spark 常用技巧总结2

zip拉链操作

def zip[U](other: org.apache.spark.rdd.RDD[U])(implicit evidence$10: scala.reflect.ClassTag[U]): org.apache.spark.rdd.RDD[(String, U)]

scala> val rdd1=sc.makeRDD(Array("apple","pear","grape","egg","elephant"))

rdd1: org.apache.spark.rdd.RDD[String] = ParallelCollectionRDD[23] at makeRDD at <console>:24

scala> val rdd2=sc.makeRDD(List(20,5,8,6,3))

rdd2: org.apache.spark.rdd.RDD[Int] = ParallelCollectionRDD[24] at makeRDD at <console>:24

scala> rdd1.zip(rdd2).collect
res35: Array[(String, Int)] = Array((apple,20), (pear,5), (grape,8), (egg,6), (elephant,3))

scala> val rdd3=rdd1 zip rdd2

rdd3: org.apache.spark.rdd.RDD[(String, Int)] = ZippedPartitionsRDD2[27] at zip at <console>:28

scala> rdd3.collect

res36: Array[(String, Int)] = Array((apple,20), (pear,5), (grape,8), (egg,6), (elephant,3))

-------------------------

def combineByKey[C](createCombiner: Int => C,mergeValue: (C, Int) => C,mergeCombiners: (C, C) => C): org.apache.spark.rdd.RDD[(String, C)]
def combineByKey[C](createCombiner: Int => C,mergeValue: (C, Int) => C,mergeCombiners: (C, C) => C,numPartitions: Int): org.apache.spark.rdd.RDD[(String, C)]
def combineByKey[C](createCombiner: Int => C,mergeValue: (C, Int) => C,mergeCombiners: (C, C) => C,partitioner: org.apache.spark.Partitioner,mapSideCombine: Boolean,serializer: org.apache.spark.serializer.Serializer): org.apache.spark.rdd.RDD[(String, C)]

def combineByKey[C](

createCombiner: V => C,

mergeValue: (C, V) => C,

mergeCombiners: (C, C) => C,       n

umPartitions: Int): RDD[(K, C)] = self.withScope {

combineByKeyWithClassTag(createCombiner, mergeValue, mergeCombiners, numPartitions)(null)

}

scala> rdd3.collect
res53: Array[(String, Int)] = Array((apple,2), (pear,1), (grape,2), (egg,1), (elephant,1))

scala> val rdd4=rdd3.combineByKey(List(_),(x:List[Int],v:Int)=>x:+v,(m:List[Int],n:List[Int])=>m++n)
rdd4: org.apache.spark.rdd.RDD[(String, List[Int])] = ShuffledRDD[35] at combineByKey at <console>:30

scala> rdd4.collect

res51: Array[(String, List[Int])] = Array((egg,List(1)), (elephant,List(1)), (pear,List(1)), (apple,List(2)), (grape,List(2)))

scala> val rdd4=rdd3.map(x=>(x._2,x._1))

rdd4: org.apache.spark.rdd.RDD[(Int, String)] = MapPartitionsRDD[33] at map at <console>:30

scala> val rdd5=rdd4.combineByKey(List(_),(x:List[String],v:String)=>x:+v,(m:List[String],n:List[String])=>m++n)
rdd5: org.apache.spark.rdd.RDD[(Int, List[String])] = ShuffledRDD[37] at combineByKey at <console>:32

scala> rdd5.collect

res52: Array[(Int, List[String])] = Array((1,List(pear, egg, elephant)), (2,List(apple, grape)))

--------------------

scala> val rdd1=sc.makeRDD(Array("apple","apple","pear","egg","hellokitty","egg","apple"))

rdd1: org.apache.spark.rdd.RDD[String] = ParallelCollectionRDD[4] at makeRDD at <console>:24

scala> rdd1.countByValue

res1: scala.collection.Map[String,Long] = Map(hellokitty -> 1, egg -> 2, pear -> 1, apple -> 3)

scala> val map1=rdd1.countByValue
map1: scala.collection.Map[String,Long] = Map(hellokitty -> 1, egg -> 2, pear -> 1, apple -> 3)

scala> val rdd2=sc.makeRDD(map1.toList)

rdd2: org.apache.spark.rdd.RDD[(String, Long)] = ParallelCollectionRDD[21] at makeRDD at <console>:28

scala> rdd2.collect

res5: Array[(String, Long)] = Array((hellokitty,1), (egg,2), (pear,1), (apple,3))

-------------------

scala> val rdd1=sc.makeRDD(Array("apple","apple","pear","egg","hellokitty","egg","apple"))

rdd1: org.apache.spark.rdd.RDD[String] = ParallelCollectionRDD[28] at makeRDD at <console>:24

scala> val rdd2=rdd1.map(x=>(x,1))

rdd2: org.apache.spark.rdd.RDD[(String, Int)] = MapPartitionsRDD[29] at map at <console>:26

scala> rdd2.collect

res33: Array[(String, Int)] = Array((apple,1), (apple,1), (pear,1), (egg,1), (hellokitty,1), (egg,1), (apple,1))

scala> rdd2.partitions.size
res34: Int = 4

scala> rdd2.reduceByKey(_+_).collect
res36: Array[(String, Int)] = Array((hellokitty,1), (egg,2), (pear,1), (apple,3))

scala> rdd2.reduceByKey(_+_,2).partitions.size //shuffile重新分为2个分区
res37: Int = 2

-------------------------------

shuffle操作可以重新分区，指定分区数

进行 shuffle 操作的是是很消耗系统资源的,需要写入到磁盘并通过网络传输,有时还需要对数据进行排序.常见的 Transformation 操作如:repartition，join，cogroup，以及任何 *By 或者 *ByKey 的 Transformation 都需要 shuffle

