Spark RDD Transformation 简单用例（二）-白红宇

Spark RDD Transformation 简单用例（二）

阅读量：6176 次

发布时间：2019-06-21

本文共 18176 字，大约阅读时间需要 60 分钟。

aggregateByKey(zeroValue)(seqOp, combOp, [numTasks])

aggregateByKey(zeroValue)(seqOp, combOp, [numTasks]) When called on a dataset of (K, V) pairs, returns a dataset of (K, U) pairs where the values for each key are aggregated using the given combine functions and a neutral "zero" value. Allows an aggregated value type that is different than the input value type, while avoiding unnecessary allocations. Like in groupByKey, the number of reduce tasks is configurable through an optional second argument.

/**  * Aggregate the values of each key, using given combine functions and a neutral "zero value".  * This function can return a different result type, U, than the type of the values in this RDD,  * V. Thus, we need one operation for merging a V into a U and one operation for merging two U's,  * as in scala.TraversableOnce. The former operation is used for merging values within a  * partition, and the latter is used for merging values between partitions. To avoid memory  * allocation, both of these functions are allowed to modify and return their first argument  * instead of creating a new U.  */ def aggregateByKey[U: ClassTag](zeroValue: U)(seqOp: (U, V) => U,     combOp: (U, U) => U): RDD[(K, U)]

/**  * Aggregate the values of each key, using given combine functions and a neutral "zero value".  * This function can return a different result type, U, than the type of the values in this RDD,  * V. Thus, we need one operation for merging a V into a U and one operation for merging two U's,  * as in scala.TraversableOnce. The former operation is used for merging values within a  * partition, and the latter is used for merging values between partitions. To avoid memory  * allocation, both of these functions are allowed to modify and return their first argument  * instead of creating a new U.  */ def aggregateByKey[U: ClassTag](zeroValue: U, numPartitions: Int)(seqOp: (U, V) => U,     combOp: (U, U) => U): RDD[(K, U)]

/**  * Aggregate the values of each key, using given combine functions and a neutral "zero value".  * This function can return a different result type, U, than the type of the values in this RDD,  * V. Thus, we need one operation for merging a V into a U and one operation for merging two U's,  * as in scala.TraversableOnce. The former operation is used for merging values within a  * partition, and the latter is used for merging values between partitions. To avoid memory  * allocation, both of these functions are allowed to modify and return their first argument  * instead of creating a new U.  */ def aggregateByKey[U: ClassTag](zeroValue: U, partitioner: Partitioner)(seqOp: (U, V) => U,     combOp: (U, U) => U): RDD[(K, U)]

def seq(a:Int,b:Int):Int={  println("seq: " + a + "\t" + b)  math.max(a,b)}def comb(a:Int,b:Int):Int = {  println("comb: " + a + "\t" + b)  a+b}val rdd = sc.parallelize(List((1,3),(1,2),(1,4),(2,3),(2,4),(2,5)))rdd.aggregateByKey(0)(seq,comb).collectrdd.aggregateByKey(6)(seq,comb).collect

scala> def seq(a:Int,b:Int):Int={     |   println("seq: " + a + "\t" + b)     |   math.max(a,b)     | }seq: (a: Int, b: Int)Intscala> scala> def comb(a:Int,b:Int):Int = {     |   println("comb: " + a + "\t" + b)     |   a+b     | }comb: (a: Int, b: Int)Intscala> val rdd = sc.parallelize(List((1,3),(1,2),(1,4),(2,3),(2,4),(2,5)))rdd: org.apache.spark.rdd.RDD[(Int, Int)] = ParallelCollectionRDD[11] at parallelize at 
     
      :26scala> rdd.aggregateByKey(0)(seq,comb).collectseq: 0    3seq: 3    2seq: 3    4seq: 0    3seq: 3    4seq: 4    5res20: Array[(Int, Int)] = Array((1,4), (2,5))scala> rdd.aggregateByKey(6)(seq,comb).collectseq: 6    3seq: 6    2seq: 6    4seq: 6    3seq: 6    4seq: 6    5res21: Array[(Int, Int)] = Array((1,6), (2,6))

但是为什么没有执行comb呢？

sortByKey([ascending], [numTasks])

sortByKey([ascending], [numTasks]) When called on a dataset of (K, V) pairs where K implements Ordered, returns a dataset of (K, V) pairs sorted by keys in ascending or descending order, as specified in the boolean ascending argument.

