spark rdd 算子

刚开始使用SPARK的同学都会因为文档说明简单无示例而导致前期开发效率较低，在网上有一位老师的博客给出了很详细的使用示例，我简单将其翻译成中文，自己顺便也熟悉一下没使用过的API。

E文版

一些注解：

• 数据分片（partitions）：执行在计算节点中的一份数据集合，包含多个数据单元

以下为翻译内容：

RDD的API示例

RDD是弹性分布式数据集的简称，RDDs在Spark系统扮演干活的角色。当我们使用的时候，可以认为RDD是一个独立的数据分片集合，这些数据可以是计算后的数据结果。但是RDD实际上远不止这些，在集群中，独立的数据分片可以分布在不同的运算节点中。RDD是访问这些数据的入口，同时也能在这些数据上做运算和数据变换。不论RDD的数据丢失了或者部分丢失了，系统都能够通过lineage信息来恢复数据。Lineage是产生当前RDD的数据变化操作序列。总的来说，Spark可以从大部分的错误中恢复。

Spark中可以用的RDDs都是间接或者直接继承至RDD类。这个类包含很多处理数据分片的方法。RDD类是抽象类，平时大家用的都是RDD的继承实现类。继承类都需要实现一些核心方法，才能被外部使用。

Spark变成一个流行的大数据处理系统的原因是它没有对RDD分片中的数据增加限制。RDD API中包含了很多非常有用的方法。但是创作者为了保证核心API适配于所有的数据类型，所以有一些方便的方法并没有加到里面。

基础的RDD API将每个数据单元稻城单独的一个输入值。但是，使用者很多时候希望输入的是一个kV对，因此Spark提供了扩展的RDD类PairRDDFunctions。现在有四个RDD API扩展类。它们是：

DoubleRDDFunctions

这个扩展类包含了很多数值的聚合方法。如果RDD的数据单元能够隐式变换成Scala的double数据类型，这些方法会非常有用

PairRDDFunctions

该扩展类中的方法输入的数据单元是一个包含两个元素的tuple结构。Spark会把其中第一个元素当成key，第二个当成value。

OrderedRDDFunctions

该扩展类的方法需要输入数据是2元tuple，并且key能够排序。

SequenceFileRDDFunctions

这个扩展类包含一些可以创建Hadoop sequence文件的方法。输入数据必须是2元tuple。但需要额外考虑tuple元素能够转换成可写类型。

下面会按照字母顺序列出RDD方法，以下简称会在相应的API后面，说明该API属于哪个扩展类。

[Double] - DoubleRDDFunctions

[Ordered] - OrderedRDDFunctions

[Pair] - PairRDDFunctions

[SeqFile] - SequenceFileRDDFunctions

aggregate

aggregate方法是一个高度定制化的RDD聚合和reduction方法，但是由于Scala和Spark在数据处理上的方式的原因，我们应该在真正使用的时候多加注意。下面是我们在使用中所观察到的要点：

• reduce和combine方法是可代替的并且相互关联的
• 如方法定义中所述，combiner的输出必须是和输入类型一致，因为Spark是链式执行
• zeroValue是U的初始值，作为 seqOp 和 combOp 执行之前的首个输入元素。你可以随着自己的需求改变它，但是为了代码执行结果的一致性，无论数据被分成多少份，每份有多大，都要保证产生同样的结果。
• 不要假设每个数据分片的执行顺序。
• 在每个数据分片上执行reduce之前，会设置一个zeroValue作为第一个输入，在执行combiner方法之前也会降zeroValue作为第一个输入。
• 为什么会有两个数据合并的输入函数？第一个函数将输入值映射到结果空间，输入值T可以和结果空间的值U不是一个类型，第二个函数将第一个函数映射后的结果再合并起来。
• 为什么有人需要输入两种数据类型的数据？此处举出一个示例，假设我们现在在考古遗址的现场，并且通过金属探测器找到重要的位置，并把位置的坐标记录下来。这样我们就可以通过聚合函数聚合这些标志出的坐标并画在地图上。我们可以设置zeroValue 为一个空的地图。GPS坐标被存储在多个数据分片中， seqOp 可以将GPS坐标转化为地图坐标，并标注在 zeroValue 表示的地图上。 combOp 将这些标注在地图上的数据以及每个数据分片所代表的地图合并成一张完整的地图。

定义

def aggregate[U: ClassTag](zeroValue: U)(seqOp: (U, T) => U, combOp: (U, U) => U): U

Examples 1

val z = sc.parallelize(List(1,2,3,4,5,6), 2)
z.aggregate(0)(math.max(_, _), _ + _)
res40: Int = 9

val z = sc.parallelize(List("a","b","c","d","e","f"),2)
z.aggregate("")(_ + _, _+_)
res115: String = abcdef

z.aggregate("x")(_ + _, _+_)
res116: String = xxdefxabc

val z = sc.parallelize(List("12","23","345","4567"),2)
z.aggregate("")((x,y) => math.max(x.length, y.length).toString, (x,y) => x + y)
res141: String = 42

z.aggregate("")((x,y) => math.min(x.length, y.length).toString, (x,y) => x + y)
res142: String = 11

val z = sc.parallelize(List("12","23","345",""),2)
z.aggregate("")((x,y) => math.min(x.length, y.length).toString, (x,y) => x + y)
res143: String = 10

Examples 2

val z = sc.parallelize(List("12","23","","345"),2)
z.aggregate("")((x,y) => math.min(x.length, y.length).toString, (x,y) => x + y)
res144: String = 11

相对于输入是数值，字符串输入会使人有一点点困惑，如果没想清楚aggregate功能的话。以示例2为例子，zeroValue为空字符串，假设第一个数据分片的输入是"12","23"，则第一次计算""与"12"的最小长度是0，第二次计算以第一次计算结果做为输入"0"和"23"则最小长度为1，则此数据分片的最终输出为"1"，同理第二个数据分片也能得到输出"1"，最后combiner合并后可得"11"。

checkpoint

checkpoint在RDD被执行的时候调用。被checkpoint的RDD被存储在一个二进制文件中，文件存储在Spark context的初始设置的“checkpoint文件夹”里面。checkpoint文件夹必须存在于所有slave机器中，或者也可以设置成HDFS的地址。

定义

def checkpoint()

Example

sc.setCheckpointDir("my_directory_name")
val a = sc.parallelize(1 to 4)
a.checkpoint
a.count
14/02/25 18:13:53 INFO SparkContext: Starting job: count at 

  
   
 :15
...
14/02/25 18:13:53 INFO MemoryStore: Block broadcast_5 stored as values to memory (estimated size 115.7 KB, free 296.3 MB)
14/02/25 18:13:53 INFO RDDCheckpointData: Done checkpointing RDD 11 to file:/home/cloudera/Documents/spark-0.9.0-incubating-bin-cdh4/bin/my_directory_name/65407913-fdc6-4ec1-82c9-48a1656b95d6/rdd-11, new parent is RDD 12
res23: Long = 4 



  

coalesce, repartition

将RDD数据重新分片，repartition是coalesce(numPartitions, shuffle = true)的缩写。

定义

def coalesce ( numPartitions : Int , shuffle : Boolean = false ): RDD [T]def repartition ( numPartitions : Int ): RDD [T]

Example

val y = sc.parallelize(1 to 10, 10)
val z = y.coalesce(2, false)
z.partitions.length
res9: Int = 2

cogroup [Pair], groupWith [Pair]

将需要cogroup的所有RDD对象放在一起，并将所有RDD中的所有tuple按key进行分组，得到的结果tuple包含当前tuple和其他tuple按可以分组后的value集合。具体参看示例。

定义

def cogroup[W](other: RDD[(K, W)]): RDD[(K, (Iterable[V], Iterable[W]))]

def cogroup[W](other: RDD[(K, W)], numPartitions: Int): RDD[(K, (Iterable[V], Iterable[W]))]

def cogroup[W](other: RDD[(K, W)], partitioner: Partitioner): RDD[(K, (Iterable[V], Iterable[W]))]

def cogroup[W1, W2](other1: RDD[(K, W1)], other2: RDD[(K, W2)]): RDD[(K, (Iterable[V], Iterable[W1], Iterable[W2]))]

def cogroup[W1, W2](other1: RDD[(K, W1)], other2: RDD[(K, W2)], numPartitions: Int): RDD[(K, (Iterable[V], Iterable[W1], Iterable[W2]))]

def cogroup[W1, W2](other1: RDD[(K, W1)], other2: RDD[(K, W2)], partitioner: Partitioner): RDD[(K, (Iterable[V], Iterable[W1], Iterable[W2]))]

def groupWith[W](other: RDD[(K, W)]): RDD[(K, (Iterable[V], Iterable[W]))]

def groupWith[W1, W2](other1: RDD[(K, W1)], other2: RDD[(K, W2)]): RDD[(K, (Iterable[V], IterableW1], Iterable[W2]))]

