項目場景：

使用sparkStream接收kafka的數(shù)據(jù)進行計算，并且打包上傳到linux進行spark任務(wù)的submit

---

錯誤集合：

1.錯誤1:

Failed to add file:/usr/local/spark-yarn/./myapp/sparkDemo04.jar to Spark environment java.io.FileNotFoundException: Jar D:\usr\local\spark-yarn\myapp\sparkDemo04.jar not found WARN ProcfsMetricsGetter: Exception when trying to compute pagesize, as a result reporting of ProcessTree metrics is stopped

2.windows下ideal中在yarn模式下運行代碼出錯，顯示如下報錯

WARN CheckpointReader: Error reading checkpoint from file hdfs://hadoop102:9000/checkpoint6/checkpoint-1637834226000 java.io.IOException: java.lang.ClassCastException: cannot assign instance of java.lang.invoke.SerializedLambda to field org.apache.spark.streaming.dstream.MappedDStream.mapFunc of type scala.Function1 in instance of org.apache.spark.streaming.dstream.MappedDStream

3.報的一些kafka包notfound的問題，這個下面就不討論了，只需要把對應(yīng)的包下載后放到spark目錄下的jars文件中即可，比如常見的

java.lang.NoClassDefFoundError: org/apache/spark/kafka010/KafkaConfigUpdater

都可以通過添加包的方式解決，如果是spark shell里面出現(xiàn)這種錯誤，則需要在輸入spark-shell命令時，在后面添加 --jars 包路徑
最初的代碼：

import com.study.stream05_kafka.SparkKafka.createSSC import org.apache.kafka.clients.consumer.{ConsumerConfig, ConsumerRecord} import org.apache.log4j.{Level, Logger} import org.apache.spark.SparkConf import org.apache.spark.rdd.RDD import org.apache.spark.streaming.dstream.{DStream, InputDStream} import org.apache.spark.streaming.kafka010.{ConsumerStrategies, KafkaUtils, LocationStrategies} import org.apache.spark.streaming.{Seconds, StreamingContext}import java.lang.System.getProperty import scala.collection.mutable.ListBufferobject stream05_kafka {object SparkKafka{def createSSC(): _root_.org.apache.spark.streaming.StreamingContext={// TODO 創(chuàng)建環(huán)境對象// StreamingContext創(chuàng)建時，第一個參數(shù)表示環(huán)境配置，第二個是數(shù)據(jù)采集周期val sparkConf = new SparkConf().setMaster("local[*]").setAppName("kafka2")sparkConf.set("spark.streaming.stopGracefullyOnShutdown","true")sparkConf.set("spark.hadoop.fs.defaultFS","hdfs://hadoop102:9000")sparkConf.set("spark.hadoop.yarn.resoursemanager.address","hadoop103:8088")val streamingContext: StreamingContext = new StreamingContext(sparkConf, Seconds(3))streamingContext.checkpoint("hdfs://hadoop102:9000/checkpoint6")val kafkaPara: Map[String, Object] = Map[String, Object](ConsumerConfig.BOOTSTRAP_SERVERS_CONFIG -> "hadoop102:9092,hadoop103:9092,hadoop104:9092",ConsumerConfig.GROUP_ID_CONFIG -> "second","key.deserializer" -> "org.apache.kafka.common.serialization.StringDeserializer","value.deserializer" -> "org.apache.kafka.common.serialization.StringDeserializer")// TODO 邏輯處理val kafkaDS: InputDStream[ConsumerRecord[String, String]] = KafkaUtils.createDirectStream[String, String](streamingContext,LocationStrategies.PreferConsistent,ConsumerStrategies.Subscribe[String, String](Set("sparkOnKafka"), kafkaPara))val num: DStream[String] = kafkaDS.map(_.value())val result = num.map(line=>{val flows = line.split(",")val up=flows(1).toIntval down=flows(2).toInt(flows(0),(up,down,up+down))}).updateStateByKey((queueValue, buffValue: Option[(Int,Int,Int)]) => {val cur=buffValue.getOrElse((0,0,0))var curUp=cur._1var curDown=cur._2for (elem <- queueValue) {curUp+=elem._1curDown+=elem._2}Option((curUp,curDown,curUp+curDown))})result.print()streamingContext}}def main(args: Array[String]): Unit = {println("**************")Logger.getLogger("org.apache.spark").setLevel(Level.WARN)System.getProperties.setProperty("HADOOP_USER_NAME", "hadoop")val streamingContext = StreamingContext.getActiveOrCreate("hdfs://hadoop102:9000/checkpoint6", ()=>createSSC())streamingContext.start()// 2.等待關(guān)閉streamingContext.awaitTermination()}}

---

原因分析：

首先，這里指出如果要打包到linux 下在yarn模式下進行spark的submit，需要設(shè)置master為yarn，至于是yarn-client還是yarn-cluster需要提交任務(wù)時指定，默認是client。我這里寫成local，所以一開始都是windows下可以正常連接kafka拿到數(shù)據(jù)進行計算，但是linux下就不行了。歸根結(jié)底沒有連接yarn。
1.錯誤1是因為windows下spark任務(wù)提交的時候，找不到你的jar包，試想一下spark的spark-submit命令，需要指定jar包以及class
2.這個是序列化問題還是廣播變量不適合于檢查點的問題，查資料發(fā)現(xiàn)廣播變量的內(nèi)容寫入hdfs后就難以恢復(fù)了，這里可以把錯誤定位到StreamingContext.getActiveOrCreate里面，這里有時候可以正常進行數(shù)據(jù)恢復(fù)，但是有時候就會報錯。解決方法還沒找到，我就直接換檢查點路徑了，一般生產(chǎn)環(huán)境下也只有代碼升級的情況下會關(guān)閉流計算，這里就沒有深究，希望大神可以解答一下。猜測是讀取檢查點數(shù)據(jù)的時候序列化出了問題