从下面的注释中可以看到在每一个partition中元素是有序的，但是在整个rdd中数据可能是无序的。 /**  * Sort the RDD by key, so that each partition contains a sorted range of the elements. Calling  * `collect` or `save` on the resulting RDD will return or output an ordered list of records  * (in the `save` case, they will be written to multiple `part-X` files in the filesystem, in  * order of the keys).  */ // TODO: this currently doesn't work on P other than Tuple2! def sortByKey(ascending: Boolean = true, numPartitions: Int = self.partitions.length)     : RDD[(K, V)]

val rdd = sc.parallelize(List((3,"sd"),(1,"fd"),(2,"dfh"),(4,"kjh"),(7,"kf"),(5,"nb"),(100,"jd"),(63,"mm"),(42,"kk"),(99,"ll"),(10,"ll"),(11,"ll"),(12,"ll")),1)val rdd1 = rdd.sortByKey(true,1)rdd1.collectval rdd2 = rdd.sortByKey(true,3) rdd2.foreachPartition( x=>{   while(x.hasNext){     println(x.next)   }   println("============") } ) val rdd2 = rdd.sortByKey(false,4) val rdd2 = rdd.sortByKey(true,3) rdd2.foreachPartition( x=>{   while(x.hasNext){     println(x.next)   }   println("============") } )

scala> val rdd = sc.parallelize(List((3,"sd"),(1,"fd"),(2,"dfh"),(4,"kjh"),(7,"kf"),(5,"nb"),(100,"jd"),(63,"mm"),(42,"kk"),(99,"ll"),(10,"ll"),(11,"ll"),(12,"ll")),1)rdd: org.apache.spark.rdd.RDD[(Int, String)] = ParallelCollectionRDD[24] at parallelize at 
     
      :26scala> val rdd1 = rdd.sortByKey(true,1)rdd1: org.apache.spark.rdd.RDD[(Int, String)] = ShuffledRDD[25] at sortByKey at 
      
       :28scala> rdd1.collectres42: Array[(Int, String)] = Array((1,fd), (2,dfh), (3,sd), (4,kjh), (5,nb), (7,kf), (10,ll), (11,ll), (12,ll), (42,kk), (63,mm), (99,ll), (100,jd))scala> val rdd2 = rdd.sortByKey(true,3)rdd2: org.apache.spark.rdd.RDD[(Int, String)] = ShuffledRDD[28] at sortByKey at 
       
        :28scala>  rdd2.foreachPartition(     |  x=>{     |    while(x.hasNext){     |      println(x.next)     |    }     |    println("============")     |  }     |  )(1,fd)(2,dfh)(3,sd)(4,kjh)(5,nb)============(7,kf)(10,ll)(11,ll)(12,ll)============(42,kk)(63,mm)(99,ll)(100,jd)============scala> val rdd2 = rdd.sortByKey(false,4)rdd2: org.apache.spark.rdd.RDD[(Int, String)] = ShuffledRDD[34] at sortByKey at 
        
         :28scala>  rdd2.foreachPartition(     |  x=>{     |    while(x.hasNext){     |      println(x.next)     |    }     |    println("============")     |  }     |  )(100,jd)(99,ll)(63,mm)============(42,kk)(12,ll)(11,ll)============(10,ll)(7,kf)(5,nb)============(4,kjh)(3,sd)(2,dfh)(1,fd)============

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

/**  * Return this RDD sorted by the given key function.  */ def sortBy[K](     f: (T) => K,     ascending: Boolean = true,     numPartitions: Int = this.partitions.length)     (implicit ord: Ordering[K], ctag: ClassTag[K]): RDD[T]

val a = Array(9,2,8,1,5,6,4,7,3)val rdd = sc.parallelize(a)rdd.collectrdd.sortBy(x=>x).collectrdd.sortBy(x=>x,false,3).collect

scala> val a = Array(9,2,8,1,5,6,4,7,3)a: Array[Int] = Array(9, 2, 8, 1, 5, 6, 4, 7, 3)scala> val rdd = sc.parallelize(a)rdd: org.apache.spark.rdd.RDD[Int] = ParallelCollectionRDD[35] at parallelize at 
     