Examples

val a = sc.parallelize(List(1, 2, 1, 3), 1)
val b = a.map((_, "b"))
val c = a.map((_, "c"))
b.cogroup(c).collect
res7: Array[(Int, (Iterable[String], Iterable[String]))] = Array(
(2,(ArrayBuffer(b),ArrayBuffer(c))),
(3,(ArrayBuffer(b),ArrayBuffer(c))),
(1,(ArrayBuffer(b, b),ArrayBuffer(c, c)))
)

val d = a.map((_, "d"))
b.cogroup(c, d).collect
res9: Array[(Int, (Iterable[String], Iterable[String], Iterable[String]))] = Array(
(2,(ArrayBuffer(b),ArrayBuffer(c),ArrayBuffer(d))),
(3,(ArrayBuffer(b),ArrayBuffer(c),ArrayBuffer(d))),
(1,(ArrayBuffer(b, b),ArrayBuffer(c, c),ArrayBuffer(d, d)))
)

val x = sc.parallelize(List((1, "apple"), (2, "banana"), (3, "orange"), (4, "kiwi")), 2)
val y = sc.parallelize(List((5, "computer"), (1, "laptop"), (1, "desktop"), (4, "iPad")), 2)
x.cogroup(y).collect
res23: Array[(Int, (Iterable[String], Iterable[String]))] = Array(
(4,(ArrayBuffer(kiwi),ArrayBuffer(iPad))), 
(2,(ArrayBuffer(banana),ArrayBuffer())), 
(3,(ArrayBuffer(orange),ArrayBuffer())),
(1,(ArrayBuffer(apple),ArrayBuffer(laptop, desktop))),
(5,(ArrayBuffer(),ArrayBuffer(computer))))

collect

将RDD转换为Scala数组。在collect中可以添加一个过滤函数对返回的数据进行初步处理，将原始数据T转化为目标数据U。

定义

def collect(): Array[T]def collect[U: ClassTag](f: PartialFunction[T, U]): RDD[U]

Example

val c = sc.parallelize(List("Gnu", "Cat", "Rat", "Dog", "Gnu", "Rat"), 2)
c.collect
res29: Array[String] = Array(Gnu, Cat, Rat, Dog, Gnu, Rat)

collectAsMap [Pair]

类似collect方法，但是作用于key-value形态的RDDs，方法会返回Scala中map数据类型的kv结构数据。

定义

def collectAsMap(): Map[K, V]

Example

val a = sc.parallelize(List(1, 2, 1, 3), 1)
val b = a.zip(a)
b.collectAsMap
res1: scala.collection.Map[Int,Int] = Map(2 -> 2, 1 -> 1, 3 -> 3)

combineByKey[Pair]

在每个partition中先创建初始combiner（createCombiner），然后将当前RDD的数据单元逐个输入combiner进行处理（如拼装），每个partition处理后，再在mergeCombiner中将各个partition的结果进行综合处理。

定义

def combineByKey[C](createCombiner: V => C, mergeValue: (C, V) => C, mergeCombiners: (C, C) => C): RDD[(K, C)]

def combineByKey[C](createCombiner: V => C, mergeValue: (C, V) => C, mergeCombiners: (C, C) => C, numPartitions: Int): RDD[(K, C)]

def combineByKey[C](createCombiner: V => C, mergeValue: (C, V) => C, mergeCombiners: (C, C) => C, partitioner: Partitioner, mapSideCombine: Boolean = true, serializerClass: String = null): RDD[(K, C)]

Example

val a = sc.parallelize(List("dog","cat","gnu","salmon","rabbit","turkey","wolf","bear","bee"), 3)
val b = sc.parallelize(List(1,1,2,2,2,1,2,2,2), 3)
val c = b.zip(a)
val d = c.combineByKey(List(_), (x:List[String], y:String) => y :: x, (x:List[String], y:List[String]) => x ::: y)
d.collect
res16: Array[(Int, List[String])] = Array((1,List(cat, dog, turkey)), (2,List(gnu, rabbit, salmon, bee, bear, wolf)))

context, sparkContext

返回SparkContext对象用来创建RDD。

定义

def compute(split: Partition, context: TaskContext): Iterator[T]

Example

val c = sc.parallelize(List("Gnu", "Cat", "Rat", "Dog"), 2)
c.context
res8: org.apache.spark.SparkContext = org.apache.spark.SparkContext@58c1c2f1

count

返回RDD中数据单元的个数。

定义

def count(): Long

Example

val c = sc.parallelize(List("Gnu", "Cat", "Rat", "Dog"), 2)
c.count
res2: Long = 4

countByKey [Pair]

针对二元组每个unique的key进行count。

定义

def countByKey(): Map[K, Long]

Example

val c = sc.parallelize(List((3, "Gnu"), (3, "Yak"), (5, "Mouse"), (3, "Dog")), 2)
c.countByKey
res3: scala.collection.Map[Int,Long] = Map(3 -> 3, 5 -> 1)

countByValue

返回unique的value所对应的个数。相对应countByKey。

定义

def countByValue(): Map[T, Long]

Example

val b = sc.parallelize(List(1,2,3,4,5,6,7,8,2,4,2,1,1,1,1,1))
b.countByValue
res27: scala.collection.Map[Int,Long] = Map(5 -> 1, 8 -> 1, 3 -> 1, 6 -> 1, 1 -> 6, 2 -> 3, 4 -> 2, 7 -> 1)

countApproxDistinct

快速计算count，但不精确。relativeSD控制精度。

定义

def countApproxDistinct(relativeSD: Double = 0.05): Long

Example

val a = sc.parallelize(1 to 10000, 20)
val b = a++a++a++a++a
b.countApproxDistinct(0.1)
res14: Long = 8224

b.countApproxDistinct(0.05)
res15: Long = 9750

b.countApproxDistinct(0.01)
res16: Long = 9947

b.countApproxDistinct(0.001)
res0: Long = 10000

countApproxDistinctByKey [Pair]

类似countApproxDistinct，此方法计算不同key下不同value的数目。但也不精确，relativeSD控制精度。

定义

def countApproxDistinctByKey(relativeSD: Double = 0.05): RDD[(K, Long)]

def countApproxDistinctByKey(relativeSD: Double, numPartitions: Int): RDD[(K, Long)]

def countApproxDistinctByKey(relativeSD: Double, partitioner: Partitioner): RDD[(K, Long)]

Example

val a = sc.parallelize(List("Gnu", "Cat", "Rat", "Dog"), 2)
val b = sc.parallelize(a.takeSample(true, 10000, 0), 20)
val c = sc.parallelize(1 to b.count().toInt, 20)
val d = b.zip(c)
d.countApproxDistinctByKey(0.1).collect
res15: Array[(String, Long)] = Array((Rat,2567), (Cat,3357), (Dog,2414), (Gnu,2494))

d.countApproxDistinctByKey(0.01).collect
res16: Array[(String, Long)] = Array((Rat,2555), (Cat,2455), (Dog,2425), (Gnu,2513))

d.countApproxDistinctByKey(0.001).collect
res0: Array[(String, Long)] = Array((Rat,2562), (Cat,2464), (Dog,2451), (Gnu,2521))

dependencies

返回当前RDD所依赖的RDD。

定义

final def dependencies: Seq[Dependency[_]]

Example

val b = sc.parallelize(List(1,2,3,4,5,6,7,8,2,4,2,1,1,1,1,1))
b: org.apache.spark.rdd.RDD[Int] = ParallelCollectionRDD[32] at parallelize at 

  
   
 :12
b.dependencies.length
Int = 0

b.map(a => a).dependencies.length
res40: Int = 1

b.cartesian(a).dependencies.length
res41: Int = 2

b.cartesian(a).dependencies
res42: Seq[org.apache.spark.Dependency[_]] = List(org.apache.spark.rdd.CartesianRDD
 
   
 
   anon$2@6d2efbbd) 



  

distinct

针对RDD中的数据进行去重。

定义

def distinct(): RDD[T]def distinct(numPartitions: Int): RDD[T]

Example

val c = sc.parallelize(List("Gnu", "Cat", "Rat", "Dog", "Gnu", "Rat"), 2)
c.distinct.collect
res6: Array[String] = Array(Dog, Gnu, Cat, Rat)

val a = sc.parallelize(List(1,2,3,4,5,6,7,8,9,10))
a.distinct(2).partitions.length
res16: Int = 2

a.distinct(3).partitions.length
res17: Int = 3

first

返回RDD数据中第一个数据元素。

定义

def first(): T

Example

val c = sc.parallelize(List("Gnu", "Cat", "Rat", "Dog"), 2)
c.first
res1: String = Gnu

filter

对RDD中的数据单元进行过滤。如果过滤函数f返回true，则当前被验证的数据单元会被过滤。

定义

def filter(f: T => Boolean): RDD[T]

Example

val a = sc.parallelize(1 to 10, 3)
a.filter(_ % 2 == 0)
b.collect
res3: Array[Int] = Array(2, 4, 6, 8, 10)

过滤函数需要和输入的数据类型匹配，如果输入数据是复合的数据类型可以利用partial functions。以下举了两个对比例子介绍。

Examples for mixed data without partial functions

val b = sc.parallelize(1 to 8)
b.filter(_ < 4).collect
res15: Array[Int] = Array(1, 2, 3)

val a = sc.parallelize(List("cat", "horse", 4.0, 3.5, 2, "dog"))
a.filter(_ < 4).collect


  
   
 :15: error: value < is not a member of Any 


  

