mapPartitions操作与 map类似，只不过映射的参数由RDD中的每一个元素变成了RDD中每一个分区的迭代器，如果映射过程需要频繁创建额外的对象，使用mapPartitions操作要比map操作效率高效许多。比如将RDD中的所有数据通过JDBC链接写入数据库，如果使用map函数，可能要为每个元素创建一个connection,开销很大。如果使用mapPartitions,那么只需要针对一个分区建立connection.
Scala中的yield的主要作用是记住每次迭代中的有关值，并逐一存入到一个数组中。
for {子句} yield {变量或表达式}
scala> val numrdd=sc.makeRDD(1 to 10,3)
numrdd: org.apache.spark.rdd.RDD[Int] = ParallelCollectionRDD[51] at makeRDD at <console>:25
scala> def sumn(iter:Iterator[Int])={val aa=for(i<-iter) yield i*2;aa.toIterator}
sumn: (iter: Iterator[Int])Iterator[Int]
scala> numrdd.mapPartitions(sumn).collect
res49: Array[Int] = Array(2, 4, 6, 8, 10, 12, 14, 16, 18, 20)
-----------------------------------------------
分区中的数值求和
scala> val numRDD=sc.makeRDD(1 to 10,3)
numRDD: org.apache.spark.rdd.RDD[Int] = ParallelCollectionRDD[210] at makeRDD at <console>:25
scala> numRDD.mapPartitions(x=>{val result=List(); var i=0;while(x.hasNext){i+=x.next()};result.::(i).toIterator}).collect
res136: Array[Int] = Array(6, 15, 34)
scala> numRDD.mapPartitions(x=>{
val result=List();
var i=0;
while(x.hasNext)
{
i+=x.next()
};
result.::(i).toIterator
}
).collect
res136: Array[Int] = Array(6, 15, 34)
-------------------------------------------------------------
scala> val numRDD=sc.makeRDD(1 to 10,3)
numRDD: org.apache.spark.rdd.RDD[Int] = ParallelCollectionRDD[4] at makeRDD at <console>:24
scala> def partionsum(iter:Iterator[Int])={var result=List[Int]();var i:Int= 0;while(iter.hasNext){var n:Int=iter.next; i += n;} ;result.::(i).toIterator}
partionsum: (iter: Iterator[Int])Iterator[Int]
scala> def partionsum(iter:Iterator[Int])={
var result=List[Int]();
var i:Int= 0;
while(iter.hasNext){
var n:Int=iter.next;
i += n;
} ;
result.::(i).toIterator
}
partionsum: (iter: Iterator[Int])Iterator[Int]
scala> numRDD.mapPartitions(partionsum).collect
res7: Array[Int] = Array(6, 15, 34)
--------------------------------------
分区内的数值进行求和，并展示分区号
scala> val numRDD=sc.makeRDD(1 to 10,3)
numRDD: org.apache.spark.rdd.RDD[Int] = ParallelCollectionRDD[4] at makeRDD at <console>:24
scala> numRDD.mapPartitionsWithIndex((x,iter)=>{val result=List(); var i=0;while(iter.hasNext){i+=iter.next()};result.::(x+"|"+i).toIterator}).collect
res138: Array[String] = Array(0|6, 1|15, 2|34)
scala> numRDD.mapPartitionsWithIndex((x,iter)=>{
val result=List();
var i=0;
while(iter.hasNext){
i+=iter.next()
};
result.::(x+"|"+i).toIterator
}).collect
res138: Array[String] = Array(0|6, 1|15, 2|34)
------------------------------
scala> val numRDD=sc.makeRDD(1 to 10,3)
numRDD: org.apache.spark.rdd.RDD[Int] = ParallelCollectionRDD[4] at makeRDD at <console>:24
scala> def partionwithindexsum(x:Int,iter:Iterator[Int])={var result=List[Int]();var i:Int= 0;while(iter.hasNext){var n:Int=iter.next; i += n;} ;result.::(x+"|"+i).toIterator} partionwithindexsum: (x: Int, iter: Iterator[Int])Iterator[Any]
scala> def partionwithindexsum(x:Int,iter:Iterator[Int])={
var result=List[Int]();
var i:Int= 0;
while(iter.hasNext){
var n:Int=iter.next;
i += n;
} ;
result.::(x+"|"+i).toIterator
}
partionwithindexsum: (x: Int, iter: Iterator[Int])Iterator[Any]
scala> numRDD.mapPartitionsWithIndex(partionwithindexsum).collect
res9: Array[Any] = Array(0|6, 1|15, 2|34)
----------------------
统计每个分区的元素数
scala> val numRDD=sc.makeRDD(1 to 10,3)
numRDD: org.apache.spark.rdd.RDD[Int] = ParallelCollectionRDD[4] at makeRDD at <console>:24
scala> def partionwithindexlength(x:Int,iter:Iterator[Int])={var result=List[Int]();var i:Int= iter.toList.length;result.::(x+"|"+i).toIterator}
partionwithindexlength: (x: Int, iter: Iterator[Int])Iterator[Any]
scala> def partionwithindexlength(x:Int,iter:Iterator[Int])={
var result=List[Int]();
var i:Int= iter.toList.length;
result.::(x+"|"+i).toIterator
}
partionwithindexlength: (x: Int, iter: Iterator[Int])Iterator[Any]
scala> numRDD.mapPartitionsWithIndex(partionwithindexlength).collect
res10: Array[Any] = Array(0|3, 1|3, 2|4)