来自Dataset的RDD导致Spark 2.x出现序列化错误

Question

我有一个RDD，我是使用Databricks笔记本从数据集创建的。

当我尝试从中获取具体值时，它只是失败并出现序列化错误消息。

这是我获取数据的地方（PageCount是Case类）：

val pcDf = spark.sql("SELECT * FROM pagecounts20160801")
val pcDs = pcDf.as[PageCount]
val pcRdd = pcDs.rdd

然后，当我这样做：

pcRdd.take(10)

我得到以下异常：

org.apache.spark.SparkException: Job aborted due to stage failure: Task 0.0 in stage 82.0 (TID 2474) had a not serializable result: org.apache.spark.sql.catalyst.expressions.GeneratedClass$SpecificUnsafeProjection

即使对数据集的同样尝试也起作用：

pcDs.take(10)

编辑：

这是完整的堆栈跟踪

Serialization stack:
    - object not serializable (class: org.apache.spark.sql.catalyst.expressions.GeneratedClass$SpecificUnsafeProjection, value: <function1>)
    - field (class: org.apache.spark.sql.execution.datasources.FileFormat$$anon$1, name: appendPartitionColumns, type: class org.apache.spark.sql.catalyst.expressions.UnsafeProjection)
    - object (class org.apache.spark.sql.execution.datasources.FileFormat$$anon$1, <function1>)
    - field (class: org.apache.spark.sql.execution.datasources.FileScanRDD, name: readFunction, type: interface scala.Function1)
    - object (class org.apache.spark.sql.execution.datasources.FileScanRDD, FileScanRDD[1095] at )
    - field (class: org.apache.spark.NarrowDependency, name: _rdd, type: class org.apache.spark.rdd.RDD)
    - object (class org.apache.spark.OneToOneDependency, org.apache.spark.OneToOneDependency@502bfe49)
    - writeObject data (class: scala.collection.immutable.List$SerializationProxy)
    - object (class scala.collection.immutable.List$SerializationProxy, scala.collection.immutable.List$SerializationProxy@51dc790)
    - writeReplace data (class: scala.collection.immutable.List$SerializationProxy)
    - object (class scala.collection.immutable.$colon$colon, List(org.apache.spark.OneToOneDependency@502bfe49))
    - field (class: org.apache.spark.rdd.RDD, name: org$apache$spark$rdd$RDD$$dependencies_, type: interface scala.collection.Seq)
    - object (class org.apache.spark.rdd.MapPartitionsRDD, MapPartitionsRDD[1096] at )
    - field (class: org.apache.spark.NarrowDependency, name: _rdd, type: class org.apache.spark.rdd.RDD)
    - object (class org.apache.spark.OneToOneDependency, org.apache.spark.OneToOneDependency@52ce8951)
    - writeObject data (class: scala.collection.immutable.List$SerializationProxy)
    - object (class scala.collection.immutable.List$SerializationProxy, scala.collection.immutable.List$SerializationProxy@57850f0)
    - writeReplace data (class: scala.collection.immutable.List$SerializationProxy)
    - object (class scala.collection.immutable.$colon$colon, List(org.apache.spark.OneToOneDependency@52ce8951))
    - field (class: org.apache.spark.rdd.RDD, name: org$apache$spark$rdd$RDD$$dependencies_, type: interface scala.collection.Seq)
    - object (class org.apache.spark.rdd.MapPartitionsRDD, MapPartitionsRDD[1097] at )
    - field (class: org.apache.spark.NarrowDependency, name: _rdd, type: class org.apache.spark.rdd.RDD)
    - object (class org.apache.spark.OneToOneDependency, org.apache.spark.OneToOneDependency@7e99329a)
    - writeObject data (class: scala.collection.immutable.List$SerializationProxy)
    - object (class scala.collection.immutable.List$SerializationProxy, scala.collection.immutable.List$SerializationProxy@792f3145)
    - writeReplace data (class: scala.collection.immutable.List$SerializationProxy)
    - object (class scala.collection.immutable.$colon$colon, List(org.apache.spark.OneToOneDependency@7e99329a))
    - field (class: org.apache.spark.rdd.RDD, name: org$apache$spark$rdd$RDD$$dependencies_, type: interface scala.collection.Seq)
    - object (class org.apache.spark.rdd.MapPartitionsRDD, MapPartitionsRDD[1098] at )
    - field (class: org.apache.spark.sql.Dataset, name: rdd, type: class org.apache.spark.rdd.RDD)
    - object (class org.apache.spark.sql.Dataset, Invalid tree; null:
null)
    - field (class: lineb9de310f01c84f49b76c6c6295a1393c121.$read$$iw$$iw$$iw$$iw, name: pcDs, type: class org.apache.spark.sql.Dataset)
    - object (class lineb9de310f01c84f49b76c6c6295a1393c121.$read$$iw$$iw$$iw$$iw, lineb9de310f01c84f49b76c6c6295a1393c121.$read$$iw$$iw$$iw$$iw@3482035d)
    - field (class: lineb9de310f01c84f49b76c6c6295a1393c121.$read$$iw$$iw$$iw$$iw$PageCount, name: $outer, type: class lineb9de310f01c84f49b76c6c6295a1393c121.$read$$iw$$iw$$iw$$iw)
    - object (class lineb9de310f01c84f49b76c6c6295a1393c121.$read$$iw$$iw$$iw$$iw$PageCount, PageCount(de.b,Spezial:Linkliste/Datei:Playing_card_diamond_9.svg,1,6053))
    - element of array (index: 0)
    - array (class [Llineb9de310f01c84f49b76c6c6295a1393c121.$read$$iw$$iw$$iw$$iw$PageCount;, size 10)
    at org.apache.spark.scheduler.DAGScheduler.org$apache$spark$scheduler$DAGScheduler$$failJobAndIndependentStages(DAGScheduler.scala:1452)
    at org.apache.spark.scheduler.DAGScheduler$$anonfun$abortStage$1.apply(DAGScheduler.scala:1440)
    at org.apache.spark.scheduler.DAGScheduler$$anonfun$abortStage$1.apply(DAGScheduler.scala:1439)
    at scala.collection.mutable.ResizableArray$class.foreach(ResizableArray.scala:59)
    at scala.collection.mutable.ArrayBuffer.foreach(ArrayBuffer.scala:48)
    at org.apache.spark.scheduler.DAGScheduler.abortStage(DAGScheduler.scala:1439)
    at org.apache.spark.scheduler.DAGScheduler$$anonfun$handleTaskSetFailed$1.apply(DAGScheduler.scala:811)
    at org.apache.spark.scheduler.DAGScheduler$$anonfun$handleTaskSetFailed$1.apply(DAGScheduler.scala:811)
    at scala.Option.foreach(Option.scala:257)
    at org.apache.spark.scheduler.DAGScheduler.handleTaskSetFailed(DAGScheduler.scala:811)
    at org.apache.spark.scheduler.DAGSchedulerEventProcessLoop.doOnReceive(DAGScheduler.scala:1665)
    at org.apache.spark.scheduler.DAGSchedulerEventProcessLoop.onReceive(DAGScheduler.scala:1620)
    at org.apache.spark.scheduler.DAGSchedulerEventProcessLoop.onReceive(DAGScheduler.scala:1609)
    at org.apache.spark.util.EventLoop$$anon$1.run(EventLoop.scala:48)
    at org.apache.spark.scheduler.DAGScheduler.runJob(DAGScheduler.scala:632)
    at org.apache.spark.SparkContext.runJob(SparkContext.scala:1868)
    at org.apache.spark.SparkContext.runJob(SparkContext.scala:1881)
    at org.apache.spark.SparkContext.runJob(SparkContext.scala:1894)
    at org.apache.spark.rdd.RDD$$anonfun$take$1.apply(RDD.scala:1311)
    at org.apache.spark.rdd.RDDOperationScope$.withScope(RDDOperationScope.scala:151)
    at org.apache.spark.rdd.RDDOperationScope$.withScope(RDDOperationScope.scala:112)
    at org.apache.spark.rdd.RDD.withScope(RDD.scala:358)
    at org.apache.spark.rdd.RDD.take(RDD.scala:1285)
    at lineb9de310f01c84f49b76c6c6295a1393c137.$read$$iw$$iw$$iw$$iw.<init>(<console>:33)
    at lineb9de310f01c84f49b76c6c6295a1393c137.$read$$iw$$iw$$iw.<init>(<console>:40)
    at lineb9de310f01c84f49b76c6c6295a1393c137.$read$$iw$$iw.<init>(<console>:42)
    at lineb9de310f01c84f49b76c6c6295a1393c137.$read$$iw.<init>(<console>:44)
    at lineb9de310f01c84f49b76c6c6295a1393c137.$eval$.$print$lzycompute(<console>:7)
    at lineb9de310f01c84f49b76c6c6295a1393c137.$eval$.$print(<console>:6)