以上是一个类型不匹配而导致失败的例子。

以下使用collect方法解决以上多类型数据的过滤问题，collect会使用isDefinedAt方法去判断当前输入是否符合，具体参见以下示例。

Examples for mixed data with partial functions

val a = sc.parallelize(List("cat", "horse", 4.0, 3.5, 2, "dog"))
a.collect({case a: Int    => "is integer" |
           case b: String => "is string" }).collect
res17: Array[String] = Array(is string, is string, is integer, is string)

val myfunc: PartialFunction[Any, Any] = {
  case a: Int    => "is integer" |
  case b: String => "is string" }
myfunc.isDefinedAt("")
res21: Boolean = true

myfunc.isDefinedAt(1)
res22: Boolean = true

myfunc.isDefinedAt(1.5)
res23: Boolean = false

小心，上面的代码可行是因为它只是检测了数据的类型。如果你如果像下例中的情况，你需要将Any类型转换为你想要的类型。否则编译器不知道应该生成什么字节码：

val myfunc2: PartialFunction[Any, Any] = {case x if (x < 4) => "x"}


  
   
 :10: error: value < is not a member of Any

val myfunc2: PartialFunction[Int, Any] = {case x if (x < 4) => "x"}
myfunc2: PartialFunction[Int,Any] = 
 
   

 
   

  

filterWith

这个方法是filter的扩展方法。它接收两组函数。第一组执行转换Int->T，这组在每个分片(partitions)上只执行一次。它会将分片中数据的index转化成T。第二组是(U,T)->Boolean。T是上一步index变换后的结果。U是RDD中的数据单元。最后这个函数返回ture或者false。

定义

def filterWith[A: ClassTag](constructA: Int => A)(p: (T, A) => Boolean): RDD[T]

Example

val a = sc.parallelize(1 to 9, 3)
val b = a.filterWith(i => i)((x,i) => x % 2 == 0 || i % 2 == 0)
b.collect
res37: Array[Int] = Array(1, 2, 3, 4, 6, 7, 8, 9)

val a = sc.parallelize(List(1,2,3,4,5,6,7,8,9,10), 5)
a.filterWith(x=> x)((a, b) =>  b == 0).collect
res30: Array[Int] = Array(1, 2)

a.filterWith(x=> x)((a, b) =>  a % (b+1) == 0).collect
res33: Array[Int] = Array(1, 2, 4, 6, 8, 10)

a.filterWith(x=> x.toString)((a, b) =>  b == "2").collect
res34: Array[Int] = Array(5, 6)

flatMap

类似map方法，但是这个方法可以输出不止一个数据单元。

定义

def flatMap[U: ClassTag](f: T => TraversableOnce[U]): RDD[U]

Example

val a = sc.parallelize(1 to 10, 5)
a.flatMap(1 to _).collect
res47: Array[Int] = Array(1, 1, 2, 1, 2, 3, 1, 2, 3, 4, 1, 2, 3, 4, 5, 1, 2, 3, 4, 5, 6, 1, 2, 3, 4, 5, 6, 7, 1, 2, 3, 4, 5, 6, 7, 8, 1, 2, 3, 4, 5, 6, 7, 8, 9, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10)

sc.parallelize(List(1, 2, 3), 2).flatMap(x => List(x, x, x)).collect
res85: Array[Int] = Array(1, 1, 1, 2, 2, 2, 3, 3, 3)

// The program below generates a random number of copies (up to 10) of the items in the list.
val x  = sc.parallelize(1 to 10, 3)
x.flatMap(List.fill(scala.util.Random.nextInt(10))(_)).collect

res1: Array[Int] = Array(1, 2, 3, 3, 3, 4, 4, 4, 4, 4, 4, 4, 4, 4, 5, 5, 6, 6, 6, 6, 6, 6, 6, 6, 7, 7, 7, 8, 8, 8, 8, 8, 8, 8, 8, 9, 9, 9, 9, 9, 10, 10, 10, 10, 10, 10, 10, 10)

flatMapValues

和mapValues很相似，此方法要求value是TraversableOnce类型的，输出结果会把TraversableOnce中的元素打散输出多个tuple出来。

定义

def flatMapValues[U](f: V => TraversableOnce[U]): RDD[(K, U)]

Example

val a = sc.parallelize(List("dog", "tiger", "lion", "cat", "panther", "eagle"), 2)
val b = a.map(x => (x.length, x))
b.flatMapValues("x" + _ + "x").collect
res6: Array[(Int, Char)] = Array((3,x), (3,d), (3,o), (3,g), (3,x), (5,x), (5,t), (5,i), (5,g), (5,e), (5,r), (5,x), (4,x), (4,l), (4,i), (4,o), (4,n), (4,x), (3,x), (3,c), (3,a), (3,t), (3,x), (7,x), (7,p), (7,a), (7,n), (7,t), (7,h), (7,e), (7,r), (7,x), (5,x), (5,e), (5,a), (5,g), (5,l), (5,e), (5,x)) 

scala> b.flatMapValues{x=>x}.collect
res3: Array[(Int, Char)] = Array((3,d), (3,o), (3,g), (5,t), (5,i), (5,g), (5,e), (5,r), (4,l), (4,i), (4,o), (4,n), (3,c), (3,a), (3,t), (7,p), (7,a), (7,n), (7,t), (7,h), (7,e), (7,r), (5,e), (5,a), (5,g), (5,l), (5,e))

fold

聚合每个数据分片的数据。每个数据分片聚合之前可以设置一个初始值zeroValue。

定义

def fold(zeroValue: T)(op: (T, T) => T): T

Example

val a = sc.parallelize(List(1,2,3), 3)
a.fold(0)(_ + _)
res59: Int = 6

foldByKey [Pair]

功能类似fold，区别在于聚合范围不再是数据分别，而是每个key。输入的RDD数据必须是包含两个元素的tuple。

定义

def foldByKey(zeroValue: V)(func: (V, V) => V): RDD[(K, V)]

def foldByKey(zeroValue: V, numPartitions: Int)(func: (V, V) => V): RDD[(K, V)]

def foldByKey(zeroValue: V, partitioner: Partitioner)(func: (V, V) => V): RDD[(K, V)]

Example

val a = sc.parallelize(List("dog", "cat", "owl", "gnu", "ant"), 2)
val b = a.map(x => (x.length, x))
b.foldByKey("")(_ + _).collect
res84: Array[(Int, String)] = Array((3,dogcatowlgnuant)

val a = sc.parallelize(List("dog", "tiger", "lion", "cat", "panther", "eagle"), 2)
val b = a.map(x => (x.length, x))
b.foldByKey("")(_ + _).collect
res85: Array[(Int, String)] = Array((4,lion), (3,dogcat), (7,panther), (5,tigereagle))

foreach

针对每个数据单元执行一个无参数的方法

定义

def foreach(f: T => Unit)

Example

val c = sc.parallelize(List("cat", "dog", "tiger", "lion", "gnu", "crocodile", "ant", "whale", "dolphin", "spider"), 3)
c.foreach(x => println(x + "s are yummy"))
lions are yummy
gnus are yummy
crocodiles are yummy
ants are yummy
whales are yummy
dolphins are yummy
spiders are yummy

foreachPartition

针对每个数据分片执行当前输入的无参数方法，每个分片的数据元素通过迭代器访问。

定义

def foreachPartition(f: Iterator[T] => Unit)

Example

val b = sc.parallelize(List(1, 2, 3, 4, 5, 6, 7, 8, 9), 3)
b.foreachPartition(x => println(x.reduce(_ + _)))
6
15
24

getCheckpointFile

获取checkpoint文件，如果RDD没有被checkpoint则返回null

定义

def getCheckpointFile: Option[String]

Example

sc.setCheckpointDir("/home/cloudera/Documents")
val a = sc.parallelize(1 to 500, 5)
val b = a++a++a++a++a
b.getCheckpointFile
res49: Option[String] = None

b.checkpoint
b.getCheckpointFile
res54: Option[String] = None

b.collect
b.getCheckpointFile
res57: Option[String] = Some(file:/home/cloudera/Documents/cb978ffb-a346-4820-b3ba-d56580787b20/rdd-40)

glom

将一个Array的数据分别分组和数据分片个数相同的Array中去，并生成相应RDD。

定义

def glom(): RDD[Array[T]]

Example

val a = sc.parallelize(1 to 100, 3)
a.glom.collect
res8: Array[Array[Int]] = Array(Array(1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31, 32, 33), Array(34, 35, 36, 37, 38, 39, 40, 41, 42, 43, 44, 45, 46, 47, 48, 49, 50, 51, 52, 53, 54, 55, 56, 57, 58, 59, 60, 61, 62, 63, 64, 65, 66), Array(67, 68, 69, 70, 71, 72, 73, 74, 75, 76, 77, 78, 79, 80, 81, 82, 83, 84, 85, 86, 87, 88, 89, 90, 91, 92, 93, 94, 95, 96, 97, 98, 99, 100))

groupBy

仅仅将数据分组，分组的key通过传入的函数生成。

定义

def groupBy[K: ClassTag](f: T => K): RDD[(K, Iterable[T])]

def groupBy[K: ClassTag](f: T => K, numPartitions: Int): RDD[(K, Iterable[T])]

def groupBy[K: ClassTag](f: T => K, p: Partitioner): RDD[(K, Iterable[T])]

