如何修复"MetadataFetchFailedException: Missing an output location for shuffle"？

Question

如果我增加word2vec模型的模型大小，我开始在我的日志中出现这种异常：

org.apache.spark.shuffle.MetadataFetchFailedException: Missing an output location for shuffle 6
    at org.apache.spark.MapOutputTracker$$anonfun$org$apache$spark$MapOutputTracker$$convertMapStatuses$2.apply(MapOutputTracker.scala:542)
    at org.apache.spark.MapOutputTracker$$anonfun$org$apache$spark$MapOutputTracker$$convertMapStatuses$2.apply(MapOutputTracker.scala:538)
    at scala.collection.TraversableLike$WithFilter$$anonfun$foreach$1.apply(TraversableLike.scala:772)
    at scala.collection.IndexedSeqOptimized$class.foreach(IndexedSeqOptimized.scala:33)
    at scala.collection.mutable.ArrayOps$ofRef.foreach(ArrayOps.scala:108)
    at scala.collection.TraversableLike$WithFilter.foreach(TraversableLike.scala:771)
    at org.apache.spark.MapOutputTracker$.org$apache$spark$MapOutputTracker$$convertMapStatuses(MapOutputTracker.scala:538)
    at org.apache.spark.MapOutputTracker.getMapSizesByExecutorId(MapOutputTracker.scala:155)
    at org.apache.spark.shuffle.BlockStoreShuffleReader.read(BlockStoreShuffleReader.scala:47)
    at org.apache.spark.rdd.ShuffledRDD.compute(ShuffledRDD.scala:98)
    at org.apache.spark.rdd.RDD.computeOrReadCheckpoint(RDD.scala:306)
    at org.apache.spark.rdd.RDD.iterator(RDD.scala:270)
    at org.apache.spark.rdd.CoalescedRDD$$anonfun$compute$1.apply(CoalescedRDD.scala:96)
    at org.apache.spark.rdd.CoalescedRDD$$anonfun$compute$1.apply(CoalescedRDD.scala:95)
    at scala.collection.Iterator$$anon$13.hasNext(Iterator.scala:371)
    at scala.collection.Iterator$$anon$11.hasNext(Iterator.scala:327)
    at scala.collection.Iterator$class.foreach(Iterator.scala:727)
    at scala.collection.AbstractIterator.foreach(Iterator.scala:1157)
    at scala.collection.generic.Growable$class.$plus$plus$eq(Growable.scala:48)
    at scala.collection.mutable.ArrayBuffer.$plus$plus$eq(ArrayBuffer.scala:103)
    at scala.collection.mutable.ArrayBuffer.$plus$plus$eq(ArrayBuffer.scala:47)
    at scala.collection.TraversableOnce$class.to(TraversableOnce.scala:273)
    at scala.collection.AbstractIterator.to(Iterator.scala:1157)
    at scala.collection.TraversableOnce$class.toBuffer(TraversableOnce.scala:265)
    at scala.collection.AbstractIterator.toBuffer(Iterator.scala:1157)
    at scala.collection.TraversableOnce$class.toArray(TraversableOnce.scala:252)
    at scala.collection.AbstractIterator.toArray(Iterator.scala:1157)
    at org.apache.spark.rdd.RDD$$anonfun$collect$1$$anonfun$12.apply(RDD.scala:927)
    at org.apache.spark.rdd.RDD$$anonfun$collect$1$$anonfun$12.apply(RDD.scala:927)
    at org.apache.spark.SparkContext$$anonfun$runJob$5.apply(SparkContext.scala:1858)
    at org.apache.spark.SparkContext$$anonfun$runJob$5.apply(SparkContext.scala:1858)
    at org.apache.spark.scheduler.ResultTask.runTask(ResultTask.scala:66)
    at org.apache.spark.scheduler.Task.run(Task.scala:89)
    at org.apache.spark.executor.Executor$TaskRunner.run(Executor.scala:213)
    at java.util.concurrent.ThreadPoolExecutor.runWorker(ThreadPoolExecutor.java:1142)
    at java.util.concurrent.ThreadPoolExecutor$Worker.run(ThreadPoolExecutor.java:617)
    at java.lang.Thread.run(Thread.java:745)

