Spark的这些事<三>——spark常用的Transformations 和Actions

Transformations

map，filter
spark最长用的两个Transformations：map，filter，下面就来介绍一下这两个。

先看下面这张图：

从上图中可以清洗的看到 map和filter都是做的什么工作，那我们就代码演示一下。

    val input = sc.parallelize(List(1,2,3,4))

    val result1 = input.map(x=>x*x)
    val result2 = input.filter(x=>x!=1)

    print(result1.collect().mkString(","))
    print("\n")
    print(result2.collect().mkString(","))
    print("\n")

执行结果如下：

16/08/17 18:48:31 INFO DAGScheduler: ResultStage 0 (collect at Map.scala:17) finished in 0.093 s
16/08/17 18:48:31 INFO DAGScheduler: Job 0 finished: collect at Map.scala:17, took 0.268871 s
1,4,9,16
........
16/08/17 18:48:31 INFO DAGScheduler: ResultStage 1 (collect at Map.scala:19) finished in 0.000 s
16/08/17 18:48:31 INFO DAGScheduler: Job 1 finished: collect at Map.scala:19, took 0.018291 s
2,3,4

再回头看下上面那张图，是不是明白什么意思了！

flatMap
另外一个常用的就是flatMap，输入一串字符，分割出每个字符

来用代码实践一下：

    val lines = sc.parallelize(List("hello world","hi"))
    val words = lines.flatMap (lines=>lines.split(" "))
    print(words.first())
    print("\n")

执行结果：

16/08/17 19:23:24 INFO DAGScheduler: Job 2 finished: first at Map.scala:24, took 0.016987 s
hello
16/08/17 19:23:24 INFO SparkContext: Invoking stop() from shutdown hook

分隔符如果改一下的话：

val words = lines.flatMap (lines=>lines.split(","))

结果会怎样呢？

16/08/17 19:33:14 INFO TaskSchedulerImpl: Removed TaskSet 2.0, whose tasks have all completed, from pool 
hello world
16/08/17 19:33:14 INFO SparkContext: Invoking stop() from shutdown hook

和想象的一样吧~

distinct，distinct，intersection，subtract
还有几个比较常用的：distinct，distinct，intersection，subtract

来看看代码实践：

val rdd1 = sc.parallelize(List("coffee","coffee","panda","monkey","tea"))
    val rdd2 = sc.parallelize(List("coffee","monkey","kitty"))

    rdd1.distinct().take(100).foreach(println)

结果：

16/08/17 19:52:29 INFO DAGScheduler: ResultStage 4 (take at Map.scala:30) finished in 0.047 s
16/08/17 19:52:29 INFO TaskSchedulerImpl: Removed TaskSet 4.0, whose tasks have all completed, from pool 
16/08/17 19:52:29 INFO DAGScheduler: Job 3 finished: take at Map.scala:30, took 0.152405 s
monkey
coffee
panda
tea
16/08/17 19:52:29 INFO SparkContext: Starting job: take at Map.scala:32

代码：

 rdd1.union(rdd2).take(100).foreach(println)

结果：

6/08/17 19:52:29 INFO DAGScheduler: Job 5 finished: take at Map.scala:32, took 0.011825 s
coffee
coffee
panda
monkey
tea
coffee
monkey
kitty
16/08/17 19:52:30 INFO SparkContext: Starting job: take at Map.scala:34
16/08/17 19:52:30 INFO DAGScheduler: Registering RDD 11 (intersection at Map.scala:34)
16/08/17 19:52:30 INFO DAGScheduler: Registering RDD 12 (intersection at Map.scala:34)

代码：

rdd1.intersection(rdd2).take(100).foreach(println)

结果：

16/08/17 19:52:30 INFO TaskSetManager: Finished task 0.0 in stage 9.0 (TID 9) in 31 ms on localhost (1/1)
16/08/17 19:52:30 INFO TaskSchedulerImpl: Removed TaskSet 9.0, whose tasks have all completed, from pool 
16/08/17 19:52:30 INFO DAGScheduler: ResultStage 9 (take at Map.scala:34) finished in 0.031 s
16/08/17 19:52:30 INFO DAGScheduler: Job 6 finished: take at Map.scala:34, took 0.060785 s
monkey
coffee
16/08/17 19:52:30 INFO SparkContext: Starting job: take at Map.scala:36

代码：

rdd1.subtract(rdd2).take(100).foreach(println)

结果：

16/08/17 19:52:30 INFO DAGScheduler: Job 6 finished: take at Map.scala:34, took 0.060785 s
monkey
coffee
16/08/17 19:52:30 INFO SparkContext: Starting job: take at Map.scala:36

