spark学习笔记—核心算子(二)

spark学习笔记—核心算子(二)

distinct算子

  /**
   * Return a new RDD containing the distinct elements in this RDD.
   */
  def distinct(numPartitions: Int)(implicit ord: Ordering[T] = null): RDD[T] = withScope {
    def removeDuplicatesInPartition(partition: Iterator[T]): Iterator[T] = {
      // Create an instance of external append only map which ignores values.
      val map = new ExternalAppendOnlyMap[T, Null, Null](
        createCombiner = _ => null,
        mergeValue = (a, b) => a,
        mergeCombiners = (a, b) => a)
      map.insertAll(partition.map(_ -> null))
      map.iterator.map(_._1)
    }
    partitioner match {
      case Some(_) if numPartitions == partitions.length =>
        mapPartitions(removeDuplicatesInPartition, preservesPartitioning = true)
      // 一般会走下面的流程，进行reduceByKey的操作，针对重复的元素只会返回第一个元素
      case _ => map(x => (x, null)).reduceByKey((x, _) => x, numPartitions).map(_._1)
    }
  }

    //使用reduceByKey实现distinct的功能
    val result: RDD[Int] = nums.map((_, null)).reduceByKey((x, y) => x).map(_._1)
    result.foreach(println)

cogroup算子

   * 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]))] = self.withScope {
    cogroup(other, defaultPartitioner(self, other))
  }


  def cogroup[W](other: RDD[(K, W)], partitioner: Partitioner)
      : RDD[(K, (Iterable[V], Iterable[W]))] = self.withScope {
    if (partitioner.isInstanceOf[HashPartitioner] && keyClass.isArray) {
      throw new SparkException("HashPartitioner cannot partition array keys.")
    }
    val cg = new CoGroupedRDD[K](Seq(self, other), partitioner)
    cg.mapValues { case Array(vs, w1s) =>
      (vs.asInstanceOf[Iterable[V]], w1s.asInstanceOf[Iterable[W]])
    }
  }

intersection算子

  /**
   * Return the intersection of this RDD and another one. The output will not contain any duplicate
   * elements, even if the input RDDs did.
   *
   * @note This method performs a shuffle internally.
   */
  def intersection(other: RDD[T]): RDD[T] = withScope {
    this.map(v => (v, null)).cogroup(other.map(v => (v, null)))
        .filter { case (_, (leftGroup, rightGroup)) => leftGroup.nonEmpty && rightGroup.nonEmpty }
        .keys
  }

使用cogroup实现intersect算子

    val rdd1: RDD[Int] = sc.makeRDD(List(1, 2, 3, 4, 5))
    val rdd2: RDD[Int] = sc.makeRDD(List(4, 5, 6, 7, 8))


    val rdd3: RDD[(Int, Null)] = rdd1.map((_, null))
    val rdd4: RDD[(Int, Null)] = rdd2.map((_, null))

    val grouped: RDD[(Int, (Iterable[Null], Iterable[Null]))] = rdd3.cogroup(rdd4)
    val res: RDD[Int] = grouped.filter(
      t => t._2._1.nonEmpty && t._2._2.nonEmpty
    ).keys

    val resultRDD: RDD[Int] = rdd1.intersection(rdd2)

使用cogroup实现join算子

val conf: SparkConf = new SparkConf().setAppName("WordCount").setMaster("local")
val sc = new SparkContext(conf)

val rdd1: RDD[(String, Int)] = sc.makeRDD(List(("tom", 1), ("tom", 2), ("jerry", 3), ("ketty", 2)))
val rdd2: RDD[(String, Int)] = sc.makeRDD(List(("jerry", 1), ("tom", 2), ("shuke", 2)))

val rdd3: RDD[(String, (Iterable[Int], Iterable[Int]))] = rdd1.cogroup(rdd2)

val result: RDD[(String, (Int, Int))] = rdd1.join(rdd2)