--------------------------------------

scala> val rdd2=rdd1.map(x=>(x,1))
rdd2: org.apache.spark.rdd.RDD[(String, Int)] = MapPartitionsRDD[29] at map at <console>:26

scala> rdd2.collect
res39: Array[(String, Int)] = Array((apple,1), (apple,1), (pear,1), (egg,1), (hellokitty,1), (egg,1), (apple,1))

scala> rdd2.combineByKey(x=>x,(c:Int,n:Int)=>c+n,(c1:Int,c2:Int)=>c1+c2).collect
res41: Array[(String, Int)] = Array((hellokitty,1), (egg,2), (pear,1), (apple,3))

scala> rdd1.countByValue()
res42: scala.collection.Map[String,Long] = Map(hellokitty -> 1, egg -> 2, pear -> 1, apple -> 3)

scala> rdd2.reduceByKey(_+_).collect
res44: Array[(String, Int)] = Array((hellokitty,1), (egg,2), (pear,1), (apple,3))

-------------------------------

scala> val rdd3=rdd1.map(x=>(1,x))

rdd3: org.apache.spark.rdd.RDD[(Int, String)] = MapPartitionsRDD[40] at map at <console>:26

scala> rdd3.collect

res45: Array[(Int, String)] = Array((1,apple), (1,apple), (1,pear), (1,egg), (1,hellokitty), (1,egg), (1,apple))

scala> rdd3.combineByKey(x=>List(x),(c:List[String],y:String)=>c:+y,(c1:List[String],c2:List[String])=>c1++c2).collect
res49: Array[(Int, List[String])] = Array((1,List(apple, apple, pear, egg, hellokitty, egg, apple)))

---------------------------------------------

scala> val rdd00=sc.makeRDD(List(("a",1),("b",1),("a",3),("ba",3),("b",1),("g",10)),2)

rdd00: org.apache.spark.rdd.RDD[(String, Int)] = ParallelCollectionRDD[44] at makeRDD at <console>:24

scala> val rdd3=rdd00.map(x=>(x._2,x._1))

rdd3: org.apache.spark.rdd.RDD[(Int, String)] = MapPartitionsRDD[45] at map at <console>:26

scala> rdd3.collect

res51: Array[(Int, String)] = Array((1,a), (1,b), (3,a), (3,ba), (1,b), (10,g))

scala> rdd3.groupByKey().collect

res53: Array[(Int, Iterable[String])] = Array((10,CompactBuffer(g)), (1,CompactBuffer(a, b, b)), (3,CompactBuffer(a, ba)))

scala> rdd3.combineByKey(x=>List(x),(c:List[String],y:String)=>c:+y,(c1:List[String],c2:List[String])=>c1++c2).collect

res54: Array[(Int, List[String])] = Array((10,List(g)), (1,List(a, b, b)), (3,List(a, ba)))

-----------------------

distinct(numPartitions:Int) 去重的同时重新分区

scala> val bb=sc.makeRDD(Array(1,1,2,1,8,6,8,4,5,4),2)

bb: org.apache.spark.rdd.RDD[Int] = ParallelCollectionRDD[81] at makeRDD at <console>:25

scala> bb.distinct(1).partitions.size

res61: Int = 1

scala> bb.distinct(3).partitions.size

res62: Int = 3

----------------------

def randomSplit(weights: Array[Double],seed: Long): Array[org.apache.spark.rdd.RDD[Int]]

randomSplit操作根据weights权重将一个RDD分割为多个RDD。权重越高，划分到的几率越大，权重的总和加起来为1，否则会不正常

scala> val split=aa.randomSplit(Array(0.1,0.2,0.3,0.4))

split: Array[org.apache.spark.rdd.RDD[Int]] = Array(MapPartitionsRDD[165] at randomSplit at <console>:27, MapPartitionsRDD[166] at randomSplit at <console>:27, MapPartitionsRDD[167] at randomSplit at <console>:27, MapPartitionsRDD[168] at randomSplit at <console>:27)

scala> split(0).count

res94: Long = 11

scala> split(1).count

res95: Long = 19

scala> split(2).count

res96: Long = 34

scala> split(3).count

res97: Long = 36

-----------------------------------------------------

def glom(): org.apache.spark.rdd.RDD[Array[Int]]

glom将每个分区中的元素放到一个数组里，变成和分区数一样多的数据

scala> val bb=sc.makeRDD(1 to 10,3)

bb: org.apache.spark.rdd.RDD[Int] = ParallelCollectionRDD[203] at makeRDD at <console>:25

scala> bb.glom().collect

res127: Array[Array[Int]] = Array(Array(1, 2, 3), Array(4, 5, 6), Array(7, 8, 9, 10))