Example

val a = sc.parallelize(1 to 9, 3)
a.groupBy(x => { if (x % 2 == 0) "even" else "odd" }).collect
res42: Array[(String, Seq[Int])] = Array((even,ArrayBuffer(2, 4, 6, 8)), (odd,ArrayBuffer(1, 3, 5, 7, 9)))

val a = sc.parallelize(1 to 9, 3)
def myfunc(a: Int) : Int =
{
  a % 2
}
a.groupBy(myfunc).collect
res3: Array[(Int, Seq[Int])] = Array((0,ArrayBuffer(2, 4, 6, 8)), (1,ArrayBuffer(1, 3, 5, 7, 9)))

val a = sc.parallelize(1 to 9, 3)
def myfunc(a: Int) : Int =
{
  a % 2
}
a.groupBy(x => myfunc(x), 3).collect
a.groupBy(myfunc(_), 1).collect
res7: Array[(Int, Seq[Int])] = Array((0,ArrayBuffer(2, 4, 6, 8)), (1,ArrayBuffer(1, 3, 5, 7, 9)))

import org.apache.spark.Partitioner
class MyPartitioner extends Partitioner {
def numPartitions: Int = 2
def getPartition(key: Any): Int =
{
    key match
    {
      case null     => 0
      case key: Int => key          % numPartitions
      case _        => key.hashCode % numPartitions
    }
  }
  override def equals(other: Any): Boolean =
  {
    other match
    {
      case h: MyPartitioner => true
      case _                => false
    }
  }
}
val a = sc.parallelize(1 to 9, 3)
val p = new MyPartitioner()
val b = a.groupBy((x:Int) => { x }, p)
val c = b.mapWith(i => i)((a, b) => (b, a))
c.collect
res42: Array[(Int, (Int, Seq[Int]))] = Array((0,(4,ArrayBuffer(4))), (0,(2,ArrayBuffer(2))), (0,(6,ArrayBuffer(6))), (0,(8,ArrayBuffer(8))), (1,(9,ArrayBuffer(9))), (1,(3,ArrayBuffer(3))), (1,(1,ArrayBuffer(1))), (1,(7,ArrayBuffer(7))), (1,(5,ArrayBuffer(5))))

groupByKey [Pair]

相比较于groupBy, groupByKey不需要传入生成key的函数，而仅将输入二元tuple中的第一个字段当成key进行分组。

定义

def groupByKey(): RDD[(K, Iterable[V])]

def groupByKey(numPartitions: Int): RDD[(K, Iterable[V])]

def groupByKey(partitioner: Partitioner): RDD[(K, Iterable[V])]

Example

val a = sc.parallelize(List("dog", "tiger", "lion", "cat", "spider", "eagle"), 2)
val b = a.keyBy(_.length)
b.groupByKey.collect
res11: Array[(Int, Seq[String])] = Array((4,ArrayBuffer(lion)), (6,ArrayBuffer(spider)), (3,ArrayBuffer(dog, cat)), (5,ArrayBuffer(tiger, eagle)))

histogram [Double]

针对数据类型为Double的RDD统计直方图。可以输入分桶数进行平均分桶，也可以输入自定义的分桶区间。平均分桶情况下会输出两个数组，第一个是每个分桶边界值，第二个是每个分桶的统计数。自定义分桶情况下只输出一个数组，即每个分桶的统计数。

定义

def histogram(bucketCount: Int): Pair[Array[Double], Array[Long]]def histogram(buckets: Array[Double], evenBuckets: Boolean = false): Array[Long]

Example with even spacing

val a = sc.parallelize(List(1.1, 1.2, 1.3, 2.0, 2.1, 7.4, 7.5, 7.6, 8.8, 9.0), 3)
a.histogram(5)
res11: (Array[Double], Array[Long]) = (Array(1.1, 2.68, 4.26, 5.84, 7.42, 9.0),Array(5, 0, 0, 1, 4))

val a = sc.parallelize(List(9.1, 1.0, 1.2, 2.1, 1.3, 5.0, 2.0, 2.1, 7.4, 7.5, 7.6, 8.8, 10.0, 8.9, 5.5), 3)
a.histogram(6)
res18: (Array[Double], Array[Long]) = (Array(1.0, 2.5, 4.0, 5.5, 7.0, 8.5, 10.0),Array(6, 0, 1, 1, 3, 4))

Example with custom spacing

val a = sc.parallelize(List(1.1, 1.2, 1.3, 2.0, 2.1, 7.4, 7.5, 7.6, 8.8, 9.0), 3)
a.histogram(Array(0.0, 3.0, 8.0))
res14: Array[Long] = Array(5, 3)

val a = sc.parallelize(List(9.1, 1.0, 1.2, 2.1, 1.3, 5.0, 2.0, 2.1, 7.4, 7.5, 7.6, 8.8, 10.0, 8.9, 5.5), 3)
a.histogram(Array(0.0, 5.0, 10.0))
res1: Array[Long] = Array(6, 9)

a.histogram(Array(0.0, 5.0, 10.0, 15.0))
res1: Array[Long] = Array(6, 8, 1)

id

查询分配给RDD的id号

定义

val id: Int

Example

val y = sc.parallelize(1 to 10, 10)
y.id
res16: Int = 19

intersection

取两个RDD的交集

定义

def intersection(other: RDD[T], numPartitions: Int): RDD[T]

def intersection(other: RDD[T], partitioner: Partitioner)(implicit ord: Ordering[T] = null): RDD[T]

def intersection(other: RDD[T]): RDD[T]

val x = sc.parallelize(1 to 20)
val y = sc.parallelize(10 to 30)
val z = x.intersection(y)

z.collect
res74: Array[Int] = Array(16, 12, 20, 13, 17, 14, 18, 10, 19, 15, 11)

isCheckpointed

显示这个RDD是否被checkpoint了。谨记只有当执行了action操作之后相应的数据才会被checkpoint。

定义

def isCheckpointed: Boolean

Example

sc.setCheckpointDir("/home/cloudera/Documents")
c.isCheckpointed
res6: Boolean = false

c.checkpoint
c.isCheckpointed
res8: Boolean = false

c.collect
c.isCheckpointed
res9: Boolean = true

join [Pair]

对两个key-value的RDD执行内连接。

定义

def join[W](other: RDD[(K, W)]): RDD[(K, (V, W))]

def join[W](other: RDD[(K, W)], numPartitions: Int): RDD[(K, (V, W))]

def join[W](other: RDD[(K, W)], partitioner: Partitioner): RDD[(K, (V, W))]

Example

val a = sc.parallelize(List("dog", "salmon", "salmon", "rat", "elephant"), 3)
val b = a.keyBy(_.length)
val c = sc.parallelize(List("dog","cat","gnu","salmon","rabbit","turkey","wolf","bear","bee"), 3)
val d = c.keyBy(_.length)
b.join(d).collect

res17: Array[(Int, (String, String))] = Array((6,(salmon,salmon)), (6,(salmon,rabbit)), (6,(salmon,turkey)), (6,(rabbit,salmon)), (6,(rabbit,rabbit)), (6,(rabbit,turkey)), (6,(turkey,salmon)), (6,(turkey,rabbit)), (6,(turkey,turkey)), (3,(dog,dog)), (3,(dog,cat)), (3,(dog,gnu)), (3,(dog,bee)), (3,(cat,dog)), (3,(cat,cat)), (3,(cat,gnu)), (3,(cat,bee)), (3,(gnu,dog)), (3,(gnu,cat)), (3,(gnu,gnu)), (3,(gnu,bee)), (3,(bee,dog)), (3,(bee,cat)), (3,(bee,gnu)), (3,(bee,bee)), (4,(wolf,wolf)), (4,(wolf,bear)), (4,(bear,wolf)), (4,(bear,bear)))

keyBy

为RDD中的每个数据单元增加一个key，即生成key-value对。输入的函数是key的构造函数。

定义

def keyBy[K](f: T => K): RDD[(K, T)]

Example

val a = sc.parallelize(List("dog", "salmon", "salmon", "rat", "elephant"), 3)
val b = a.keyBy(_.length)
b.collect
res26: Array[(Int, String)] = Array((3,dog), (6,salmon), (6,salmon), (3,rat), (8,elephant))

keys [Pair]

从RDD中抽取所有元素的key并生成新的RDD。

定义

def keys: RDD[K]

Example

val a = sc.parallelize(List("dog", "tiger", "lion", "cat", "panther", "eagle"), 2)
val b = a.map(x => (x.length, x))
b.keys.collect
res2: Array[Int] = Array(3, 5, 4, 3, 7, 5)

leftOuterJoin [Pair]

针对两个key-value数据形式的RDD进行左外连接操作。

定义

def leftOuterJoin[W](other: RDD[(K, W)]): RDD[(K, (V, Option[W]))]

def leftOuterJoin[W](other: RDD[(K, W)], numPartitions: Int): RDD[(K, (V, Option[W]))]

def leftOuterJoin[W](other: RDD[(K, W)], partitioner: Partitioner): RDD[(K, (V, Option[W]))]