//使用cogroup实现join的效果
val rdd4: RDD[(String, (Int, Int))] = rdd3.flatMapValues(t => {
  for (x <- t._1.iterator; y <- t._2.iterator) yield (x, y)
})

val rdd1: RDD[(String, Int)] = sc.makeRDD(List(("tom", 1), ("tom", 2), ("jerry", 3), ("ketty", 2)))
val rdd2: RDD[(String, Int)] = sc.makeRDD(List(("jerry", 1), ("tom", 2), ("shuke", 2)))

val rdd3: RDD[(String, (Iterable[Int], Iterable[Int]))] = rdd1.cogroup(rdd2)

val result: RDD[(String, (Int, Int))] = rdd1.join(rdd2)

//使用cogroup实现join
val rdd4: RDD[(String, (Int, Int))] = rdd3.flatMapValues(t => {
  for (x <- t._1.iterator; y <- t._2.iterator) yield (x, y)
})


val leftJoinRDD: RDD[(String, (Int, Option[Int]))] = rdd1.leftOuterJoin(rdd2)
leftJoinRDD.collect().foreach(println)
//使用cogroup实现leftJoin
val rdd5: RDD[(String, (Int, Option[Int]))] = rdd3.flatMapValues((t: (Iterable[Int], Iterable[Int])) => {
  if (t._2.isEmpty) {
    //这里的t._1有可能会有多个元素或者是empty需要进行map操作
    t._1.map((_, None))
  } else {
    for (x <- t._1.iterator; y <- t._2.iterator) yield (x, Some(y))
  }
})

//使用cogroup实现rightOuterJoin
val value: RDD[(String, (Option[Int], Int))] = rdd1.rightOuterJoin(rdd2)
val value1: RDD[(String, (Option[Int], Int))] = rdd3.flatMapValues(
  t => {
    if (t._1.isEmpty) {
      t._2.map((None, _))
    } else {
      for (x <- t._1.iterator; y <- t._2.iterator) yield (Some(x), y)
    }
  }
)
value.collect().foreach(println)

// 使用cogroup算子实现fullOuterJoin
val fullOuterJoinRDD: RDD[(String, (Option[Int], Option[Int]))] = rdd3.flatMapValues {
  case (i1, Seq()) => i1.iterator.map(x => (Some(x), None))
  case (Seq(), i2) => i2.iterator.map(x => (None, Some(x)))
  case (i1, i2) => for (a <- i1.iterator; b <- i2.iterator) yield (Some(a), Some(b))
}

count算子

/**
 * Run a job on all partitions in an RDD and return the results in an array.
 *
 * @param rdd target RDD to run tasks on
 * @param func a function to run on each partition of the RDD
 * @return in-memory collection with a result of the job (each collection element will contain
 * a result from one partition)
 */
// 这里的array里面有每个分区进行计算得到的结果
def runJob[T, U: ClassTag](rdd: RDD[T], func: Iterator[T] => U): Array[U] = {
  runJob(rdd, func, 0 until rdd.partitions.length)
}

rdd中的cache操作

• 一个application多次触发action，为了服用前面RDD计算好的数据，避免反复读取HDFS(数据源)中的数据或者重复计算
• 缓存，可以将数据缓存到内存或者磁盘(executor所在的磁盘),第一次触发action才放入在内存或者磁盘，以后再触发action会读取缓存的RDD的数据进行操作并且复用缓存的数据
• 一个RDD多次触发action缓存才有意义
• 如果将数据缓存在内存，内存不够，以分区为单位，只缓存部分分区的数据
• 支持多种StorageLevel，可以将数据序列化，默认放入内存使用的是java对象存储，但是占用空间大，优点是速度快，也可以使用其他的序列化的方式
• cache底层调用的是persist方法，可以指定其他的存储级别
• cache和persist方法，严格来说不是Transformation，因为没有生成新的rdd，只是标记当前的rdd需要cache或者persist
• 原始的数据，经过整理过滤后再进行cache或者persist效果会更佳

rdd中的checkpoint操作

• 使用场景:适合复杂的计算(机器学习、迭代计算)为了避免丢失数据重复计算、可以将宝贵的中间结果保存在hdfs中、保证中间结果的安全

• 在调用rdd的checkpoint方法之前，一定要指定checkpoint的目录，即sc.setCheckPointDir

• 为了保证中间结果的安全，将数据保存在HDFS、分布式文件系统中可以保证数据不丢

• 第一次触发action，才做checkpoint，会额外触发一个job，这个job的母的就是将中间结果保存在HDFS中

• 如果rdd做了checkpoint、这个rdd之前的依赖关系就不再使用了

• 触发多次action，checkpoint才有意义、多用于迭代计算

• checkpoint严格的说，不是transformation，只是标记当前rdd要做checkpoint

• 如果checkpoint前，对rdd进行了cache，可以避免数据重复计算，如果有cache的数据优先使用cache，没有再使用checkpoint，如果checkpoint过保存在hdfs中的数据丢了，在对相关的数据进行操作时会报错