我尝试编写自己的"保存模型"版本，如下所示：

  def save(model: Word2VecModel, sc: SparkContext, path: String): Unit = {
    println("Saving model as CSV ..")
    val vectorSize = model.getVectors.values.head.size
    println("vectorSize="+vectorSize)
    val SEPARATOR_TOKEN = " "    
    val dataArray = model.getVectors.toSeq.map { case (w, v) => Data(w, v) }
    println("Got dataArray ..")
    println("parallelize(dataArray, 10)")
    val par = sc.parallelize(dataArray, 10)
          .map(d => {
            val sb = new mutable.StringBuilder()
            sb.append(d.word)
            sb.append(SEPARATOR_TOKEN)
            for(v <- d.vector) {
              sb.append(v)
              sb.append(SEPARATOR_TOKEN)
            }
            sb.setLength(sb.length - 1)
            sb.append("n")
            sb.toString()
          })
    println("repartition(1)")
    val rep = par.repartition(1)
    println("collect()")
    val vectorsAsString = rep.collect()
    println("Collected serialized vectors ..")    
    val cfile = new mutable.StringBuilder()
    cfile.append(vectorsAsString.length)
    cfile.append(" ")
    cfile.append(vectorSize)
    cfile.append("n")
    val sb = new StringBuilder
    sb.append("word,")
    for(i <- 0 until vectorSize) {
      sb.append("v")
      sb.append(i.toString)
      sb.append(",")
    }
    sb.setLength(sb.length - 1)
    sb.append("n")
    for(vectorString <- vectorsAsString) {
      sb.append(vectorString)
      cfile.append(vectorString)
    }
    println("Saving file to " + new Path(path, "data").toUri.toString)
    sc.parallelize(sb.toString().split("n"), 1).saveAsTextFile(new Path(path+".csv", "data").toUri.toString)
    sc.parallelize(cfile.toString().split("n"), 1).saveAsTextFile(new Path(path+".cs", "data").toUri.toString)
  }

显然，它的工作方式与他们当前的实现类似 - 它没有。

我想得到一个word2vec模型。它适用于小文件，但如果模型变大，则不适用。

Answer 1

当执行程序上的MapOutputTracker找不到本地缓存中分区的请求随机映射输出并尝试从驱动程序的MapOutputTracker远程获取它们时，会抛出MetadataFetchFailedException。

这可能会导致一些结论：

1. 驱动程序的内存问题
2. 执行程序的内存问题
3. 遗嘱执行人丢失

请查看日志，查找报告为"执行程序丢失"信息消息的问题和/或查看 Web UI 的执行程序页面，了解执行程序的工作原理。

执行程序

丢失的根本原因也可能是集群管理器决定杀死行为不良的执行程序（可能占用的内存多于请求的内存）。

有关更多见解，请参阅另一个问题 FetchFailedException 或 MetadataFetchFailedException 处理大数据集时。

Answer 2

简而言之，配置可能会有所帮助：

--conf spark.blacklist.enabled=true # blacklist bad machine
--conf spark.reducer.maxReqsInFlight=10 # limit concurrent requests from reducer 
--conf spark.shuffle.io.retryWait=10s # increase retry wait
--conf spark.shuffle.io.maxRetries=10 # increase retry times
--conf spark.shuffle.io.backLog=4096 # increase tcp connection wait queue length

长解释如下。

对于 MetadataFetchFailedException，它通常发生在一个执行器突然被杀死或终止时，但这个执行器有一些随机输出，然后当另一个执行器尝试获取此随机输出的元数据时，会发生异常。

1. 在大多数情况下，这是由 Yarn 杀死的容器引起的超出内存限制。所以你需要在原木。

2. 最常见的解决方法是增加内存开销，默认值是 0.1 * 执行程序内存。对于大多数情况来说，这太小了。我会建议将其设为0.2 *执行器内存。如果您有大量的执行程序或运行另一个子进程，您需要更大的值这。