      :28scala> rdd.collectres46: Array[Int] = Array(9, 2, 8, 1, 5, 6, 4, 7, 3)scala> rdd.sortBy(x=>x).collectres49: Array[Int] = Array(1, 2, 3, 4, 5, 6, 7, 8, 9)scala> rdd.sortBy(x=>x,false,3).collectres50: Array[Int] = Array(9, 8, 7, 6, 5, 4, 3, 2, 1)

join(otherDataset, [numTasks])

join(otherDataset, [numTasks]) When called on datasets of type (K, V) and (K, W), returns a dataset of (K, (V, W)) pairs with all pairs of elements for each key. Outer joins are supported through leftOuterJoin, rightOuterJoin, and fullOuterJoin.

同SQL语句中join,leftOuterJoin同SQL中left outer join,rightOuterJoin同SQL语句中right outer join,fullOuterJoin同SQL语句中的full outer join

scala> val a = List((1,"a"),(2,"b"),(3,"c"))a: List[(Int, String)] = List((1,a), (2,b), (3,c))scala> val rdd1 = sc.parallelize(a)rdd1: org.apache.spark.rdd.RDD[(Int, String)] = ParallelCollectionRDD[47] at parallelize at 
     
      :28scala> val b = List((1,"A"),(2,"B"),(4,"D"))b: List[(Int, String)] = List((1,A), (2,B), (4,D))scala> val rdd2 = sc.parallelize(b)rdd2: org.apache.spark.rdd.RDD[(Int, String)] = ParallelCollectionRDD[48] at parallelize at 
      
       :28scala> val rdd = rdd1.join(rdd2)rdd: org.apache.spark.rdd.RDD[(Int, (String, String))] = MapPartitionsRDD[51] at join at 
       
        :34scala> rdd.collectres51: Array[(Int, (String, String))] = Array((1,(a,A)), (2,(b,B)))scala> rdd1.leftOuterJoin(rdd2)res52: org.apache.spark.rdd.RDD[(Int, (String, Option[String]))] = MapPartitionsRDD[54] at leftOuterJoin at 
        
         :35scala> rdd1.leftOuterJoin(rdd2).collectres53: Array[(Int, (String, Option[String]))] = Array((1,(a,Some(A))), (3,(c,None)), (2,(b,Some(B))))scala> rdd1.rightOuterJoin(rdd2).collectres54: Array[(Int, (Option[String], String))] = Array((4,(None,D)), (1,(Some(a),A)), (2,(Some(b),B)))scala> rdd1.fullOuterJoin(rdd2).collectres55: Array[(Int, (Option[String], Option[String]))] = Array((4,(None,Some(D))), (1,(Some(a),Some(A))), (3,(Some(c),None)), (2,(Some(b),Some(B))))

不管是join,leftOuterJoin,rightOuterJoin还是fullOuterJoin，除上述入参为otherDataset外，还包含下面两种方式

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

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

cogroup(otherDataset, [numTasks])

cogroup(otherDataset, [numTasks]) When called on datasets of type (K, V) and (K, W), returns a dataset of (K, (Iterable<V>, Iterable<W>)) tuples. This operation is also called groupWith.

/**  * For each key k in `this` or `other`, return a resulting RDD that contains a tuple with the  * list of values for that key in `this` as well as `other`.  */ def cogroup[W](other: RDD[(K, W)]): RDD[(K, (Iterable[V], Iterable[W]))]

scala> val rdd1 = sc.parallelize(List((1,"a"),(2,"b"),(3,"c"),(1,"z")))rdd1: org.apache.spark.rdd.RDD[(Int, String)] = ParallelCollectionRDD[0] at parallelize at 
     
      :24scala> val rdd2 = sc.parallelize(List((1,"A"),(2,"B"),(2,"C"),(4,"D")))rdd2: org.apache.spark.rdd.RDD[(Int, String)] = ParallelCollectionRDD[1] at parallelize at 
      
       :24scala> val rdd = rdd1.cogroup(rdd2)rdd: org.apache.spark.rdd.RDD[(Int, (Iterable[String], Iterable[String]))] = MapPartitionsRDD[3] at cogroup at 
       
        :28scala> rdd.collectres0: Array[(Int, (Iterable[String], Iterable[String]))] = Array((4,(CompactBuffer(),CompactBuffer(D))), (1,(CompactBuffer(a, z),CompactBuffer(A))), (3,(CompactBuffer(c),CompactBuffer())), (2,(CompactBuffer(b),CompactBuffer(B, C))))

cartesian(otherDataset)

cartesian(otherDataset)

When called on datasets of types T and U, returns a dataset of (T, U) pairs (all pairs of elements).