解決方案：

錯誤1的解決：所以如果要在windows下運行，需要先使用mvn package或者build artifacts對程序進行打包，然后對sparkConf.setJars指定包的路徑，這樣在windows下就可以正常運行了
錯誤2的解決：這里我就換檢查點了
最后貼一下我最終成功運行的代碼

import com.study.stream05_kafka.SparkKafka.createSSC import org.apache.kafka.clients.consumer.{ConsumerConfig, ConsumerRecord} import org.apache.log4j.{Level, Logger} import org.apache.spark.SparkConf import org.apache.spark.rdd.RDD import org.apache.spark.streaming.dstream.{DStream, InputDStream} import org.apache.spark.streaming.kafka010.{ConsumerStrategies, KafkaUtils, LocationStrategies} import org.apache.spark.streaming.{Seconds, StreamingContext}import java.lang.System.getProperty import scala.collection.mutable.ListBufferobject stream05_kafka {object SparkKafka{def createSSC(): _root_.org.apache.spark.streaming.StreamingContext={// TODO 創(chuàng)建環(huán)境對象// StreamingContext創(chuàng)建時，第一個參數(shù)表示環(huán)境配置，第二個是數(shù)據(jù)采集周期val sparkConf = new SparkConf().setMaster("yarn").setAppName("kafka2").set("spark.serializer", "org.apache.spark.serializer.KryoSerializer")sparkConf.set("spark.streaming.stopGracefullyOnShutdown","true")sparkConf.set("spark.hadoop.fs.defaultFS","hdfs://hadoop102:9000")sparkConf.set("spark.hadoop.yarn.resoursemanager.address","hadoop103:8088")val streamingContext: StreamingContext = new StreamingContext(sparkConf, Seconds(3))streamingContext.checkpoint("hdfs://hadoop102:9000/checkpoint7")val kafkaPara: Map[String, Object] = Map[String, Object](ConsumerConfig.BOOTSTRAP_SERVERS_CONFIG -> "hadoop102:9092,hadoop103:9092,hadoop104:9092",ConsumerConfig.GROUP_ID_CONFIG -> "second","key.deserializer" -> "org.apache.kafka.common.serialization.StringDeserializer","value.deserializer" -> "org.apache.kafka.common.serialization.StringDeserializer")// TODO 邏輯處理val kafkaDS: InputDStream[ConsumerRecord[String, String]] = KafkaUtils.createDirectStream[String, String](streamingContext,LocationStrategies.PreferConsistent,ConsumerStrategies.Subscribe[String, String](Set("sparkOnKafka"), kafkaPara))val num: DStream[String] = kafkaDS.map(_.value())val result = num.map(line=>{val flows = line.split(",")val up=flows(1).toIntval down=flows(2).toInt(flows(0),(up,down,up+down))}).updateStateByKey((queueValue, buffValue: Option[(Int,Int,Int)]) => {val cur=buffValue.getOrElse((0,0,0))var curUp=cur._1var curDown=cur._2for (elem <- queueValue) {curUp+=elem._1curDown+=elem._2}Option((curUp,curDown,curUp+curDown))})result.print()streamingContext}}def main(args: Array[String]): Unit = {println("**************")Logger.getLogger("org.apache.spark").setLevel(Level.WARN)System.getProperties.setProperty("HADOOP_USER_NAME", "hadoop")val streamingContext = StreamingContext.getActiveOrCreate("hdfs://hadoop102:9000/checkpoint7", ()=>createSSC()) // new Thread(new MonitorStop(streamingContext)).start()streamingContext.start()// 2.等待關(guān)閉streamingContext.awaitTermination()}}

另外，打包的時候不要添加setJars，否則還是會報錯，報的是什么已經(jīng)忘了，這篇博客也是在我解決問題之后寫的，沒有記錄太多報錯，如果我沒記錯的話可能會報這種錯誤

cannot assign instance of java.lang.invoke.SerializedLambda to field org.apache.spark.rdd.MapPartitionsRDD.f of type scala.Function3 in instance of org.apache.spark.rdd.MapPartitionsRDD

困惑：

為了解決這個bug，也是在yarn日志和spark日志來回看，看了一天，最讓我頭疼的就是spark-submit使用control+z退出后，spark-submit進行還會在后臺運行，我都懷疑是不是我的kill -9 操作使檢查點損壞導(dǎo)致數(shù)據(jù)恢復(fù)失敗的，請問各路大神怎么才能結(jié)束sparkSubmit進程？

總結(jié)

以上是生活随笔為你收集整理的spark on yarn模式下SparkStream整合kafka踩的各种坑(已解决)_fqzzzzz的博客的全部內(nèi)容，希望文章能夠幫你解決所遇到的問題。

如果覺得生活随笔網(wǎng)站內(nèi)容還不錯，歡迎將生活随笔推薦給好友。