文件）上转换动作。只有当发生一个要求返回结果给Driver的动作或者将结果写入到外部存储中，这写转换才会真正的运行，这种设计让Spark更加有效率的运行。

1.2 Action

Action(动作)：对RDD结果计算返回一个数值value给驱动程序，或者把结果存储到外部存储系统中
Action算子返回结果或保存结果，如count，collect，save等，Action操作是返回结果或将结果写入存储的操作，Action是Spark应用程序真正执行的触发动作

2. Transformation算子示例

2.1 map(func)

说明:返回一个新的RDD，该RDD由每一个输入元素经过func函数转换后组成

var source  = sc.parallelize(1 to 10).map(_ * 2)

2.2 mapPartitions(func)

l类似于map，但独立地在RDD的每一个分片上运行，因此在类型为T的RDD上运行时，func的函数类型必须是Iterator[T] => Iterator[U]。假设有N个元素，有M个分区，那么map的函数的将被调用N次,而mapPartitions被调用M次,一个函数一次处理所有分区

scala> val rdd = sc.parallelize(List(("kpop","female"),("zorro","male"),("mobin","male"),("lucy","female")))
rdd: org.apache.spark.rdd.RDD[(String, String)] = ParallelCollectionRDD[16] at parallelize at <console>:24

scala> :paste
// Entering paste mode (ctrl-D to finish)
def partitionsFun(iter : Iterator[(String,String)]) : Iterator[String] = {
  var woman = List[String]()
  while (iter.hasNext){
    val next = iter.next()
    next match {
       case (_,"female") => woman = next._1 :: woman
       case _ =>
    }
  }
  woman.iterator
}
// Exiting paste mode, now interpreting.

partitionsFun: (iter: Iterator[(String, String)])Iterator[String]

scala> val result = rdd.mapPartitions(partitionsFun)
result: org.apache.spark.rdd.RDD[String] = MapPartitionsRDD[17] at mapPartitions at <console>:28

scala> result.collect()
res13: Array[String] = Array(kpop, lucy)

2.3 glom

将每一个分区形成一个数组，形成新的RDD类型时RDD[Array[T]]

scala> val rdd = sc.parallelize(1 to 16,4)
rdd: org.apache.spark.rdd.RDD[Int] = ParallelCollectionRDD[65] at parallelize at <console>:24

scala> rdd.glom().collect()
res25: Array[Array[Int]] = Array(Array(1, 2, 3, 4), Array(5, 6, 7, 8), Array(9, 10, 11, 12), Array(13, 14, 15, 16))

2.4 flatMap(func)

类似于map，但是每一个输入元素可以被映射为0或多个输出元素（所以func应该返回一个序列，而不是单一元素）

scala> val data = sc.parallelize(1 to 5)
data: org.apache.spark.rdd.RDD[Int] = ParallelCollectionRDD[8] at parallelize at <console>:24

scala> data.collect()
res3: Array[Int] = Array(1, 2, 3, 4, 5)

scala> val flagmap = data.flatMap(1 to _)
flagmap: org.apache.spark.rdd.RDD[Int] = MapPartitionsRDD[9] at flatMap at <console>:25

scala> flagmap.collect()
res4: Array[Int] = Array(1, 1, 2, 1, 2, 3, 1, 2, 3, 4, 1, 2, 3, 4, 5)

2.5 filter(func)

返回一个新的RDD，该RDD由经过func函数计算后返回值为true的输入元素组成

scala> var data= sc.parallelize(Array("xiaoming","xiaojiang","xiaohe","dazhi"))
sourceFilter: org.apache.spark.rdd.RDD[String] = ParallelCollectionRDD[10] at parallelize at <console>:24

scala> val filter = sourceFilter.filter(_.contains("xiao"))
filter: org.apache.spark.rdd.RDD[String] = MapPartitionsRDD[11] at filter at <console>:26

scala> data.collect()
res9: Array[String] = Array(xiaoming, xiaojiang, xiaohe, dazhi)

scala> filter.collect()
res10: Array[String] = Array(xiaoming, xiaojiang, xiaohe)

2.6 mapPartitionsWithIndex(func)