Example

val a = sc.parallelize(List("dog", "salmon", "salmon", "rat", "elephant"), 3)
val b = a.keyBy(_.length)
val c = sc.parallelize(List("dog","cat","gnu","salmon","rabbit","turkey","wolf","bear","bee"), 3)
val d = c.keyBy(_.length)
b.leftOuterJoin(d).collect

res1: Array[(Int, (String, Option[String]))] = Array((6,(salmon,Some(salmon))), (6,(salmon,Some(rabbit))), (6,(salmon,Some(turkey))), (6,(salmon,Some(salmon))), (6,(salmon,Some(rabbit))), (6,(salmon,Some(turkey))), (3,(dog,Some(dog))), (3,(dog,Some(cat))), (3,(dog,Some(gnu))), (3,(dog,Some(bee))), (3,(rat,Some(dog))), (3,(rat,Some(cat))), (3,(rat,Some(gnu))), (3,(rat,Some(bee))), (8,(elephant,None)))

lookup

遍历RDD中所有的key，找到符合输入key的value，并输出scala的seq数据。

定义

def lookup(key: K): Seq[V]

Example

val a = sc.parallelize(List("dog", "tiger", "lion", "cat", "panther", "eagle"), 2)
val b = a.map(x => (x.length, x))
b.lookup(5)
res0: Seq[String] = WrappedArray(tiger, eagle)

map

针对RDD中的每个数据元素执行map函数，并返回处理后的新RDD。

定义

def map[U: ClassTag](f: T => U): RDD[U]

Example

val a = sc.parallelize(List("dog", "salmon", "salmon", "rat", "elephant"), 3)
val b = a.map(_.length)
val c = a.zip(b)
c.collect
res0: Array[(String, Int)] = Array((dog,3), (salmon,6), (salmon,6), (rat,3), (elephant,8))

mapPartitions

这是个特殊的map，它在每个数据分片上只执行一次。分片上的数据作为数据序列输入到处理函数，输入数据类型是迭代器(Iterarator[T])。处理函数返回另外一个迭代器。最终mapPartitions会调用合并程序将多个数据分片返回的数据合并成一个新的RDD。

定义

def mapPartitions[U: ClassTag](f: Iterator[T] => Iterator[U], preservesPartitioning: Boolean = false): RDD[U]

Example 1

val a = sc.parallelize(1 to 9, 3)
def myfunc[T](iter: Iterator[T]) : Iterator[(T, T)] = {
  var res = List[(T, T)]()
  var pre = iter.next
  while (iter.hasNext)
  {
    val cur = iter.next;
    res .::= (pre, cur)
    pre = cur;
  }
  res.iterator
}
a.mapPartitions(myfunc).collect
res0: Array[(Int, Int)] = Array((2,3), (1,2), (5,6), (4,5), (8,9), (7,8))

Example 2

val x = sc.parallelize(List(1, 2, 3, 4, 5, 6, 7, 8, 9,10), 3)
def myfunc(iter: Iterator[Int]) : Iterator[Int] = {
  var res = List[Int]()
  while (iter.hasNext) {
    val cur = iter.next;
    res = res ::: List.fill(scala.util.Random.nextInt(10))(cur)
  }
  res.iterator
}
x.mapPartitions(myfunc).collect
// some of the number are not outputted at all. This is because the random number generated for it is zero.
res8: Array[Int] = Array(1, 2, 2, 2, 2, 3, 3, 3, 3, 3, 3, 3, 3, 3, 4, 4, 4, 4, 4, 4, 4, 5, 7, 7, 7, 9, 9, 10)

上面的示例也可以用 flatMap 实现，示例如下：

Example 2 using flatmap

val x  = sc.parallelize(1 to 10, 3)
x.flatMap(List.fill(scala.util.Random.nextInt(10))(_)).collect

res1: Array[Int] = Array(1, 2, 3, 3, 3, 4, 4, 4, 4, 4, 4, 4, 4, 4, 5, 5, 6, 6, 6, 6, 6, 6, 6, 6, 7, 7, 7, 8, 8, 8, 8, 8, 8, 8, 8, 9, 9, 9, 9, 9, 10, 10, 10, 10, 10, 10, 10, 10)

mapPartitionsWithContext (developer API)

功能类似于mapPartitions，但是处理函数会带有当前节点的相关信息。

定义

def mapPartitionsWithContext[U: ClassTag](f: (TaskContext, Iterator[T]) => Iterator[U], preservesPartitioning: Boolean = false): RDD[U]

Example

val a = sc.parallelize(1 to 9, 3)
import org.apache.spark.TaskContext
def myfunc(tc: TaskContext, iter: Iterator[Int]) : Iterator[Int] = {
  tc.addOnCompleteCallback(() => println(
    "Partition: "     + tc.partitionId +
    ", AttemptID: "   + tc.attemptId ))
  
  iter.toList.filter(_ % 2 == 0).iterator
}
a.mapPartitionsWithContext(myfunc).collect

14/04/01 23:05:48 INFO SparkContext: Starting job: collect at 

  
   
 :20
...
14/04/01 23:05:48 INFO Executor: Running task ID 0
Partition: 0, AttemptID: 0, Interrupted: false
...
14/04/01 23:05:48 INFO Executor: Running task ID 1
14/04/01 23:05:48 INFO TaskSetManager: Finished TID 0 in 470 ms on localhost (progress: 0/3)
...
14/04/01 23:05:48 INFO Executor: Running task ID 2
14/04/01 23:05:48 INFO TaskSetManager: Finished TID 1 in 23 ms on localhost (progress: 1/3)
14/04/01 23:05:48 INFO DAGScheduler: Completed ResultTask(0, 1)

?
res0: Array[Int] = Array(2, 6, 4, 8)


  

mapPartitionsWithIndex

功能和mapPartitions类似，但是有两个输入参数。第一个参数是数据分片的索引值，第二个参数是当前分片数据的迭代器。输出是当前分片数据处理后的数据迭代器。

定义

def mapPartitionsWithIndex[U: ClassTag](f: (Int, Iterator[T]) => Iterator[U], preservesPartitioning: Boolean = false): RDD[U]

Example

val x = sc.parallelize(List(1,2,3,4,5,6,7,8,9,10), 3)
def myfunc(index: Int, iter: Iterator[Int]) : Iterator[String] = {
  iter.toList.map(x => index + "," + x).iterator
}
x.mapPartitionsWithIndex(myfunc).collect()
res10: Array[String] = Array(0,1, 0,2, 0,3, 1,4, 1,5, 1,6, 2,7, 2,8, 2,9, 2,10)

mapValues [Pair]

将输入的二元tuple数据的value值逐一转换处理，并输出包含原有key和处理后value的二元tuple，最终输出处理后二元tuple的RDD。

定义

def mapValues[U](f: V => U): RDD[(K, U)]

Example

val a = sc.parallelize(List("dog", "tiger", "lion", "cat", "panther", "eagle"), 2)
val b = a.map(x => (x.length, x))
b.mapValues("x" + _ + "x").collect
res5: Array[(Int, String)] = Array((3,xdogx), (5,xtigerx), (4,xlionx), (3,xcatx), (7,xpantherx), (5,xeaglex))

max

返回RDD中最大的元素。

定义

def max()(implicit ord: Ordering[T]): T

Example

val y = sc.parallelize(10 to 30)
y.max
res75: Int = 30

mean [Double], meanApprox [Double]

计算RDD中的均值，meanApprox提供近似算法，提高计算速度但是不精确。

定义

def mean(): Doubledef meanApprox(timeout: Long, confidence: Double = 0.95): PartialResult[BoundedDouble]

Example

val a = sc.parallelize(List(9.1, 1.0, 1.2, 2.1, 1.3, 5.0, 2.0, 2.1, 7.4, 7.5, 7.6, 8.8, 10.0, 8.9, 5.5), 3)
a.mean
res0: Double = 5.3

min

返回RDD中最小的元素

定义

def min()(implicit ord: Ordering[T]): T

Example

val y = sc.parallelize(10 to 30)
y.min
res75: Int = 10 

val y = sc.parallelize(Array("1","2","!","a"))
y.min
res4: String = !

name, setName

给RDD打标签

定义

@transient var name: Stringdef setName(_name: String)

Example

val y = sc.parallelize(1 to 10, 10)
y.name
res13: String = null
y.setName("Fancy RDD Name")
y.name
res15: String = Fancy RDD Name

partitionBy [Pair]

将数据重新分片

定义

def partitionBy(partitioner: Partitioner): RDD[(K, V)]

partitioner

设置默认的partitioner，这个默认partitioner会在groupBy,subtract,reduceByKey等方法中用到。

定义

@transient val partitioner: Option[Partitioner]

partitions

返回与当前RDD关联的partition对象。

定义

final def partitions: Array[Partition]

Example

val b = sc.parallelize(List("Gnu", "Cat", "Rat", "Dog", "Gnu", "Rat"), 2)
b.partitions
res48: Array[org.apache.spark.Partition] = Array(org.apache.spark.rdd.ParallelCollectionPartition@18aa, org.apache.spark.rdd.ParallelCollectionPartition@18ab)

persist, cache

这个函数可以用来调整RDD的存储等级。释放内存的时候，Spark需要根据这些存储等级标签决定释放谁。persist()和cache()是persist(StorageLevel.MEMORY_ONLY)的缩写。（Warning:一旦存储等级被改变，则它不能被再次改变！）

