Scala 使用org.apache.hadoop.conf.Configuration设置Spark记录分隔符时，操作RDD失败

我想用Spark处理一个大文本文件“mydata.txt”（实际文件大小约30GB）。它的记录分隔符是“\\；”后跟“\n”。由于加载文件（按“sc.textFile”）的默认记录分隔符为“\n”，因此我将org.apache.hadoop.conf.conf.Configuration的“textinputformat.record.delimiter”属性设置为“\\ 124n”以指定记录分隔符：

AAAAA_|BBBBB_|
CCCCC\
DDDDD
EEEEE_FFFFFFFFFFFF\ |
GGGGG_|HHHHH_|
IIIII\
GGGGG\
KKKKK_|LLLLLLLLLLL\ |
MMMM_|NNNNN_|OOOOO\ |

接下来，我在spark shell中执行了以下代码：

import org.apache.hadoop.io.LongWritable
import org.apache.hadoop.io.Text
import org.apache.hadoop.conf.Configuration
import org.apache.hadoop.mapreduce.lib.input.TextInputFormat

val LINE_DELIMITER = "\\ |\n"
val FIELD_SEP = "_\\|"

val conf = new Configuration
conf.set("textinputformat.record.delimiter", LINE_DELIMITER)
val raw_data = sc.newAPIHadoopFile("mydata.txt", classOf[TextInputFormat], classOf[LongWritable], classOf[Text], conf).map(_._2.toString)

到目前为止还不错。但是,

scala> val data = raw_data.filter(x => x.split(FIELD_SEP).size >= 3)
data: org.apache.spark.rdd.RDD[String] = FilteredRDD[4] at filter at <console>:22

scala> data.collect
org.apache.spark.SparkException: Job aborted due to stage failure: Task not serializable: java.io.NotSerializableException: org.apache.hadoop.conf.Configuration
    at org.apache.spark.scheduler.DAGScheduler.org$apache$spark$scheduler$DAGScheduler$$failJobAndIndependentStages(DAGScheduler.scala:1049)
    at org.apache.spark.scheduler.DAGScheduler$$anonfun$abortStage$1.apply(DAGScheduler.scala:1033)
    at org.apache.spark.scheduler.DAGScheduler$$anonfun$abortStage$1.apply(DAGScheduler.scala:1031)
    at scala.collection.mutable.ResizableArray$class.foreach(ResizableArray.scala:59)
    at scala.collection.mutable.ArrayBuffer.foreach(ArrayBuffer.scala:47)
    at org.apache.spark.scheduler.DAGScheduler.abortStage(DAGScheduler.scala:1031)
    at org.apache.spark.scheduler.DAGScheduler.org$apache$spark$scheduler$DAGScheduler$$submitMissingTasks(DAGScheduler.scala:772)
    at org.apache.spark.scheduler.DAGScheduler.org$apache$spark$scheduler$DAGScheduler$$submitStage(DAGScheduler.scala:715)
    at org.apache.spark.scheduler.DAGScheduler.handleJobSubmitted(DAGScheduler.scala:699)
    at org.apache.spark.scheduler.DAGSchedulerEventProcessActor$$anonfun$receive$2.applyOrElse(DAGScheduler.scala:1203)
    at akka.actor.ActorCell.receiveMessage(ActorCell.scala:498)
    at akka.actor.ActorCell.invoke(ActorCell.scala:456)
    at akka.dispatch.Mailbox.processMailbox(Mailbox.scala:237)
    at akka.dispatch.Mailbox.run(Mailbox.scala:219)
    at akka.dispatch.ForkJoinExecutorConfigurator$AkkaForkJoinTask.exec(AbstractDispatcher.scala:386)
    at scala.concurrent.forkjoin.ForkJoinTask.doExec(ForkJoinTask.java:260)
    at scala.concurrent.forkjoin.ForkJoinPool$WorkQueue.runTask(ForkJoinPool.java:1339)
    at scala.concurrent.forkjoin.ForkJoinPool.runWorker(ForkJoinPool.java:1979)
    at scala.concurrent.forkjoin.ForkJoinWorkerThread.run(ForkJoinWorkerThread.java:107)

scala> data.foreach(println)
org.apache.spark.SparkException: Job aborted due to stage failure: Task not serializable: java.io.NotSerializableException: org.apache.hadoop.conf.Configuration
    ...