对两个RDD中元素进行笛卡尔积运算。 /**  * Return the Cartesian product of this RDD and another one, that is, the RDD of all pairs of  * elements (a, b) where a is in `this` and b is in `other`.  */ def cartesian[U: ClassTag](other: RDD[U]): RDD[(T, U)]

scala> val rdd1 = sc.parallelize(Array(1,2,3,4,5))rdd1: org.apache.spark.rdd.RDD[Int] = ParallelCollectionRDD[4] at parallelize at 
     
      :24scala> val rdd2 = sc.parallelize(Array("A","B","C"))rdd2: org.apache.spark.rdd.RDD[String] = ParallelCollectionRDD[5] at parallelize at 
      
       :24scala> val rdd = rdd1.cartesian(rdd2)rdd: org.apache.spark.rdd.RDD[(Int, String)] = CartesianRDD[6] at cartesian at 
       
        :28scala> rdd.collectres1: Array[(Int, String)] = Array((1,A), (1,B), (1,C), (2,A), (2,B), (2,C), (3,A), (3,B), (3,C), (4,A), (4,B), (4,C), (5,A), (5,B), (5,C))

pipe(command, [envVars])

pipe(command, [envVars])

Pipe each partition of the RDD through a shell command, e.g. a Perl or bash script. RDD elements are written to the process's stdin and lines output to its stdout are returned as an RDD of strings.

通过pipe运行外部程序，每个分区中的元素作为外部程序入参运行一次外部程序，而外部程序的输出有创建一个新的RDD。 /**  * Return an RDD created by piping elements to a forked external process.  */ def pipe(command: String): RDD[String]

[root@localhost home]# more /home/test.sh #!/bin/bashecho "Running shell script"RESULT=""while read LINEdo   if [ -z ${LINE} ] then    break fi RESULT=${RESULT}" "${LINE}doneecho ${RESULT} >> /home/out.txtecho "========" >>/home/out.txt

val rdd = sc.parallelize(List("ab","cd","ef","gh","ij"),2)rdd.pipe("/home/test.sh").collect

结果：

rdd有两个分区，test.sh每次运行会输出一个“Running shell script”字符串，元素输出至/home/out.txt中。

scala> val rdd = sc.parallelize(List("ab","cd","ef","gh","ij"),2)rdd: org.apache.spark.rdd.RDD[String] = ParallelCollectionRDD[8] at parallelize at 
     
      :24scala> rdd.pipe("/home/test.sh").collectres6: Array[String] = Array(Running shell script, Running shell script)

[root@localhost home]# more out.txt ab cd========ef gh ij========

coalesce(numPartitions)

coalesce(numPartitions)

Decrease the number of partitions in the RDD to numPartitions. Useful for running operations more efficiently after filtering down a large dataset.

减少RDD的partition数量，对过滤掉大量数据后进行算子操作高效运行非常有用。 /**  * Return a new RDD that is reduced into `numPartitions` partitions.  *  * This results in a narrow dependency, e.g. if you go from 1000 partitions  * to 100 partitions, there will not be a shuffle, instead each of the 100  * new partitions will claim 10 of the current partitions.  *  * However, if you're doing a drastic coalesce, e.g. to numPartitions = 1,  * this may result in your computation taking place on fewer nodes than  * you like (e.g. one node in the case of numPartitions = 1). To avoid this,  * you can pass shuffle = true. This will add a shuffle step, but means the  * current upstream partitions will be executed in parallel (per whatever  * the current partitioning is).  *  * Note: With shuffle = true, you can actually coalesce to a larger number  * of partitions. This is useful if you have a small number of partitions,  * say 100, potentially with a few partitions being abnormally large. Calling  * coalesce(1000, shuffle = true) will result in 1000 partitions with the  * data distributed using a hash partitioner.  */ def coalesce(numPartitions: Int, shuffle: Boolean = false,              partitionCoalescer: Option[PartitionCoalescer] = Option.empty)             (implicit ord: Ordering[T] = null)     : RDD[T]

scala> val rdd = sc.parallelize(1 to 1000,1000)rdd: org.apache.spark.rdd.RDD[Int] = ParallelCollectionRDD[10] at parallelize at 
     