类似于mapPartitions，但func带有一个整数参数表示分片的索引值，因此在类型为T的RDD上运行时，func的函数类型必须是(Int, Interator[T]) => Iterator[U]

scala> val data = sc.parallelize(List(("kpop","female"),("zorro","male"),("mobin","male"),("lucy","female")))
data: org.apache.spark.rdd.RDD[(String, String)] = ParallelCollectionRDD[12] at parallelize at <console>:24

scala> :paste
// Entering paste mode (ctrl-D to finish)

def partitionsFun(index : Int, iter : Iterator[(String,String)]) : Iterator[String] = {
  var woman = List[String]()
  while (iter.hasNext){
    val next = iter.next()
    next match {
       case (_,"female") => woman = "["+index+"]"+next._1 :: woman
       case _ =>
    }
  }
  woman.iterator
}

// Exiting paste mode, now interpreting.

partitionsFun: (index: Int, iter: Iterator[(String, String)])Iterator[String]

scala> val result = data.mapPartitionsWithIndex(partitionsFun)
result: org.apache.spark.rdd.RDD[String] = MapPartitionsRDD[13] at mapPartitionsWithIndex at <console>:27

scala> result.collect()
res7: Array[String] = Array([0]lucy, [0]kpop)

2.7 sample(withReplacement, fraction, seed)

以指定的随机种子随机抽样出数量为fraction的数据，withReplacement表示是抽出的数据是否放回，true为有放回的抽样，false为无放回的抽样，seed用于指定随机数生成器种子。例子从RDD中随机且有放回的抽出50%的数据，随机种子值为3（即可能以1 2 3的其中一个起始值）

scala> val data = sc.parallelize(1 to 10)
data: org.apache.spark.rdd.RDD[Int] = ParallelCollectionRDD[4] at parallelize at <console>:24

scala> data.collect()
res3: Array[Int] = Array(1, 2, 3, 4, 5, 6, 7, 8, 9, 10)

scala> val sample = data.sample(true, 0.4, 2)
sample: org.apache.spark.rdd.RDD[Int] = PartitionwiseSampledRDD[5] at sample at <console>:25

scala> sample.collect()
res4: Array[Int] = Array(1, 2, 2, 6, 6, 10)

scala> val sample2 = data.sample(false, 0.2, 3)
sample2: org.apache.spark.rdd.RDD[Int] = PartitionwiseSampledRDD[6] at sample at <console>:25

scala> sample2.collect()
res5: Array[Int] = Array(1)

2.8 distinct([numTasks]))

对源RDD进行去重后返回一个新的RDD. 默认情况下，只有8个并行任务来操作，但是可以传入一个可选的numTasks参数改变它。

scala> val data = sc.parallelize(List(1,2,1,5,2,9,6,1))
data: org.apache.spark.rdd.RDD[Int] = ParallelCollectionRDD[0] at parallelize at <console>:24

scala> val distinct = data.distinct()
distinct: org.apache.spark.rdd.RDD[Int] = MapPartitionsRDD[3] at distinct at <console>:25

scala> distinct.collect()
res0: Array[Int] = Array(1, 6, 9, 5, 2)

scala> val distinct = data.distinct(2)
distinct: org.apache.spark.rdd.RDD[Int] = MapPartitionsRDD[6] at distinct at <console>:25

scala> distinct.collect()
res1: Array[Int] = Array(6, 2, 1, 9, 5)

2.9 partitionBy

对RDD进行分区操作，如果原有的partionRDD和现有的partionRDD是一致的话就不进行分区， 否则会生成ShuffleRDD

scala> val data = sc.parallelize(Array((1,"aaa"),(2,"bbb"),(3,"ccc"),(4,"ddd")),4)
data: org.apache.spark.rdd.RDD[(Int, String)] = ParallelCollectionRDD[7] at parallelize at <console>:24

scala> data.partitions.size
res2: Int = 4

scala> var data2 = data.partitionBy(new org.apache.spark.HashPartitioner(2))
data2: org.apache.spark.rdd.RDD[(Int, String)] = ShuffledRDD[8] at partitionBy at <console>:25

scala> data2.partitions.size
res3: Int = 2

2.10 coalesce(numPartitions)

