spark 累加器

累加器原理图：

 累加器创建：

sc.longaccumulator("")                sc.longaccumulator

sc.collectionaccumulator()            sc.collectionaccumulator

sc.doubleaccumulator()                sc.doubleaccumulator

累加器累加：

l.add(1L)

累加器结果获取：

l.value

demo

long累加器

spark.sparkContext.setLogLevel("error")

val data=spark.sparkContext.parallelize(List(" "," "," "," "))
//var l=spark.sparkContext.longAccumulator
var l=spark.sparkContext.longAccumulator("test")
data.map(x=>{
  l.add(3L)
  x
}).count()
//count 函数仅仅用于触发执行
println(l.value)

data循环4次，每次加3，输出结果为12

Double累加器

collection累加器

重复累计问题：

val data=spark.sparkContext.parallelize(List(" "," "," "," "))
    //var l=spark.sparkContext.longAccumulator
    var l=spark.sparkContext.longAccumulator("test")
    val res=data.map(x=>{
      l.add(3L)
      x
    })
      res.count()
    //count 函数仅仅用于触发执行
    println(l.value)
    res.collect()
    println(l.value)

 连续两次调用了action算子，所以这里累加器进行了两次重复的累加，也就是说，累加器实在遇到action算子的时候才进行累加操作的

正确写法在累加器结束后加入cache

 spark.sparkContext.setLogLevel("error")

    val data=spark.sparkContext.parallelize(List(" "," "," "," "))
    //var l=spark.sparkContext.longAccumulator
    var l=spark.sparkContext.longAccumulator("test")
    val res=data.map(x=>{
      l.add(3L)
      x
    }).cache()
      res.count()
    //count 函数仅仅用于触发执行
    println(l.value)
    res.collect()
    println(l.value)