Answer 1

1. PageCount类肯定具有非可序列化的引用（一些非瞬态非可序列化成员，或者可能是具有相同问题的父类型）。 无法序列化给定的对象让Spark尝试序列化封闭范围，直到越来越多的成员，包括在某个地方的FileFormat成员， - Janino生成的投影（设计不可序列化）。 这只是坏目标对象（PageCount）序列化的副作用。

来自spark FileFormat.scala的相关代码（如果“appendPartitionColumns”一旦实现，应该用@transient标记以确保避免序列化）

// Using lazy val to avoid serialization
  private lazy val appendPartitionColumns =
    GenerateUnsafeProjection.generate(fullSchema, fullSchema)

上述“非预期”序列化在常规方案中永远不会发生，直到用户定义类型序列化成功为止。

2. Spark RDD （原始类型！Spark的已知全局模式）序列化涉及实现期间的完整对象序列化（对象数据和对象“模式”，类型）。 序列化的默认机制是Java序列化程序（因此您可以尝试使用Java序列化程序序列化PageCount ，这可能会揭示此类型的问题），并且可能会被更高效的Kryo序列化程序替换（它会将对象序列化为blob，所以我们将松开架构，并且无法应用需要列访问的sqls）。 这就是RDD访问触发序列化问题的原因

3. Dataframes / Datasets是强类型的，它们绑定到Spark已知的模式。 因此，对于Spark，不需要在节点之间传递对象结构，只传递数据。
   这就是为什么实现底层对象类型PageCount的数据集/数据帧没有问题的原因。