定义

def cache(): RDD[T]

def persist(): RDD[T]

def persist(newLevel: StorageLevel): RDD[T]

Example

val c = sc.parallelize(List("Gnu", "Cat", "Rat", "Dog", "Gnu", "Rat"), 2)
c.getStorageLevel
res0: org.apache.spark.storage.StorageLevel = StorageLevel(false, false, false, false, 1)
c.cache
c.getStorageLevel
res2: org.apache.spark.storage.StorageLevel = StorageLevel(false, true, false, true, 1)

pipe

将RDD的每个数据分片都接到shell-command的标准输入上。经过shell-command的输出数据会重新生成新的RDD，新RDD是string类型的RDD。

定义

def pipe(command: String): RDD[String]

def pipe(command: String, env: Map[String, String]): RDD[String]

def pipe(command: Seq[String], env: Map[String, String] = Map(), printPipeContext: (String => Unit) => Unit = null, printRDDElement: (T, String => Unit) => Unit = null): RDD[String]

Example

val a = sc.parallelize(1 to 9, 3)
a.pipe("head -n 1").collect
res2: Array[String] = Array(1, 4, 7)

randomSplit

数据集切分方法。随机将RDD的数据切分到多个小RDD中。切分比率由输入的权重Array确定。切分不会严格和输入的比率相同，同时也可以设定随机seed。

定义

def randomSplit(weights: Array[Double], seed: Long = Utils.random.nextLong): Array[RDD[T]]

Example

val y = sc.parallelize(1 to 10)
val splits = y.randomSplit(Array(0.6, 0.4), seed = 11L)
val training = splits(0)
val test = splits(1)
training.collect
res:85 Array[Int] = Array(1, 4, 5, 6, 8, 10)
test.collect
res86: Array[Int] = Array(2, 3, 7, 9)

val y = sc.parallelize(1 to 10)
val splits = y.randomSplit(Array(0.1, 0.3, 0.6))

val rdd1 = splits(0)
val rdd2 = splits(1)
val rdd3 = splits(2)

rdd1.collect
res87: Array[Int] = Array(4, 10)
rdd2.collect
res88: Array[Int] = Array(1, 3, 5, 8)
rdd3.collect
res91: Array[Int] = Array(2, 6, 7, 9)

reduce

这个方法是Spark中很有名的方法，需要注意的是，其中输入的f的输入输出参数类型一致。

定义

def reduce(f: (T, T) => T): T

Example

val a = sc.parallelize(1 to 100, 3)
a.reduce(_ + _)
res41: Int = 5050

reduceByKey [Pair], reduceByKeyLocally [Pair]

相对于reduce方法，这个方法会在同一key下进行reduce操作。reduceByKeyLocally与reduceByKey的区别是前者将数据返回到master节点中，并返回map结果，后者只返回RDD结果。

定义

def reduceByKey(func: (V, V) => V): RDD[(K, V)]

def reduceByKey(func: (V, V) => V, numPartitions: Int): RDD[(K, V)]

def reduceByKey(partitioner: Partitioner, func: (V, V) => V): RDD[(K, V)]

def reduceByKeyLocally(func: (V, V) => V): Map[K, V]

Example

val a = sc.parallelize(List("dog", "cat", "owl", "gnu", "ant"), 2)
val b = a.map(x => (x.length, x))
b.reduceByKey(_ + _).collect
res86: Array[(Int, String)] = Array((3,dogcatowlgnuant))

val a = sc.parallelize(List("dog", "tiger", "lion", "cat", "panther", "eagle"), 2)
val b = a.map(x => (x.length, x))
b.reduceByKey(_ + _).collect
res87: Array[(Int, String)] = Array((4,lion), (3,dogcat), (7,panther), (5,tigereagle))

rightOuterJoin [Pair]

执行两个(key-value)RDD的右外连接。

定义

def rightOuterJoin[W](other: RDD[(K, W)]): RDD[(K, (Option[V], W))]

def rightOuterJoin[W](other: RDD[(K, W)], numPartitions: Int): RDD[(K, (Option[V], W))]

def rightOuterJoin[W](other: RDD[(K, W)], partitioner: Partitioner): RDD[(K, (Option[V], W))]

Example

val a = sc.parallelize(List("dog", "salmon", "salmon", "rat", "elephant"), 3)
val b = a.keyBy(_.length)
val c = sc.parallelize(List("dog","cat","gnu","salmon","rabbit","turkey","wolf","bear","bee"), 3)
val d = c.keyBy(_.length)
b.rightOuterJoin(d).collect

res2: Array[(Int, (Option[String], String))] = Array((6,(Some(salmon),salmon)), (6,(Some(salmon),rabbit)), (6,(Some(salmon),turkey)), (6,(Some(salmon),salmon)), (6,(Some(salmon),rabbit)), (6,(Some(salmon),turkey)), (3,(Some(dog),dog)), (3,(Some(dog),cat)), (3,(Some(dog),gnu)), (3,(Some(dog),bee)), (3,(Some(rat),dog)), (3,(Some(rat),cat)), (3,(Some(rat),gnu)), (3,(Some(rat),bee)), (4,(None,wolf)), (4,(None,bear)))

sample

随机取样

定义

def sample(withReplacement: Boolean, fraction: Double, seed: Int): RDD[T]

Example

val a = sc.parallelize(1 to 10000, 3)
a.sample(false, 0.1, 0).count
res24: Long = 960

a.sample(true, 0.3, 0).count
res25: Long = 2888

a.sample(true, 0.3, 13).count
res26: Long = 2985

saveAsObjectFile

将RDD存入二进制文件

定义

def saveAsObjectFile(path: String)

val x = sc.parallelize(1 to 100, 3)
x.saveAsObjectFile("objFile")
val y = sc.objectFile[Array[Int]]("objFile")
y.collect
res52: Array[Int] = Array(67, 68, 69, 70, 71, 72, 73, 74, 75, 76, 77, 78, 79, 80, 81, 82, 83, 84, 85, 86, 87, 88, 89, 90, 91, 92, 93, 94, 95, 96, 97, 98, 99, 100, 34, 35, 36, 37, 38, 39, 40, 41, 42, 43, 44, 45, 46, 47, 48, 49, 50, 51, 52, 53, 54, 55, 56, 57, 58, 59, 60, 61, 62, 63, 64, 65, 66, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31, 32, 33)

saveAsSequenceFile [SeqFile]

保存RDD为一个Hadoop的sequence文件

定义

def saveAsSequenceFile(path: String, codec: Option[Class[_ <: CompressionCodec]] = None)

Example

val v = sc.parallelize(Array(("owl",3), ("gnu",4), ("dog",1), ("cat",2), ("ant",5)), 2)
v.saveAsSequenceFile("hd_seq_file")
14/04/19 05:45:43 INFO FileOutputCommitter: Saved output of task 'attempt_201404190545_0000_m_000001_191' to file:/home/cloudera/hd_seq_file

[cloudera@localhost ~]$ ll ~/hd_seq_file
total 8
-rwxr-xr-x 1 cloudera cloudera 117 Apr 19 05:45 part-00000
-rwxr-xr-x 1 cloudera cloudera 133 Apr 19 05:45 part-00001
-rwxr-xr-x 1 cloudera cloudera   0 Apr 19 05:45 _SUCCESS

saveAsTextFile

保存RDD为文本文件。一次一行

定义

def saveAsTextFile(path: String)def saveAsTextFile(path: String, codec: Class[_ <: CompressionCodec])

Example without compression

val a = sc.parallelize(1 to 10000, 3)
a.saveAsTextFile("mydata_a")
14/04/03 21:11:36 INFO FileOutputCommitter: Saved output of task 'attempt_201404032111_0000_m_000002_71' to file:/home/cloudera/Documents/spark-0.9.0-incubating-bin-cdh4/bin/mydata_a


[cloudera@localhost ~]$ head -n 5 ~/Documents/spark-0.9.0-incubating-bin-cdh4/bin/mydata_a/part-00000
1
2
3
4
5

// Produces 3 output files since we have created the a RDD with 3 partitions
[cloudera@localhost ~]$ ll ~/Documents/spark-0.9.0-incubating-bin-cdh4/bin/mydata_a/
-rwxr-xr-x 1 cloudera cloudera 15558 Apr  3 21:11 part-00000
-rwxr-xr-x 1 cloudera cloudera 16665 Apr  3 21:11 part-00001
-rwxr-xr-x 1 cloudera cloudera 16671 Apr  3 21:11 part-00002

Example with compression

import org.apache.hadoop.io.compress.GzipCodec
a.saveAsTextFile("mydata_b", classOf[GzipCodec])