与repartition的区别: repartition(numPartitions:Int):RDD[T]和coalesce(numPartitions:Int，shuffle:Boolean=false):RDD[T] repartition只是coalesce接口中shuffle为true的实现.
缩减分区数，用于大数据集过滤后，提高小数据集的执行效率。

scala> val rdd = sc.parallelize(1 to 16,4)
rdd: org.apache.spark.rdd.RDD[Int] = ParallelCollectionRDD[9] at parallelize at <console>:24

scala> rdd.partitions.size
res4: Int = 4

scala> val coalesceRDD = rdd.coalesce(3)
coalesceRDD: org.apache.spark.rdd.RDD[Int] = CoalescedRDD[10] at coalesce at <console>:25

scala> coalesceRDD.partitions.size
res5: Int = 3

2.11 repartition(numPartitions)

根据分区数，从新通过网络随机洗牌所有数据

scala> val data = sc.parallelize(1 to 16,4)
data: org.apache.spark.rdd.RDD[Int] = ParallelCollectionRDD[11] at parallelize at <console>:24

scala> data.partitions.size
res6: Int = 4

scala> val re = data.repartition(2)
re: org.apache.spark.rdd.RDD[Int] = MapPartitionsRDD[15] at repartition at <console>:25

scala> re.partitions.size
res7: Int = 2

scala> val re = data.repartition(4)
re: org.apache.spark.rdd.RDD[Int] = MapPartitionsRDD[19] at repartition at <console>:25

scala> re.partitions.size
res8: Int = 4

2.12 repartitionAndSortWithinPartitions(partitioner)

repartitionAndSortWithinPartitions函数是repartition函数的变种，与repartition函数不同的是，repartitionAndSortWithinPartitions在给定的partitioner内部进行排序，性能比repartition要高

2.13sortBy(func,[ascending], [numTasks])

用func先对数据进行处理，按照处理后的数据比较结果排序!

scala> val rdd = sc.parallelize(List(1,2,3,4))
rdd: org.apache.spark.rdd.RDD[Int] = ParallelCollectionRDD[21] at parallelize at <console>:24

scala> rdd.sortBy(x => x).collect()
res11: Array[Int] = Array(1, 2, 3, 4)

scala> rdd.sortBy(x => x%3).collect()
res12: Array[Int] = Array(3, 4, 1, 2)

2.14 union(otherDataset)

对源RDD和参数RDD求并集后返回一个新的RDD 不去重

scala> val rdd1 = sc.parallelize(1 to 5)
rdd1: org.apache.spark.rdd.RDD[Int] = ParallelCollectionRDD[23] at parallelize at <console>:24

scala> val rdd2 = sc.parallelize(5 to 10)
rdd2: org.apache.spark.rdd.RDD[Int] = ParallelCollectionRDD[24] at parallelize at <console>:24

scala> val rdd3 = rdd1.union(rdd2)
rdd3: org.apache.spark.rdd.RDD[Int] = UnionRDD[25] at union at <console>:28

scala> rdd3.collect()
res18: Array[Int] = Array(1, 2, 3, 4, 5, 5, 6, 7, 8, 9, 10)

2.15 subtract (otherDataset)

计算差的一种函数，去除两个RDD中相同的元素，不同的RDD将保留下来

scala> val rdd = sc.parallelize(3 to 8)
rdd: org.apache.spark.rdd.RDD[Int] = ParallelCollectionRDD[70] at parallelize at <console>:24

scala> val rdd1 = sc.parallelize(1 to 5)
rdd1: org.apache.spark.rdd.RDD[Int] = ParallelCollectionRDD[71] at parallelize at <console>:24

scala> rdd.subtract(rdd1).collect()
res27: Array[Int] = Array(8, 6, 7)

2.16 intersection(otherDataset)

对源RDD和参数RDD求交集后返回一个新的RDD

scala> val rdd1 = sc.parallelize(1 to 7)
rdd1: org.apache.spark.rdd.RDD[Int] = ParallelCollectionRDD[26] at parallelize at <console>:24

scala> val rdd2 = sc.parallelize(5 to 10)
rdd2: org.apache.spark.rdd.RDD[Int] = ParallelCollectionRDD[27] at parallelize at <console>:24

scala> val rdd3 = rdd1.intersection(rdd2)
rdd3: org.apache.spark.rdd.RDD[Int] = MapPartitionsRDD[33] at intersection at <console>:28

scala> rdd3.collect()
res19: Array[Int] = Array(6, 7, 5)

2.17 cartesian(otherDataset)

