SparkStreaming：updateStateByKey

updateStateByKey（func）

从名字上来看，该函数会更新根据key聚合，并不断更新value值

要想使用该函数，Dstream之前的类型必须是K，V形式的二元组。
经过执行func函数后，会返回一个key的所有的聚合值得状态。

以word count为例，对于每一个批的数据进行分解聚合，会得到当前的这个批的状态,经过聚合后得到值的，假设有(word1,10),(word2,15),(word3,1);
而对于之前的累积值，(word1,100),(word2,10),(word3，15),(word4,10)
则下一步的状态值为(word1,100+10),(word2,10+15),(word3,1+15),(word4,0+10)
等待下一个批次的数据到了，继续按照这个逻辑进行处理。

代码示例：

package com.lgh.sparkstreaming
 
import kafka.serializer.StringDecoder
import org.apache.spark.SparkConf
import org.apache.spark.streaming.{Seconds, StreamingContext}
import org.apache.spark.streaming.dstream.DStream
import org.apache.spark.streaming.kafka.KafkaUtils
 
/**
  * Created by lgh on 2017/8/24.
  */
object UpdateStateByKey {
  def main(args: Array[String]): Unit = {
    val brokers = "mtime-bigdata00:9092,mtime-bigdata01:9092";
    val topics = "testkafka";
    val batchseconds = "10";
    val checkpointDirectory = "./upbykey";
    val ssc = StreamingContext.getOrCreate(checkpointDirectory, () => createcontext(brokers, topics, batchseconds, checkpointDirectory))
    ssc.start()
    ssc.awaitTermination()
  }
 
  def createcontext(brokers: String, topics: String, batchseconds: String, checkpointDirectory: String): StreamingContext = {
    val sparkconf = new SparkConf().setAppName("TestUpStateByKey").setMaster("local[3]")
    val ssc = new StreamingContext(sparkconf, Seconds(batchseconds.toInt))
    val topicsSet = topics.split(",").toSet
    val kafkaParams = Map[String, String]("metadata.broker.list" -> brokers);
 
    val messages = KafkaUtils.createDirectStream[String, String, StringDecoder, StringDecoder](ssc, kafkaParams, topicsSet)
 
    val lines: DStream[String] = messages.map(_._2)
 
    val message: DStream[(String, Long)] = lines.flatMap(_.split(" ")).map(x => (x, 1L)).reduceByKey(_+_)
 
    val keyvalue=message.updateStateByKey[Long](addFunction _)
    keyvalue.print(100)
 
    ssc.checkpoint(checkpointDirectory)
    ssc
  }
 
  //curValues:对于每一个key的所有value值得集合Seq
  //preVauleState: 之前的每一个key的值对应的value值；这这个例子中是单词计数的之前的累计值
  //
  def addFunction(currValues: Seq[Long], preVauleState: Option[Long]): Option[Long] = {
    //计算得到当前key在这个Dstream中的值
    val currentSum:Long = currValues.sum
    //得到当前key以前的累积值
    val previousSum:Long = preVauleState.getOrElse(0)
    //计算出当前key对应的新的值
    val nowvalue:Long=currentSum + previousSum;
    //返回结果值；在scala中，Some和None是Option的子类
    Some(nowvalue)
  }
 
}

原文参考：https://www.2cto.com/kf/201708/673330.html