[cloudera@localhost ~]$ ll ~/Documents/spark-0.9.0-incubating-bin-cdh4/bin/mydata_b/
total 24
-rwxr-xr-x 1 cloudera cloudera 7276 Apr  3 21:29 part-00000.gz
-rwxr-xr-x 1 cloudera cloudera 6517 Apr  3 21:29 part-00001.gz
-rwxr-xr-x 1 cloudera cloudera 6525 Apr  3 21:29 part-00002.gz

val x = sc.textFile("mydata_b")
x.count
res2: Long = 10000

Example writing into HDFS

val x = sc.parallelize(List(1,2,3,4,5,6,6,7,9,8,10,21), 3)
x.saveAsTextFile("hdfs://localhost:8020/user/cloudera/test");

val sp = sc.textFile("hdfs://localhost:8020/user/cloudera/sp_data")
sp.flatMap(_.split(" ")).saveAsTextFile("hdfs://localhost:8020/user/cloudera/sp_x")

stats [Double]

同时计算均值，方差，以及标准差。

定义

def stats(): StatCounter

Example

val x = sc.parallelize(List(1.0, 2.0, 3.0, 5.0, 20.0, 19.02, 19.29, 11.09, 21.0), 2)
x.stats
res16: org.apache.spark.util.StatCounter = (count: 9, mean: 11.266667, stdev: 8.126859)

sortBy

这个方法将RDD的数据排序并存入新的RDD，第一个参数传入方法指定排序key，第二个参数指定是逆序还是顺序。

定义

def sortBy[K](f: (T) ⇒ K, ascending: Boolean = true, numPartitions: Int = this.partitions.size)(implicit ord: Ordering[K], ctag: ClassTag[K]): RDD[T]

Example

val y = sc.parallelize(Array(5, 7, 1, 3, 2, 1))
y.sortBy(c => c, true).collect
res101: Array[Int] = Array(1, 1, 2, 3, 5, 7)

y.sortBy(c => c, false).collect
res102: Array[Int] = Array(7, 5, 3, 2, 1, 1)

val z = sc.parallelize(Array(("H", 10), ("A", 26), ("Z", 1), ("L", 5)))
z.sortBy(c => c._1, true).collect
res109: Array[(String, Int)] = Array((A,26), (H,10), (L,5), (Z,1))

z.sortBy(c => c._2, true).collect
res108: Array[(String, Int)] = Array((Z,1), (L,5), (H,10), (A,26))

sortByKey [Ordered]

这个方法按Key进行排序，并输出新RDD。输出的RDD数据是经过shuffled之后的，因为在在输出的reducer里数据已经被shuffle了。

定义

def sortByKey(ascending: Boolean = true, numPartitions: Int = self.partitions.size): RDD[P]

Example

val a = sc.parallelize(List("dog", "cat", "owl", "gnu", "ant"), 2)
val b = sc.parallelize(1 to a.count.toInt, 2)
val c = a.zip(b)
c.sortByKey(true).collect
res74: Array[(String, Int)] = Array((ant,5), (cat,2), (dog,1), (gnu,4), (owl,3))
c.sortByKey(false).collect
res75: Array[(String, Int)] = Array((owl,3), (gnu,4), (dog,1), (cat,2), (ant,5))

val a = sc.parallelize(1 to 100, 5)
val b = a.cartesian(a)
val c = sc.parallelize(b.takeSample(true, 5, 13), 2)
val d = c.sortByKey(false)
res56: Array[(Int, Int)] = Array((96,9), (84,76), (59,59), (53,65), (52,4))

stdev [Double], sampleStdev [Double]

调用stats方法并抽取其中stdev结果或者sampleStdev结果。

定义

def stdev(): Doubledef sampleStdev(): Double

Example

val d = sc.parallelize(List(0.0, 0.0, 0.0), 3)
d.stdev
res10: Double = 0.0
d.sampleStdev
res11: Double = 0.0

val d = sc.parallelize(List(0.0, 1.0), 3)
d.stdev
d.sampleStdev
res18: Double = 0.5
res19: Double = 0.7071067811865476

val d = sc.parallelize(List(0.0, 0.0, 1.0), 3)
d.stdev
res14: Double = 0.4714045207910317
d.sampleStdev
res15: Double = 0.5773502691896257

subtract

实现差集

定义

def subtract(other: RDD[T]): RDD[T]

def subtract(other: RDD[T], numPartitions: Int): RDD[T]

def subtract(other: RDD[T], p: Partitioner): RDD[T]

Example

val a = sc.parallelize(1 to 9, 3)
val b = sc.parallelize(1 to 3, 3)
val c = a.subtract(b)
c.collect
res3: Array[Int] = Array(6, 9, 4, 7, 5, 8)

subtractByKey [Pair]

实现差集，但按key求差。

定义

def subtractByKey[W: ClassTag](other: RDD[(K, W)]): RDD[(K, V)]

def subtractByKey[W: ClassTag](other: RDD[(K, W)], numPartitions: Int): RDD[(K, V)]

def subtractByKey[W: ClassTag](other: RDD[(K, W)], p: Partitioner): RDD[(K, V)]

Example

val a = sc.parallelize(List("dog", "tiger", "lion", "cat", "spider", "eagle"), 2)
val b = a.keyBy(_.length)
val c = sc.parallelize(List("ant", "falcon", "squid"), 2)
val d = c.keyBy(_.length)
b.subtractByKey(d).collect
res15: Array[(Int, String)] = Array((4,lion))

sum [Double], sumApprox [Double]

求RDD中所有value的和值，sumApprox是近似方法，不精确但速度快。

定义

def sum(): Doubledef sumApprox(timeout: Long, confidence: Double = 0.95): PartialResult[BoundedDouble]

Example

val x = sc.parallelize(List(1.0, 2.0, 3.0, 5.0, 20.0, 19.02, 19.29, 11.09, 21.0), 2)
x.sum
res17: Double = 101.39999999999999

take

将RDD中前n个数据单元抽取出来作为array返回。

定义

def take(num: Int): Array[T]

Example

val b = sc.parallelize(List("dog", "cat", "ape", "salmon", "gnu"), 2)
b.take(2)
res18: Array[String] = Array(dog, cat)

val b = sc.parallelize(1 to 10000, 5000)
b.take(100)
res6: Array[Int] = Array(1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31, 32, 33, 34, 35, 36, 37, 38, 39, 40, 41, 42, 43, 44, 45, 46, 47, 48, 49, 50, 51, 52, 53, 54, 55, 56, 57, 58, 59, 60, 61, 62, 63, 64, 65, 66, 67, 68, 69, 70, 71, 72, 73, 74, 75, 76, 77, 78, 79, 80, 81, 82, 83, 84, 85, 86, 87, 88, 89, 90, 91, 92, 93, 94, 95, 96, 97, 98, 99, 100)

takeOrdered

用RDD数据单元本身隐含的排序方法进行排序，排序后返回前n个元素组成的array。

定义

def takeOrdered(num: Int)(implicit ord: Ordering[T]): Array[T]

Example

val b = sc.parallelize(List("dog", "cat", "ape", "salmon", "gnu"), 2)
b.takeOrdered(2)
res19: Array[String] = Array(ape, cat)

takeSample

以下是和sample的区别：

* 返回具体个数的样本（第二个参数指定）

* 直接返回array而不是RDD

* 内部会将返回结果随机打散

定义

def takeSample(withReplacement: Boolean, num: Int, seed: Int): Array[T]

Example

val x = sc.parallelize(1 to 1000, 3)
x.takeSample(true, 100, 1)
res3: Array[Int] = Array(339, 718, 810, 105, 71, 268, 333, 360, 341, 300, 68, 848, 431, 449, 773, 172, 802, 339, 431, 285, 937, 301, 167, 69, 330, 864, 40, 645, 65, 349, 613, 468, 982, 314, 160, 675, 232, 794, 577, 571, 805, 317, 136, 860, 522, 45, 628, 178, 321, 482, 657, 114, 332, 728, 901, 290, 175, 876, 227, 130, 863, 773, 559, 301, 694, 460, 839, 952, 664, 851, 260, 729, 823, 880, 792, 964, 614, 821, 683, 364, 80, 875, 813, 951, 663, 344, 546, 918, 436, 451, 397, 670, 756, 512, 391, 70, 213, 896, 123, 858)

toDebugString

以string的形式返回RDD和它所依赖的RDD的调试信息

定义

def toDebugString: String

Example

val a = sc.parallelize(1 to 9, 3)
val b = sc.parallelize(1 to 3, 3)
val c = a.subtract(b)
c.toDebugString
res6: String = 
MappedRDD[15] at subtract at 

  
   
 :16 (3 partitions)
  SubtractedRDD[14] at subtract at 
 
   
    
  :16 (3 partitions)
    MappedRDD[12] at subtract at 
  
    
     
   :16 (3 partitions)
      ParallelCollectionRDD[10] at parallelize at 
   
     
      
    :12 (3 partitions)
    MappedRDD[13] at subtract at 
    
      
       
     :16 (3 partitions)
      ParallelCollectionRDD[11] at parallelize at 
     
       
        
      :12 (3 partitions) 


     
       
    
      
   
     
  
    
 
   

  

toJavaRDD

将RDD对象嵌入JavaRDD对象中并返回JavaRDD。

定义

def toJavaRDD() : JavaRDD[T]