笛卡尔积

scala> val rdd1 = sc.parallelize(1 to 3)
rdd1: org.apache.spark.rdd.RDD[Int] = ParallelCollectionRDD[44] at parallelize at <console>:24

scala> val rdd2 = sc.parallelize(2 to 5)
rdd2: org.apache.spark.rdd.RDD[Int] = ParallelCollectionRDD[45] at parallelize at <console>:24

scala> rdd1.cartesian(rdd2).collect()
res14: Array[(Int, Int)] = Array((1,2), (1,3), (1,4), (1,5), (2,2), (2,3), (2,4), (2,5), (3,2), (3,3), (3,4), (3,5))

2.18 pipe(command, [envVars])

管道，对于每个分区，都执行一个perl或者shell脚本，返回输出的RDD

Shell脚本
#!/bin/sh
echo "AA"
while read LINE; do
   echo ">>>"${LINE}
done
scala> val rdd = sc.parallelize(List("hi","Hello","how","are","you"),1)
rdd: org.apache.spark.rdd.RDD[String] = ParallelCollectionRDD[50] at parallelize at <console>:24

scala> rdd.pipe("/home/bigdata/pipe.sh").collect()
res18: Array[String] = Array(AA, >>>hi, >>>Hello, >>>how, >>>are, >>>you)

scala> val rdd = sc.parallelize(List("hi","Hello","how","are","you"),2)
rdd: org.apache.spark.rdd.RDD[String] = ParallelCollectionRDD[52] at parallelize at <console>:24

scala> rdd.pipe("/home/bigdata/pipe.sh").collect()
res19: Array[String] = Array(AA, >>>hi, >>>Hello, AA, >>>how, >>>are, >>>you)

pipe.sh:
#!/bin/sh
echo "AA"
while read LINE; do
   echo ">>>"${LINE}
done

2.19 join(otherDataset, [numTasks])

在类型为(K,V)和(K,W)的RDD上调用，返回一个相同key对应的所有元素对在一起的(K,(V,W))的RDD

scala> val rdd = sc.parallelize(Array((1,"a"),(2,"b"),(3,"c")))
rdd: org.apache.spark.rdd.RDD[(Int, String)] = ParallelCollectionRDD[32] at parallelize at <console>:24

scala> val rdd1 = sc.parallelize(Array((1,4),(2,5),(3,6)))
rdd1: org.apache.spark.rdd.RDD[(Int, Int)] = ParallelCollectionRDD[33] at parallelize at <console>:24

scala> rdd.join(rdd1).collect()
res13: Array[(Int, (String, Int))] = Array((1,(a,4)), (2,(b,5)), (3,(c,6)))

2.20 cogroup(otherDataset, [numTasks])

在类型为(K,V)和(K,W)的RDD上调用，返回一个(K,(Iterable,Iterable))类型的RDD

scala> val rdd = sc.parallelize(Array((1,"a"),(2,"b"),(3,"c")))
rdd: org.apache.spark.rdd.RDD[(Int, String)] = ParallelCollectionRDD[52] at parallelize at <console>:24

scala> val rdd1 = sc.parallelize(Array((1,4),(2,5),(3,6)))
rdd1: org.apache.spark.rdd.RDD[(Int, Int)] = ParallelCollectionRDD[53] at parallelize at <console>:24

scala> rdd.cogroup(rdd1).collect()
res16: Array[(Int, (Iterable[String], Iterable[Int]))] = Array((1,(CompactBuffer(a),CompactBuffer(4))), (3,(CompactBuffer(c),CompactBuffer(6))), (2,(CompactBuffer(b),CompactBuffer(5))))

scala> val rdd2 = sc.parallelize(Array((4,4),(2,5),(3,6)))
rdd2: org.apache.spark.rdd.RDD[(Int, Int)] = ParallelCollectionRDD[56] at parallelize at <console>:24

scala> rdd.cogroup(rdd2).collect()
res17: Array[(Int, (Iterable[String], Iterable[Int]))] = Array((4,(CompactBuffer(),CompactBuffer(4))), (1,(CompactBuffer(a),CompactBuffer())), (3,(CompactBuffer(c),CompactBuffer(6))), (2,(CompactBuffer(b),CompactBuffer(5))))

scala> val rdd3 = sc.parallelize(Array((1,"a"),(1,"d"),(2,"b"),(3,"c")))
rdd3: org.apache.spark.rdd.RDD[(Int, String)] = ParallelCollectionRDD[59] at parallelize at <console>:24

scala> rdd3.cogroup(rdd2).collect()
res18: Array[(Int, (Iterable[String], Iterable[Int]))] = Array((4,(CompactBuffer(),CompactBuffer(4))), (1,(CompactBuffer(a, d),CompactBuffer())), (3,(CompactBuffer(c),CompactBuffer(6))), (2,(CompactBuffer(b),CompactBuffer(5))))

