spark-streaming-kafka怎樣通過(guò)KafkaUtils.createDirectStream的方式處理數(shù)據(jù)

spark-streaming-kafka怎樣通過(guò)KafkaUtils.createDirectStream的方式處理數(shù)據(jù)，相信很多沒(méi)有經(jīng)驗(yàn)的人對(duì)此束手無(wú)策，為此本文總結(jié)了問(wèn)題出現(xiàn)的原因和解決方法，通過(guò)這篇文章希望你能解決這個(gè)問(wèn)題。

package hgs.spark.streaming
import org.apache.spark.SparkConf
import org.apache.spark.SparkContext
import org.apache.spark.streaming.StreamingContext
import org.apache.spark.streaming.Seconds
import org.apache.spark.streaming.kafka.KafkaUtils
import org.apache.spark.streaming.kafka.KafkaCluster
import scala.collection.immutable.Map
import java.util.NoSuchElementException
import org.apache.spark.SparkException
import kafka.common.TopicAndPartition
import kafka.message.MessageAndMetadata
import org.codehaus.jackson.map.deser.std.PrimitiveArrayDeserializers.StringDeser
import kafka.serializer.StringDecoder
import org.apache.spark.streaming.kafka.DirectKafkaInputDStream
import org.apache.spark.rdd.RDD
import org.apache.spark.streaming.kafka.HasOffsetRanges
import org.apache.spark.HashPartitioner
object SparkStreamingKafkaDirectWordCount {
  def main(args: Array[String]): Unit = {
     val conf = new SparkConf().setAppName("KafkaWordCount").setMaster("local[5]")
     conf.set("spark.streaming.kafka.maxRatePerPartition", "1")
     val sc = new SparkContext(conf)
     val ssc = new StreamingContext(sc,Seconds(1)) 
     ssc.checkpoint("d:\\checkpoint")
     val kafkaParams = Map[String,String](
         "metadata.broker.list"->"bigdata01:9092,bigdata02:9092,bigdata03:9092",
         "group.id"->"group_hgs",
         "zookeeper.connect"->"bigdata01:2181,bigdata02:2181,bigdata03:2181")
     val kc = new KafkaCluster(kafkaParams)
     val topics = Set[String]("test")
     //每個(gè)rdd返回的數(shù)據(jù)是(K,V)類型的，該函數(shù)規(guī)定了函數(shù)返回?cái)?shù)據(jù)的類型
     val mmdFunct = (mmd: MessageAndMetadata[String, String])=>(mmd.topic+" "+mmd.partition,mmd.message())
     
     val rds = KafkaUtils.createDirectStream[String,String,StringDecoder,StringDecoder,(String,String)](ssc, kafkaParams, getOffsets(topics,kc,kafkaParams),mmdFunct)
     val updateFunc=(iter:Iterator[(String,Seq[Int],Option[Int])])=>{
        //iter.flatMap(it=>Some(it._2.sum+it._3.getOrElse(0)).map((it._1,_)))//方式一
        //iter.flatMap{case(x,y,z)=>{Some(y.sum+z.getOrElse(0)).map((x,_))}}//方式二
        iter.flatMap(it=>Some(it._1,(it._2.sum.toInt+it._3.getOrElse(0))))//方式三
      }
     val words = rds.flatMap(x=>x._2.split(" ")).map((_,1))
     //val wordscount = words.map((_,1)).updateStateByKey(updateFunc, new HashPartitioner(sc.defaultMinPartitions), true)
     //println(getOffsets(topics,kc,kafkaParams))
     rds.foreachRDD(rdd=>{
       if(!rdd.isEmpty()){
         //對(duì)每個(gè)dataStreamoffset進(jìn)行更新
         upateOffsets(topics,kc,rdd,kafkaParams)
       }
     }   
    )
     
     words.print()
     ssc.start()
     ssc.awaitTermination()
     