再看看上面的图，很容易理解吧

Actions

常用的Transformations就介绍到这里，下面介绍下常用的Action:
reduce,countByValue,takeOrdered,takeSample,aggregate

首先看一下：reduce

    val rdd5 = sc.parallelize(List(1,2,3,4))
    print("reduce action:"+rdd5.reduce((x,y)=>x+y)+"\n")

16/08/18 11:51:16 INFO DAGScheduler: Job 15 finished: reduce at Function.scala:55, took 0.012698 s
reduce action:10
16/08/18 11:51:16 INFO SparkContext: Starting job: aggregate at Function.scala:57

countByValue

print(rdd1.countByValue() + "\n")

16/08/18 11:51:16 INFO DAGScheduler: Job 11 finished: countByValue at Function.scala:48, took 0.031726 s
Map(monkey -> 1, coffee -> 2, panda -> 1, tea -> 1)
16/08/18 11:51:16 INFO SparkContext: Starting job: takeOrdered at Function.scala:50

takeOrdered

rdd1.takeOrdered(10).take(100).foreach(println)

16/08/18 11:51:16 INFO DAGScheduler: Job 12 finished: takeOrdered at Function.scala:50, took 0.026160 s
coffee
coffee
monkey
panda
tea
16/08/18 11:51:16 INFO SparkContext: Starting job: takeSample at Function.scala:52

aggregate
这个要重点介绍一下：

Spark文档中aggregate函数定义如下
def aggregate[U](zeroValue: U)(seqOp: (U, T) ⇒ U, combOp: (U, U) ⇒ U)(implicit arg0: ClassTag[U]): U
Aggregate the elements of each partition, and then the results for all the partitions, using given combine functions and a neutral “zero value”. This function can return a different result type, U, than the type of this RDD, T. Thus, we need one operation for merging a T into an U and one operation for merging two U’s, as in scala.TraversableOnce. Both of these functions are allowed to modify and return their first argument instead of creating a new U to avoid memory allocation.

seqOp操作会聚合各分区中的元素，然后combOp操作把所有分区的聚合结果再次聚合，两个操作的初始值都是zeroValue. seqOp的操作是遍历分区中的所有元素(T)，第一个T跟zeroValue做操作，结果再作为与第二个T做操作的zeroValue，直到遍历完整个分区。combOp操作是把各分区聚合的结果，再聚合。aggregate函数返回一个跟RDD不同类型的值。因此，需要一个操作seqOp来把分区中的元素T合并成一个U，另外一个操作combOp把所有U聚合。

val rdd5 = sc.parallelize(List(1,2,3,4))
val rdd6 = rdd5.aggregate((0, 0))  ((x, y) =>(x._1 + y, x._2+1),  (x, y) =>(x._1 + y._1, x._2 + y._2))
    print ("aggregate action : " + rdd6 + "\n"  )

我们看一下结果：

16/08/18 11:51:16 INFO DAGScheduler: Job 16 finished: aggregate at Function.scala:57, took 0.011686 s
aggregate action : (10,4)
16/08/18 11:51:16 INFO SparkContext: Invoking stop() from shutdown hook

我们可以根据以上执行的例子来理解aggregate 用法：

• 第一步：将rdd5中的元素与初始值遍历进行聚合操作
  
  • 第二步：将初始值加1进行遍历聚合
  • 第三步：将结果进行聚合
  • 根据本次的RDD 背部实现如下：
  • 第一步：其实是0+1
  • 1+2
  • 3+3
  • 6+4
  • 然后执行：0+1
  • 1+1
  • 2+1
  • 3+1
  • 此时返回(10,4)
  • 本次执行是一个节点，如果在集群中的话，多个节点，会先把数据打到不同的分区上，比如(1,2) (3,4)
  • 得到的结果就会是(3,2) (7,2)
  • 然后进行第二步combine就得到 (10,4)

这样你应该能理解aggregate这个函数了吧

以上就是对常用的Transformations 和Actions介绍，对于初学者来说，动手代码实践各个函数，才是明白其功能最好的方法。

PS ：源码

Spark的这些事系列文章：
Spark的这些事<一>——Windows下spark开发环境搭建
Spark的这些事<二>——几个概念
Spark的这些事<三>——spark常用的Transformations 和Actions

<script type="text/javascript"> $(function () { $('pre.prettyprint code').each(function () { var lines = $(this).text().split('\n').length; var $numbering = $('<ul/>').addClass('pre-numbering').hide(); $(this).addClass('has-numbering').parent().append($numbering); for (i = 1; i <= lines; i++) { $numbering.append($('<li/>').text(i)); }; $numbering.fadeIn(1700); }); }); </script>