2.21 reduceByKey(func, [numTasks])

在一个(K,V)的RDD上调用，返回一个(K,V)的RDD，使用指定的reduce函数，将相同key的值聚合到一起，reduce任务的个数可以通过第二个可选的参数来设置

scala> val rdd = sc.parallelize(List(("female",1),("male",5),("female",5),("male",2)))
rdd: org.apache.spark.rdd.RDD[(String, Int)] = ParallelCollectionRDD[62] at parallelize at <console>:24

scala> val reduce = rdd.reduceByKey((x,y) => x+y)
reduce: org.apache.spark.rdd.RDD[(String, Int)] = ShuffledRDD[63] at reduceByKey at <console>:25

scala> reduce.collect()
res19: Array[(String, Int)] = Array((male,7), (female,6))

2.22 groupByKey

groupByKey也是对每个key进行操作，但只生成一个sequence

scala> val words = Array("one", "two", "two", "three", "three", "three")
words: Array[String] = Array(one, two, two, three, three, three)

scala> val wordPairsRDD = sc.parallelize(words).map(word => (word, 1))
wordPairsRDD: org.apache.spark.rdd.RDD[(String, Int)] = MapPartitionsRDD[1] at map at <console>:26

scala> wordPairsRDD.collect()
res0: Array[(String, Int)] = Array((one,1), (two,1), (two,1), (three,1), (three,1), (three,1))

scala> val group = wordPairsRDD.groupByKey()
group: org.apache.spark.rdd.RDD[(String, Iterable[Int])] = ShuffledRDD[2] at groupByKey at <console>:25

scala> group.collect()
res1: Array[(String, Iterable[Int])] = Array((two,CompactBuffer(1, 1)), (one,CompactBuffer(1)), (three,CompactBuffer(1, 1, 1)))

scala> group.map(t => (t._1, t._2.sum))
res2: org.apache.spark.rdd.RDD[(String, Int)] = MapPartitionsRDD[3] at map at <console>:26

scala> res2.collect()
res3: Array[(String, Int)] = Array((two,2), (one,1), (three,3))

scala> val map = group.map(t => (t._1, t._2.sum))
map: org.apache.spark.rdd.RDD[(String, Int)] = MapPartitionsRDD[4] at map at <console>:25

scala> map.collect()
res4: Array[(String, Int)] = Array((two,2), (one,1), (three,3))

2.23 combineByKey[C]

createCombiner: V => C,  mergeValue: (C, V) => C,  mergeCombiners: (C, C) => C)
对相同K，把V合并成一个集合。
createCombiner: combineByKey() 会遍历分区中的所有元素，因此每个元素的键要么还没有遇到过，要么就 和之前的某个元素的键相同。如果这是一个新的元素,combineByKey() 会使用一个叫作 createCombiner() 的函数来创建 
那个键对应的累加器的初始值
mergeValue: 如果这是一个在处理当前分区之前已经遇到的键， 它会使用 mergeValue() 方法将该键的累加器对应的当前值与这个新的值进行合并
mergeCombiners: 由于每个分区都是独立处理的， 因此对于同一个键可以有多个累加器。如果有两个或者更多的分区都有对应同一个键的累加器， 就需要使用用户提供的 mergeCombiners() 方法将各个分区的结果进行合并

scala> val scores = Array(("Fred", 88), ("Fred", 95), ("Fred", 91), ("Wilma", 93), ("Wilma", 95), ("Wilma", 98))
scores: Array[(String, Int)] = Array((Fred,88), (Fred,95), (Fred,91), (Wilma,93), (Wilma,95), (Wilma,98))

scala> val input = sc.parallelize(scores)
input: org.apache.spark.rdd.RDD[(String, Int)] = ParallelCollectionRDD[52] at parallelize at <console>:26

scala> val combine = input.combineByKey(
     |     {(v)=>(v,1)},
     |     {(acc:(Int,Int),v)=>(acc._1+v,acc._2+1)},
     |     {(acc1:(Int,Int),acc2:(Int,Int))=>(acc1._1+acc2._1,acc1._2+acc2._2)})
combine: org.apache.spark.rdd.RDD[(String, (Int, Int))] = ShuffledRDD[53] at combineByKey at <console>:28

scala> val result = combine.map{
     |     {case (key,value) => (key,value._1/value._2.toDouble)}}
result: org.apache.spark.rdd.RDD[(String, Double)] = MapPartitionsRDD[54] at map at <console>:30

scala> result.collect()
res33: Array[(String, Double)] = Array((Wilma,95.33333333333333), (Fred,91.33333333333333))

