spark mapPartitionWithindex && repartition && coalesce

mapPartitionWithindex  transformation算子,每次输入是一个分区的数据,并且传入数据的分区号

spark.sparkContext.setLogLevel("error")
val kzc=spark.sparkContext.parallelize(List(("hive",8),("apache",8),("hive",30),("hadoop",18)),2)
val bd=spark.sparkContext.parallelize(List(("hive",8),("test",2),("spark",20)),1)
val result=bd.union(kzc)
def fun(x:Int,y:Iterator[(String,Int)]):Iterator[(Int,String,Int)]={
val l=new scala.collection.mutable.ListBuffer[(Int,String,Int)]()
while (y.hasNext){
var tmpy=y.next()
l.append((x,tmpy._1,tmpy._2))
}
l.iterator
}
val result2=result.mapPartitionsWithIndex(fun)
result2.collect().foreach(println(_))

 repartition     transformation算子,从新定义分区,(多个分区分到一个分区不会产生shuffle)

spark.sparkContext.setLogLevel("error")
    val kzc=spark.sparkContext.parallelize(List(("hive",8),("apache",8),("hive",30),("hadoop",18)),2)
    val bd=spark.sparkContext.parallelize(List(("hive",8),("test",2),("spark",20)),1)
    val result=bd.union(kzc).repartition(4)
    def fun(x:Int,y:Iterator[(String,Int)]):Iterator[(Int,String,Int)]={
      val l=new scala.collection.mutable.ListBuffer[(Int,String,Int)]()
      while (y.hasNext){
        var tmpy=y.next()
        l.append((x,tmpy._1,tmpy._2))
      }
      l.iterator
    }
    val result2=result.mapPartitionsWithIndex(fun)
    result2.collect().foreach(println(_))

coalesce常用来减少分区,第二个参数是减少分区的过程中是否产生shuffle。true为产生shuffle,false不产生shuffle。默认是false。如果coalesce设置的分区数比原来的RDD的分区数还多的话,第二个参数设置为false不会起作用,如果设置成true,效果和repartition一样。

spark.sparkContext.setLogLevel("error")
    val kzc=spark.sparkContext.parallelize(List(("hive",8),("apache",8),("hive",30),("hadoop",18)),2)
    val bd=spark.sparkContext.parallelize(List(("hive",8),("test",2),("spark",20)),1)
    val result=bd.union(kzc).coalesce(2,true)
    def fun(x:Int,y:Iterator[(String,Int)]):Iterator[(Int,String,Int)]={
      val l=new scala.collection.mutable.ListBuffer[(Int,String,Int)]()
      while (y.hasNext){
        var tmpy=y.next()
        l.append((x,tmpy._1,tmpy._2))
      }
      l.iterator
    }
    val result2=result.mapPartitionsWithIndex(fun)
    result2.collect().foreach(println(_))

原文地址:https://www.cnblogs.com/students/p/14262279.html