统计连续登录的三天及以上的用户

这个问题可以拓展到很多相似的问题：连续几个月充值会员、连续天数有商品卖出、连续打滴滴、连续逾期

测试数据:用户id、登入日期

原始数据

guid01,2018-02-28
guid01,2018-03-01
guid01,2018-03-02
guid01,2018-03-04
guid01,2018-03-05
guid01,2018-03-06
guid01,2018-03-07
guid02,2018-03-01
guid02,2018-03-02
guid02,2018-03-03
guid02,2018-03-06

object UserContinuedLogin {
  def main(args: Array[String]): Unit = {
    val conf: SparkConf = new SparkConf().setAppName(this.getClass.getCanonicalName).setMaster("local[*]")
    val sc = new SparkContext(conf)
    val rdd: RDD[String] = sc.textFile("data/login.log")

    // 转换为(uid,date)类型的格式
    val mapRDD: RDD[(String, String)] = rdd.map(line => {
      val strings: Array[String] = line.split(",")
      (strings(0), strings(1))
    })

    val groupRDD: RDD[(String, Iterable[String])] = mapRDD.groupByKey()

    val flatMapRDD: RDD[(String, (String, String))] = groupRDD.flatMapValues(it => {
      val sorted: List[String] = it.toSet.toList.sortBy((x: String) => x)
      val calendar: Calendar = Calendar.getInstance()
      var sdf = new SimpleDateFormat("yyyy-MM-dd")
      var index = 0
      sorted.map(dateStr => {
        // 减去行号，如果结果相同的话说明连续登录
        val date: Date = sdf.parse(dateStr)
        calendar.setTime(date)
        calendar.add(Calendar.DATE, -index)
        index += 1
        (dateStr, sdf.format(calendar.getTime))
      })
    })
    
    val result: RDD[(String, Int, String, String)] = flatMapRDD.map(t => ((t._1, t._2._2), t._2._1)).groupByKey().mapValues(it => {
      val list = it.toList.sorted
      val times = list.size
      val beginTime = list.head
      val endTime = list.last
      (times, beginTime, endTime)
    }).filter(t => t._2._1 >= 3).map(t => {
      (t._1._1, t._2._1, t._2._2, t._2._3)
    })
    println(result.collect().toBuffer)
  }
}

统计每门学科最受欢迎的老师前三名

原始数据如下

http://bigdata.edu360.cn/laozhang
http://bigdata.edu360.cn/laozhang
http://bigdata.edu360.cn/laozhao
http://bigdata.edu360.cn/laozhao
http://bigdata.edu360.cn/laozhao
http://bigdata.edu360.cn/laozhao
http://bigdata.edu360.cn/laozhao

直接toList进行排序

    val sc: SparkConf = new SparkConf().setMaster("local").setAppName("TopTeacher");
    val context = new SparkContext(sc)
    val rdd: RDD[String] = context.textFile("data/teacher.log")
    val mapRdd: RDD[((String, String), Int)] = rdd.map({
      line: String => {
        val strings: Array[String] = line.split("/")
        val teacher: String = strings(3)
        val course: String = strings(2).split("\\.")(0)
        ((course, teacher), 1)
      }
    })
    val reduceRdd: RDD[((String, String), Int)] = mapRdd.reduceByKey(_ + _)
    val groupRdd: RDD[(String, Iterable[((String, String), Int)])] = reduceRdd.groupBy(_._1._1)
    va l result: RDD[(String, List[((String, String), Int)])] = 								groupRdd.mapValues(_.toList.sortBy(_._2).reverse.take(3))