2.24 aggregateByKey

(zeroValue:U,[partitioner: Partitioner]) (seqOp: (U, V) => U,combOp: (U, U) => U)
在kv对的RDD中，，按key将value进行分组合并，合并时，将每个value和初始值作为seq函数的参数，进行计算，返回的结果作为一个新的kv对，然后再将结果按照key进行合并，最后将每个分组的value传递给combine函数进行计算（先将前两个value进行计算，将返回结果和下一个value传给combine函数，以此类推），将key与计算结果作为一个新的kv对输出。
seqOp函数用于在每一个分区中用初始值逐步迭代value，combOp函数用于合并每个分区中的结果

scala> val rdd = sc.parallelize(List((1,3),(1,2),(1,4),(2,3),(3,6),(3,8)),3)
rdd: org.apache.spark.rdd.RDD[(Int, Int)] = ParallelCollectionRDD[7] at parallelize at <console>:24


scala> val agg = rdd.aggregateByKey(0)(math.max(_,_),_+_)
agg: org.apache.spark.rdd.RDD[(Int, Int)] = ShuffledRDD[8] at aggregateByKey at <console>:25

scala> agg.collect()
res5: Array[(Int, Int)] = Array((3,8), (1,7), (2,3))

scala> agg.partitions.size
res6: Int = 3

scala> val rdd = sc.parallelize(List((1,3),(1,2),(1,4),(2,3),(3,6),(3,8)),1)
rdd: org.apache.spark.rdd.RDD[(Int, Int)] = ParallelCollectionRDD[9] at parallelize at <console>:24

scala> val agg = rdd.aggregateByKey(0)(math.max(_,_),_+_).collect()
agg: Array[(Int, Int)] = Array((1,4), (3,8), (2,3))

2.25 foldByKey

(zeroValue: V)(func: (V, V) => V): RDD[(K, V)]
aggregateByKey的简化操作，seqop和combop相同

scala> val rdd = sc.parallelize(List((1,3),(1,2),(1,4),(2,3),(3,6),(3,8)),3)
rdd: org.apache.spark.rdd.RDD[(Int, Int)] = ParallelCollectionRDD[11] at parallelize at <console>:24

scala> val agg = rdd.foldByKey(0)(_+_)
agg: org.apache.spark.rdd.RDD[(Int, Int)] = ShuffledRDD[12] at foldByKey at <console>:25

scala> agg.collect()
res7: Array[(Int, Int)] = Array((3,14), (1,9), (2,3))

2.26 sortByKey([ascending], [numTasks])

在一个(K,V)的RDD上调用，K必须实现Ordered接口，返回一个按照key进行排序的(K,V)的RDD

scala> val rdd = sc.parallelize(Array((3,"aa"),(6,"cc"),(2,"bb"),(1,"dd")))
rdd: org.apache.spark.rdd.RDD[(Int, String)] = ParallelCollectionRDD[13] at parallelize at <console>:24

scala> rdd.sortByKey(true).collect()
res8: Array[(Int, String)] = Array((1,dd), (2,bb), (3,aa), (6,cc))

scala> rdd.sortByKey(false).collect()
res9: Array[(Int, String)] = Array((6,cc), (3,aa), (2,bb), (1,dd))

2.27 mapValues

针对于(K,V)形式的类型只对V进行操作

scala> val rdd3 = sc.parallelize(Array((1,"a"),(1,"d"),(2,"b"),(3,"c")))
rdd3: org.apache.spark.rdd.RDD[(Int, String)] = ParallelCollectionRDD[16] at parallelize at <console>:24

scala> rdd3.mapValues(_+"|||").collect()
res10: Array[(Int, String)] = Array((1,a|||), (1,d|||), (2,b|||), (3,c|||))

3. Action算子示例

3.1 reduce(func)