Example

val c = sc.parallelize(List("Gnu", "Cat", "Rat", "Dog"), 2)
c.toJavaRDD
res3: org.apache.spark.api.java.JavaRDD[String] = ParallelCollectionRDD[6] at parallelize at 

  
   
 :12 



  

top

依着输入数据默认的排序机制，取出排序后的前n个value，并组成array返回。

定义

ddef top(num: Int)(implicit ord: Ordering[T]): Array[T]

Example

val c = sc.parallelize(Array(6, 9, 4, 7, 5, 8), 2)
c.top(2)
res28: Array[Int] = Array(9, 8)

toString

将RDD组装出一个可读的文本。

定义

override def toString: String

Example

val a = sc.parallelize(1 to 9, 3)
val b = sc.parallelize(1 to 3, 3)
val c = a.subtract(b)
c.toString
res7: String = MappedRDD[15] at subtract at 

  
   
 :16


  

union, ++

执行集合合并操作

定义

def ++(other: RDD[T]): RDD[T]def union(other: RDD[T]): RDD[T]

Example

val a = sc.parallelize(1 to 3, 1)
val b = sc.parallelize(5 to 7, 1)
(a ++ b).collect
res0: Array[Int] = Array(1, 2, 3, 5, 6, 7)

unpersist

释放RDD，将存入磁盘和内存的数据元素释放。但是RDD对象会被保留，在后续使用它的时候，Spark会重新依据依赖图计算。

定义

def unpersist(blocking: Boolean = true): RDD[T]

Example

val y = sc.parallelize(1 to 10, 10)
val z = (y++y)
z.collect
z.unpersist(true)
14/04/19 03:04:57 INFO UnionRDD: Removing RDD 22 from persistence list
14/04/19 03:04:57 INFO BlockManager: Removing RDD 22

values

抽取所有tuple中的value并组装成新的RDD返回。

定义

def values: RDD[V]

Example

val a = sc.parallelize(List("dog", "tiger", "lion", "cat", "panther", "eagle"), 2)
val b = a.map(x => (x.length, x))
b.values.collect
res3: Array[String] = Array(dog, tiger, lion, cat, panther, eagle)

variance [Double], sampleVariance [Double]

调用stats方法，并返回其中variance值或者sampleVariance值

定义

def variance(): Doubledef sampleVariance(): Double

Example

val a = sc.parallelize(List(9.1, 1.0, 1.2, 2.1, 1.3, 5.0, 2.0, 2.1, 7.4, 7.5, 7.6, 8.8, 10.0, 8.9, 5.5), 3)
a.variance
res70: Double = 10.605333333333332

val x = sc.parallelize(List(1.0, 2.0, 3.0, 5.0, 20.0, 19.02, 19.29, 11.09, 21.0), 2)
x.variance
res14: Double = 66.04584444444443

x.sampleVariance
res13: Double = 74.30157499999999

zip

将两个RDD中第i个元素组成一个tuple，进而形成(key-value)的PairRDD。

定义

def zip[U: ClassTag](other: RDD[U]): RDD[(T, U)]

Example

val a = sc.parallelize(1 to 100, 3)
val b = sc.parallelize(101 to 200, 3)
a.zip(b).collect
res1: Array[(Int, Int)] = Array((1,101), (2,102), (3,103), (4,104), (5,105), (6,106), (7,107), (8,108), (9,109), (10,110), (11,111), (12,112), (13,113), (14,114), (15,115), (16,116), (17,117), (18,118), (19,119), (20,120), (21,121), (22,122), (23,123), (24,124), (25,125), (26,126), (27,127), (28,128), (29,129), (30,130), (31,131), (32,132), (33,133), (34,134), (35,135), (36,136), (37,137), (38,138), (39,139), (40,140), (41,141), (42,142), (43,143), (44,144), (45,145), (46,146), (47,147), (48,148), (49,149), (50,150), (51,151), (52,152), (53,153), (54,154), (55,155), (56,156), (57,157), (58,158), (59,159), (60,160), (61,161), (62,162), (63,163), (64,164), (65,165), (66,166), (67,167), (68,168), (69,169), (70,170), (71,171), (72,172), (73,173), (74,174), (75,175), (76,176), (77,177), (78,...

val a = sc.parallelize(1 to 100, 3)
val b = sc.parallelize(101 to 200, 3)
val c = sc.parallelize(201 to 300, 3)
a.zip(b).zip(c).map((x) => (x._1._1, x._1._2, x._2 )).collect
res12: Array[(Int, Int, Int)] = Array((1,101,201), (2,102,202), (3,103,203), (4,104,204), (5,105,205), (6,106,206), (7,107,207), (8,108,208), (9,109,209), (10,110,210), (11,111,211), (12,112,212), (13,113,213), (14,114,214), (15,115,215), (16,116,216), (17,117,217), (18,118,218), (19,119,219), (20,120,220), (21,121,221), (22,122,222), (23,123,223), (24,124,224), (25,125,225), (26,126,226), (27,127,227), (28,128,228), (29,129,229), (30,130,230), (31,131,231), (32,132,232), (33,133,233), (34,134,234), (35,135,235), (36,136,236), (37,137,237), (38,138,238), (39,139,239), (40,140,240), (41,141,241), (42,142,242), (43,143,243), (44,144,244), (45,145,245), (46,146,246), (47,147,247), (48,148,248), (49,149,249), (50,150,250), (51,151,251), (52,152,252), (53,153,253), (54,154,254), (55,155,255)...

zipParititions

功能与zip相近，但是可以对zip过程提供更多的控制。

定义

def zipPartitions[B: ClassTag, V: ClassTag](rdd2: RDD[B])(f: (Iterator[T], Iterator[B]) => Iterator[V]): RDD[V]

def zipPartitions[B: ClassTag, V: ClassTag](rdd2: RDD[B], preservesPartitioning: Boolean)(f: (Iterator[T], Iterator[B]) => Iterator[V]): RDD[V]

def zipPartitions[B: ClassTag, C: ClassTag, V: ClassTag](rdd2: RDD[B], rdd3: RDD[C])(f: (Iterator[T], Iterator[B], Iterator[C]) => Iterator[V]): RDD[V]

def zipPartitions[B: ClassTag, C: ClassTag, V: ClassTag](rdd2: RDD[B], rdd3: RDD[C], preservesPartitioning: Boolean)(f: (Iterator[T], Iterator[B], Iterator[C]) => Iterator[V]): RDD[V]

def zipPartitions[B: ClassTag, C: ClassTag, D: ClassTag, V: ClassTag](rdd2: RDD[B], rdd3: RDD[C], rdd4: RDD[D])(f: (Iterator[T], Iterator[B], Iterator[C], Iterator[D]) => Iterator[V]): RDD[V]

def zipPartitions[B: ClassTag, C: ClassTag, D: ClassTag, V: ClassTag](rdd2: RDD[B], rdd3: RDD[C], rdd4: RDD[D], preservesPartitioning: Boolean)(f: (Iterator[T], Iterator[B], Iterator[C], Iterator[D]) => Iterator[V]): RDD[V]

Example

val a = sc.parallelize(0 to 9, 3)
val b = sc.parallelize(10 to 19, 3)
val c = sc.parallelize(100 to 109, 3)
def myfunc(aiter: Iterator[Int], biter: Iterator[Int], citer: Iterator[Int]): Iterator[String] =
{
  var res = List[String]()
  while (aiter.hasNext && biter.hasNext && citer.hasNext)
  {
    val x = aiter.next + " " + biter.next + " " + citer.next
    res ::= x
  }
  res.iterator
}
a.zipPartitions(b, c)(myfunc).collect
res50: Array[String] = Array(2 12 102, 1 11 101, 0 10 100, 5 15 105, 4 14 104, 3 13 103, 9 19 109, 8 18 108, 7 17 107, 6 16 106)

zipWithIndex

将输入数据标号，标号从0开始。如果RDD有多个数据分片，则会启动一个Spark任务来处理。

定义

def zipWithIndex(): RDD[(T, Long)]

Example

val z = sc.parallelize(Array("A", "B", "C", "D"))
val r = z.zipWithIndex
res110: Array[(String, Long)] = Array((A,0), (B,1), (C,2), (D,3))

val z = sc.parallelize(100 to 120, 5)
val r = z.zipWithIndex
r.collect
res11: Array[(Int, Long)] = Array((100,0), (101,1), (102,2), (103,3), (104,4), (105,5), (106,6), (107,7), (108,8), (109,9), (110,10), (111,11), (112,12), (113,13), (114,14), (115,15), (116,16), (117,17), (118,18), (119,19), (120,20))

zipWithUniqueId

与zipWithIndex不同的是这个方法会给每个数据单元一个独立的id，但是和数据单元的实际顺序的index无关。即使RDD存在多个数据分片，这个方法也不会启动spark任务去处理。

定义

def zipWithUniqueId(): RDD[(T, Long)]

val z = sc.parallelize(100 to 120, 5)
val r = z.zipWithUniqueId
r.collect

res12: Array[(Int, Long)] = Array((100,0), (101,5), (102,10), (103,15), (104,1), (105,6), (106,11), (107,16), (108,2), (109,7), (110,12), (111,17), (112,3), (113,8), (114,13), (115,18), (116,4), (117,9), (118,14), (119,19), (120,24))