      :24scala> val rdd1 = rdd.filter(_%3 == 0)rdd1: org.apache.spark.rdd.RDD[Int] = MapPartitionsRDD[11] at filter at 
      
       :26scala> rdd1.partitions.lengthres7: Int = 1000scala> rdd1.coalesce(3,false).partitions.lengthres9: Int = 3

repartition(numPartitions)

repartition(numPartitions)

Reshuffle the data in the RDD randomly to create either more or fewer partitions and balance it across them. This always shuffles all data over the network.

该函数其实内部调用就是coalesce(numPartitions, shuffle = true)。

/**  * Return a new RDD that has exactly numPartitions partitions.  * Can increase or decrease the level of parallelism in this RDD. Internally, this uses  * a shuffle to redistribute data.  * If you are decreasing the number of partitions in this RDD, consider using `coalesce`,  * which can avoid performing a shuffle.  */ def repartition(numPartitions: Int)(implicit ord: Ordering[T] = null): RDD[T] = withScope {
     coalesce(numPartitions, shuffle = true) }

repartitionAndSortWithinPartitions(partitioner)

repartitionAndSortWithinPartitions(partitioner) Repartition the RDD according to the given partitioner and, within each resulting partition, sort records by their keys. This is more efficient than calling repartition and then sorting within each partition because it can push the sorting down into the shuffle machinery.

/**  * Repartition the RDD according to the given partitioner and, within each resulting partition,  * sort records by their keys.  *  * This is more efficient than calling `repartition` and then sorting within each partition  * because it can push the sorting down into the shuffle machinery.  */ def repartitionAndSortWithinPartitions(partitioner: Partitioner): RDD[(K, V)]

class MyPartitioner(numParts:Int) extends org.apache.spark.Partitioner{  override def numPartitions: Int = numParts  override def getPartition(key: Any): Int = {    key.toString.toInt%numPartitions  }}val rdd1 = sc.makeRDD(1 to 10,2)val rdd2 = sc.makeRDD(1 to 10,2)val rdd = rdd1.zip(rdd2)rdd.foreachPartition( x=>{   while(x.hasNext){     println(x.next)   }   println("============") } ) val rdd3 = rdd.repartitionAndSortWithinPartitions(new MyPartitioner(3))rdd3.foreachPartition( x=>{   while(x.hasNext){     println(x.next)   }   println("============") } )

scala> class MyPartitioner(numParts:Int) extends org.apache.spark.Partitioner{     |   override def numPartitions: Int = numParts     |   override def getPartition(key: Any): Int = {     |     key.toString.toInt%numPartitions     |   }     | }defined class MyPartitionerscala> val rdd1 = sc.makeRDD(1 to 10,2)rdd1: org.apache.spark.rdd.RDD[Int] = ParallelCollectionRDD[37] at makeRDD at 
     
      :24scala> val rdd2 = sc.makeRDD(1 to 10,2)rdd2: org.apache.spark.rdd.RDD[Int] = ParallelCollectionRDD[38] at makeRDD at 
      
       :24scala> val rdd = rdd1.zip(rdd2)rdd: org.apache.spark.rdd.RDD[(Int, Int)] = ZippedPartitionsRDD2[39] at zip at 
       
        :28scala> rdd.foreachPartition(     |  x=>{     |    while(x.hasNext){     |      println(x.next)     |    }     |    println("============")     |  }     |  )(1,1)(2,2)(3,3)(4,4)(5,5)============(6,6)(7,7)(8,8)(9,9)(10,10)============                                                                                scala> val rdd3 = rdd.repartitionAndSortWithinPartitions(new MyPartitioner(3))rdd3: org.apache.spark.rdd.RDD[(Int, Int)] = ShuffledRDD[40] at repartitionAndSortWithinPartitions at 
        
         :31scala> rdd3.foreachPartition(     |  x=>{     |    while(x.hasNext){     |      println(x.next)     |    }     |    println("============")     |  }     |  )[Stage 17:>                                                         (0 + 1) / 3](3,3)(6,6)(9,9)============(1,1)(4,4)(7,7)(10,10)============(2,2)(5,5)(8,8)============