scala>val data=raw\u data.filter（x=>x.split（FIELD\u SEP.size>=3）
数据：org.apache.spark.rdd.rdd[String]=filtereddd[4]位于22处的过滤器
scala>data.collect
org.apache.spark.sparkeexception:作业因阶段失败而中止：任务不可序列化：java.io.notserializableeexception:org.apache.hadoop.conf.conf
位于org.apache.spark.scheduler.DAGScheduler.org$apache$spark$scheduler$DAGScheduler$$failJobAndIndependentStages（DAGScheduler.scala:1049）
位于org.apache.spark.scheduler.DAGScheduler$$anonfun$abortStage$1.apply（DAGScheduler.scala:1033）
位于org.apache.spark.scheduler.DAGScheduler$$anonfun$abortStage$1.apply（DAGScheduler.scala:1031）
位于scala.collection.mutable.resizeblearray$class.foreach（resizeblearray.scala:59）
位于scala.collection.mutable.ArrayBuffer.foreach（ArrayBuffer.scala:47）
位于org.apache.spark.scheduler.DAGScheduler.abortStage（DAGScheduler.scala:1031）
位于org.apache.spark.scheduler.DAGScheduler.org$apache$spark$scheduler$DAGScheduler$$submitmissingstasks（DAGScheduler.scala:772）
位于org.apache.spark.scheduler.DAGScheduler.org$apache$spark$scheduler$DAGScheduler$$submitStage（DAGScheduler.scala:715）
位于org.apache.spark.scheduler.DAGScheduler.handleJobSubmitted（DAGScheduler.scala:699）
位于org.apache.spark.scheduler.DAGSchedulerEventProcessActor$$anonfun$receive$2.applyOrElse（DAGScheduler.scala:1203）
在akka.actor.ActorCell.receiveMessage（ActorCell.scala:498）
在akka.actor.ActorCell.invoke（ActorCell.scala:456）
位于akka.dispatch.Mailbox.processMailbox（Mailbox.scala:237）
在akka.dispatch.Mailbox.run（Mailbox.scala:219）
在akka.dispatch.ForkJoinExecutorConfigurator$AkkaForkJoinTask.exec（AbstractDispatcher.scala:386）
位于scala.concurrent.forkjoin.ForkJoinTask.doExec（ForkJoinTask.java:260）
位于scala.concurrent.forkjoin.ForkJoinPool$WorkQueue.runTask（ForkJoinPool.java:1339）
位于scala.concurrent.forkjoin.ForkJoinPool.runWorker（ForkJoinPool.java:1979）
在scala.concurrent.forkjoin.ForkJoinWorkerThread.run（ForkJoinWorkerThread.java:107）中
scala>data.foreach（println）
org.apache.spark.sparkeexception:作业因阶段失败而中止：任务不可序列化：java.io.notserializableeexception:org.apache.hadoop.conf.conf
...

为什么我不能操作RDD“数据”，而使用

sc.textFile（“mydata.txt”）

时一切正常？

---

以及如何修复它？

您得到这个异常是因为您正在关闭

org.apache.hadoop.conf.Configuration

，但它不是

可序列化的

Caused by: java.io.NotSerializableException: org.apache.hadoop.conf.Configuration
    at java.io.ObjectOutputStream.writeObject0(ObjectOutputStream.java:1183)
    at java.io.ObjectOutputStream.defaultWriteFields(ObjectOutputStream.java:1547)
    at java.io.ObjectOutputStream.writeSerialData(ObjectOutputStream.java:1508)
    at java.io.ObjectOutputStream.writeOrdinaryObject(ObjectOutputStream.java:1431)
    at java.io.ObjectOutputStream.writeObject0(ObjectOutputStream.java:1177)
    at java.io.ObjectOutputStream.defaultWriteFields(ObjectOutputStream.java:1547)
    at java.io.ObjectOutputStream.writeSerialData(ObjectOutputStream.java:1508)
    at java.io.ObjectOutputStream.writeOrdinaryObject(ObjectOutputStream.java:1431)
    at java.io.ObjectOutputStream.writeObject0(ObjectOutputStream.java:1177)