通过func函数聚集RDD中的所有元素，这个功能必须是可交换且可并联的

scala> val rdd1 = sc.makeRDD(1 to 10,2)
rdd1: org.apache.spark.rdd.RDD[Int] = ParallelCollectionRDD[0] at makeRDD at <console>:24

scala> rdd1.reduce(_+_)
res0: Int = 55

scala> val rdd2 = sc.makeRDD(Array(("a",1),("a",3),("c",3),("d",5)))
rdd2: org.apache.spark.rdd.RDD[(String, Int)] = ParallelCollectionRDD[1] at makeRDD at <console>:24

scala> rdd2.reduce((x,y)=>(x._1 + y._1,x._2 + y._2))
res1: (String, Int) = (aacd,12)

3.2 collect()

在驱动程序中，以数组的形式返回数据集的所有元素

3.3 count()

返回RDD的元素个数

3.4 first()

返回RDD的第一个元素（类似于take(1)）

3.5 take(n)

返回一个由数据集的前n个元素组成的数组

3.6 takeSample(withReplacement,num, [seed])

返回一个数组，该数组由从数据集中随机采样的num个元素组成，可以选择是否用随机数替换不足的部分，seed用于指定随机数生成器种子

3.7 takeOrdered(n)

返回前几个的排序

3.8 aggregate

(zeroValue: U)(seqOp: (U, T) ⇒ U, combOp: (U, U) ⇒ U)
aggregate函数将每个分区里面的元素通过seqOp和初始值进行聚合，然后用combine函数将每个分区的结果和初始值(zeroValue)进行combine操作。这个函数最终返回的类型不需要和RDD中元素类型一致

scala> var rdd1 = sc.makeRDD(1 to 10,2)
rdd1: org.apache.spark.rdd.RDD[Int] = ParallelCollectionRDD[88] at makeRDD at <console>:24

scala>:
res56: Int = 58

scala> rdd1.aggregate(1)(
     | {(x : Int,y : Int) => x * y},
     | {(a : Int,b : Int) => a + b}
     | )
res57: Int = 30361

3.9 fold(num)(func)

折叠操作，aggregate的简化操作，seqop和combop一样

scala> var rdd1 = sc.makeRDD(1 to 4,2)
rdd1: org.apache.spark.rdd.RDD[Int] = ParallelCollectionRDD[2] at makeRDD at <console>:24

scala> :paste
// Entering paste mode (ctrl-D to finish)

rdd1.aggregate(1)(
     | {(x : Int,y : Int) => x + y},
     | {(a : Int,b : Int) => a + b}
     | )

// Exiting paste mode, now interpreting.

res5: Int = 13

scala> rdd1.fold(1)(_+_)
res6: Int = 13

3.10 saveAsTextFile(path)

将数据集的元素以textfile的形式保存到HDFS文件系统或者其他支持的文件系统，对于每个元素，Spark将会调用toString方法，将它装换为文件中的文本

3.11 saveAsSequenceFile(path)

将数据集中的元素以Hadoop sequencefile的格式保存到指定的目录下，可以使HDFS或者其他Hadoop支持的文件系统

3.12 saveAsObjectFile(path)

用于将RDD中的元素序列化成对象，存储到文件中

3.13 countByKey()

针对(K,V)类型的RDD，返回一个(K,Int)的map，表示每一个key对应的元素个数

scala> val rdd = sc.parallelize(List((1,3),(1,2),(1,4),(2,3),(3,6),(3,8)),3)
rdd: org.apache.spark.rdd.RDD[(Int, Int)] = ParallelCollectionRDD[3] at parallelize at <console>:24

scala> rdd.countByKey()
res7: scala.collection.Map[Int,Long] = Map(3 -> 2, 1 -> 3, 2 -> 1)

3.14 foreach(func)

在数据集的每一个元素上，运行函数func进行更新

scala> var rdd = sc.makeRDD(1 to 10,2)
rdd: org.apache.spark.rdd.RDD[Int] = ParallelCollectionRDD[0] at makeRDD at <console>:24

scala> var sum = sc.accumulator(0)
warning: there were two deprecation warnings; re-run with -deprecation for details
sum: org.apache.spark.Accumulator[Int] = 0

scala> rdd.foreach(sum+=_)

scala> sum.value
res1: Int = 55

scala> rdd.collect().foreach(println)
1
2
3
4
5
6
7
8
9
10