过滤后求topN

    val sc: SparkConf = new SparkConf().setMaster("local").setAppName("TopTeacher");
    val context = new SparkContext(sc)
    val rdd: RDD[String] = context.textFile("data/teacher.log")
    val mapRdd: RDD[((String, String), Int)] = rdd.map({
      line: String => {
        val strings: Array[String] = line.split("/")
        val teacher: String = strings(3)
        val course: String = strings(2).split("\\.")(0)
        ((course, teacher), 1)
      }
    })
		
    val reduceRdd: RDD[((String, String), Int)] = mapRdd.reduceByKey(_ + _)
       
		val subject = List("bigdata", "javaee", "kafka", "hive")

    for (sb <- subject){
      val filtered: RDD[((String, String), Int)] = reduceRdd.filter((_: ((String, String), Int))._1._1 == sb)
      val favTeacher: Array[((String, String), Int)] = filtered.sortBy((_: ((String, String), Int))._2, ascending = false).take(3)
      //上面的sortBy所用的是全局排序的方式，实际上要取TopN没有必要进行全局排序
      implicit val orderRules:Ordering[((String,String),Int)] = Ordering[Int].on(t => t._2)
      //下面在进行top操作的过程中需要用到隐式参数
      val res = reduceRdd.top(2)
      print(res.toBuffer)
    }

//对于reduceRdd可以进行repartition进行不同分区的重新组合，
// 使不同的学科的数据都在不同的组内，不会产生数据倾斜的情况

// 首先统计不同学科的数量放在一个array中
val subjects: Array[String] = reduceRdd.map(_._1._2).distinct().collect()
val subjectPartitioner = new SubjectPartitioner(subjects)
val partitioned: RDD[((String, String), Int)] = reduceRdd.partitionBy(subjectPartitioner)

//对于自定义分区器重新分区后的数据进行取topN操作
class SubjectPartitioner(val subjects: Array[String]) extends Partitioner{
  val nameToNum = new mutable.HashMap[String,Int]()
  var i = 0
  for (sub <- subjects){
    nameToNum(sub) = i
    i += 1
  }
  override def numPartitions: Int = subjects.length

  //在Executors中的Task中，shuffleWrite之前会被调用
  override def getPartition(key: Any): Int = {
    val tuple: (String, String) = key.asInstanceOf[(String, String)]
    nameToNum(tuple._1)
  }
}
val partitionedRDD: RDD[((String, String), Int)] = partitioned.mapPartitions(it => {
  it.toList.sortBy(-_._2).take(2).iterator
})

//可以对上面操作进行优化，利用有界优先队列
val value2: RDD[((String, String), Int)] = partitioned.mapPartitions(
  it => {
    val value1: Ordering[((String, String), Int)] = Ordering[Int].on[((String, String), Int)](-_._2)
    val sorter = new mutable.TreeSet[((String, String), Int)]()
    it.foreach(
      e => {
        sorter.add(e)
        if (sorter.size > 2) {
          sorter -= sorter.last
        }
      }
    )
    sorter.iterator
  }
)
println(value2.collect().toBuffer)

reduce中传入自定义的分区器减少shuffle的数量

    val sc: SparkConf = new SparkConf().setMaster("local").setAppName("TopTeacher");
    val context = new SparkContext(sc)
    val rdd: RDD[String] = context.textFile("data/teacher.log")
    val mapRdd: RDD[((String, String), Int)] = rdd.map({
      line: String => {
        val strings: Array[String] = line.split("/")
        val teacher: String = strings(3)
        val course: String = strings(2).split("\\.")(0)
        ((course, teacher), 1)
      }
    })
  val subjectPartitioner = new SubjectPartitioner(subjects)
  val reduceRdd: RDD[((String, String), Int)] = mapRdd.reduceByKey(subjectPartitioner, _ + _)
  //可以对上面的value的操作进行优化，利用有界优先队列
  val value2: RDD[((String, String), Int)] = reduceRdd.mapPartitions(
    it => {
      val value1: Ordering[((String, String), Int)] = Ordering[Int].on[((String, String), Int)](-_._2)
      val sorter = new mutable.TreeSet[((String, String), Int)]()
      it.foreach(
        e => {
          sorter.add(e)
          if (sorter.size > 2) {
            sorter -= sorter.last
          }
        }
      )
      sorter.iterator
    }
  )