    
  }
  
  
  
  def getOffsets(topics : Set[String],kc:KafkaCluster,kafkaParams:Map[String,String]):Map[TopicAndPartition, Long]={
   
    val topicAndPartitionsOrNull =  kc.getPartitions(topics)
    if(topicAndPartitionsOrNull.isLeft){
      throw new SparkException(s"$topics in the set may not found")
    }
    else{
      val topicAndPartitions = topicAndPartitionsOrNull.right.get
      val groups = kafkaParams.get("group.id").get
      val offsetOrNull = kc.getConsumerOffsets(groups, topicAndPartitions)
      if(offsetOrNull.isLeft){
        println(s"$groups you assignment may not exists!now redirect to zero!")
        //如果沒(méi)有消費(fèi)過(guò)，則從最開(kāi)始的位置消費(fèi)
        val erliestOffset = kc.getEarliestLeaderOffsets(topicAndPartitions)
        if(erliestOffset.isLeft)
          throw new SparkException(s"Topics and Partions not definded not found!")
        else
          erliestOffset.right.get.map(x=>(x._1,x._2.offset))
      }
      else{
        //如果消費(fèi)組已經(jīng)存在則從記錄的地方開(kāi)始消費(fèi)
        offsetOrNull.right.get
      }
    }
    
  }
  
  //每次拉取數(shù)據(jù)后存儲(chǔ)offset到ZK
  def upateOffsets(topics : Set[String],kc:KafkaCluster,directRDD:RDD[(String,String)],kafkaParams:Map[String,String]){
    val offsetRanges =  directRDD.asInstanceOf[HasOffsetRanges].offsetRanges
    for(offr <-offsetRanges){
      val topicAndPartitions = TopicAndPartition(offr.topic,offr.partition)
      val yesOrNo = kc.setConsumerOffsets(kafkaParams.get("group.id").get, Map(topicAndPartitions->offr.untilOffset))
      if(yesOrNo.isLeft){
        println(s"Error when update offset of $topicAndPartitions")
      }
    }
  }
  
  
 
}
/* val conf = new SparkConf().setAppName("KafkaWordCount").setMaster("local[2]")
     val sc = new SparkContext(conf)
     val ssc = new StreamingContext(sc,Seconds(4)) 
    val kafkaParams = Map[String,String](
         "metadata.broker.list"->"bigdata01:9092,bigdata02:9092,bigdata03:9092")
     val kc = new KafkaCluster(kafkaParams)
    
     //獲取topic與paritions的信息
     //val tmp = kc.getPartitions(Set[String]("test7"))
     //結(jié)果：topicAndPartitons=Set([test7,0], [test7,1], [test7,2])
     //val topicAndPartitons = tmp.right.get
     //println(topicAndPartitons)
    
    //每個(gè)分區(qū)對(duì)應(yīng)的leader信息
    //val tmp = kc.getPartitions(Set[String]("test7"))
    //val topicAndPartitons = tmp.right.get
    //結(jié)果：leadersPerPartitions= Right(Map([test7,0] -> (bigdata03,9092), [test7,1] -> (bigdata01,9092), [test7,2] -> (bigdata02,9092)))
    //val leadersPerPartitions = kc.findLeaders(topicAndPartitons)
    //println(leadersPerPartitions)
    
    //每增加一條消息，對(duì)應(yīng)的partition的offset都會(huì)加1,即LeaderOffset(bigdata02,9092,23576)第三個(gè)參數(shù)會(huì)加一
    //val tmp = kc.getPartitions(Set[String]("test"))
    //val topicAndPartitons = tmp.right.get
    //結(jié)果t=	Right(Map([test7,0] -> LeaderOffset(bigdata03,9092,23568), [test7,2] -> LeaderOffset(bigdata02,9092,23576), [test7,1] -> LeaderOffset(bigdata01,9092,23571)))
    //val  t = kc.getLatestLeaderOffsets(topicAndPartitons)
    // println(t)
    
    
    //findLeader需要兩個(gè)參數(shù) topic 分區(qū)編號(hào)
    //val tmp = kc.findLeader("test7",0)
    //結(jié)果leader=RightProjection(Right((bigdata03,9092)))
    //val leader = tmp.right
    //val tp = leader.flatMap(x=>{Either.cond(false, None,(x._1,x._2))})  
    
    val tmp = kc.getPartitions(Set[String]("test"))
    val ttp = tmp.right.get
    
    
    while(true){
      try{
      val tp = kc.getConsumerOffsets("group_test1", ttp)
      val maps = tp.right.get
      println(maps)
      Thread.sleep(2000)
      }
      catch{
        case ex:NoSuchElementException=>{println("test")}
      }
        
    }*/

看完上述內(nèi)容，你們掌握spark-streaming-kafka怎樣通過(guò)KafkaUtils.createDirectStream的方式處理數(shù)據(jù)的方法了嗎？如果還想學(xué)到更多技能或想了解更多相關(guān)內(nèi)容，歡迎關(guān)注億速云行業(yè)資訊頻道，感謝各位的閱讀！