spark解决hash碰撞

问题背景：pairRDD

调用spark存入文件的api时，最后结果文件的个数（就是分区的个数）取决于PariRDD中的key的hash值,

解决后可是相同key的数据到同一个partitioner中

解决方法：

1. 自己重新定义一个partitioner

//一般内部通过map来区分最好
class HostPartitioner(hosts:Array[String]) extends Partitioner{
  //从host数组中进行partitioner编号的确定

  val hostMap = new mutable.HashMap[String,Int]()
  var count = 0
  for (host <- hosts){
    hostMap +=  (host -> count)
    count += 1
  }

  override def numPartitions: Int = hosts.length

  //key 是前面的host 有三个spark会自动将key传进来判断
  override def getPartition(key: Any): Int = {
    hostMap.getOrElse(key.toString,0)
  }
}

整个代码如下：

package flowanalysis

import java.net.URL

import org.apache.spark.{Partitioner, SparkConf, SparkContext}

import scala.collection.mutable

/**
  * Created by willian on 2017/3/18.
  * 解决存放进hadoop中的产生的hash碰撞，导致文件的内容没有按key进分区
  */
object FlowAnalysisPartitioner {
  def main(args: Array[String]): Unit = {
    val conf = new SparkConf().setAppName("flow_analysis").setMaster("local")
    val sc = new SparkContext(conf)
    val rdd = sc.textFile("/Users/willian/Desktop/project/spark/wordcount/src/main/resources/itcast.log").map(line =>{
      val f = line.split("	")
      (f(i = 1),1)
    })
    val rdd1 = rdd.reduceByKey(_+_)
    val rdd3 = rdd1.map(tuple =>{
      val url = tuple._1
      val host = new URL(url).getHost
      (host,(url,tuple._2))
    })
    val hostrdd = rdd3.map(_._1).distinct().collect()
//    rdd3.repartition(3).saveAsTextFile("/Users/willian/Desktop/project/spark/wordcount/src/main/output")
//    println(hostrdd.collect.toBuffer)
    val partitioner = new HostPartitioner(hostrdd)

    rdd3.partitionBy(partitioner).mapPartitions(it=>{
      it.toList.sortBy(_._2._2).reverse.take(3).iterator
    }).saveAsTextFile("/Users/willian/Desktop/project/spark/wordcount/src/main/output")

  }
}

class HostPartitioner(hosts:Array[String]) extends Partitioner{
  //从host数组中进行partitioner编号的确定

  val hostMap = new mutable.HashMap[String,Int]()
  var count = 0
  for (host <- hosts){
    hostMap +=  (host -> count)
    count += 1
  }

  override def numPartitions: Int = hosts.length

  //key 是前面的host 有三个spark会自动将key传进来判断
  override def getPartition(key: Any): Int = {
    hostMap.getOrElse(key.toString,0)
  }
}