你可以做两件事： 1.使用Kyro序列化程序或 2.只需将conf变量标记为

transient

，这基本上告诉Spark不要随闭包一起发送

scala> @transient val conf = new Configuration
conf: org.apache.hadoop.conf.Configuration = Configuration: core-default.xml, core-site.xml, mapred-default.xml, mapred-site.xml

scala> val raw_data = sc.newAPIHadoopFile("../test.txt", classOf[TextInputFormat], classOf[LongWritable], classOf[Text], conf).map(_._2.toString)
14/11/28 00:54:03 INFO MemoryStore: ensureFreeSpace(32937) called with curMem=70594, maxMem=278302556
14/11/28 00:54:03 INFO MemoryStore: Block broadcast_4 stored as values in memory (estimated size 32.2 KB, free 265.3 MB)
raw_data: org.apache.spark.rdd.RDD[String] = MappedRDD[5] at map at <console>:18

scala> val data = raw_data.filter{x => x.split(FIELD_SEP).size >= 3}
data: org.apache.spark.rdd.RDD[String] = FilteredRDD[6] at filter at <console>:22

scala> data.count
14/11/28 00:54:16 INFO FileInputFormat: Total input paths to process : 1
14/11/28 00:54:16 INFO SparkContext: Starting job: count at <console>:25
14/11/28 00:54:16 INFO DAGScheduler: Got job 2 (count at <console>:25) with 1 output partitions (allowLocal=false)
14/11/28 00:54:16 INFO DAGScheduler: Final stage: Stage 2(count at <console>:25)
14/11/28 00:54:16 INFO DAGScheduler: Parents of final stage: List()
14/11/28 00:54:16 INFO DAGScheduler: Missing parents: List()
14/11/28 00:54:16 INFO DAGScheduler: Submitting Stage 2 (FilteredRDD[6] at filter at <console>:22), which has no missing parents
14/11/28 00:54:16 INFO MemoryStore: ensureFreeSpace(4488) called with curMem=103531, maxMem=278302556
14/11/28 00:54:16 INFO MemoryStore: Block broadcast_5 stored as values in memory (estimated size 4.4 KB, free 265.3 MB)
14/11/28 00:54:16 INFO DAGScheduler: Submitting 1 missing tasks from Stage 2 (FilteredRDD[6] at filter at <console>:22)
14/11/28 00:54:16 INFO TaskSchedulerImpl: Adding task set 2.0 with 1 tasks
14/11/28 00:54:16 INFO TaskSetManager: Starting task 0.0 in stage 2.0 (TID 2, localhost, PROCESS_LOCAL, 1223 bytes)
14/11/28 00:54:16 INFO Executor: Running task 0.0 in stage 2.0 (TID 2)
14/11/28 00:54:16 INFO NewHadoopRDD: Input split: file:/Users/ssimanta/spark/test.txt:0+123
14/11/28 00:54:16 INFO Executor: Finished task 0.0 in stage 2.0 (TID 2). 1731 bytes result sent to driver
14/11/28 00:54:16 INFO TaskSetManager: Finished task 0.0 in stage 2.0 (TID 2) in 19 ms on localhost (1/1)
14/11/28 00:54:16 INFO DAGScheduler: Stage 2 (count at <console>:25) finished in 0.019 s
14/11/28 00:54:16 INFO TaskSchedulerImpl: Removed TaskSet 2.0, whose tasks have all completed, from pool 
14/11/28 00:54:16 INFO DAGScheduler: Job 2 finished: count at <console>:25, took 0.025300 s
res5: Long = 1

scala> data.collect
14/11/28 00:55:16 INFO SparkContext: Starting job: collect at <console>:25
14/11/28 00:55:16 INFO DAGScheduler: Got job 3 (collect at <console>:25) with 1 output partitions (allowLocal=false)
14/11/28 00:55:16 INFO DAGScheduler: Final stage: Stage 3(collect at <console>:25)
14/11/28 00:55:16 INFO DAGScheduler: Parents of final stage: List()
14/11/28 00:55:16 INFO DAGScheduler: Missing parents: List()
14/11/28 00:55:16 INFO DAGScheduler: Submitting Stage 3 (FilteredRDD[6] at filter at <console>:22), which has no missing parents
14/11/28 00:55:16 INFO MemoryStore: ensureFreeSpace(4504) called with curMem=108019, maxMem=278302556
14/11/28 00:55:16 INFO MemoryStore: Block broadcast_6 stored as values in memory (estimated size 4.4 KB, free 265.3 MB)
14/11/28 00:55:16 INFO DAGScheduler: Submitting 1 missing tasks from Stage 3 (FilteredRDD[6] at filter at <console>:22)
14/11/28 00:55:16 INFO TaskSchedulerImpl: Adding task set 3.0 with 1 tasks
14/11/28 00:55:16 INFO TaskSetManager: Starting task 0.0 in stage 3.0 (TID 3, localhost, PROCESS_LOCAL, 1223 bytes)
14/11/28 00:55:16 INFO Executor: Running task 0.0 in stage 3.0 (TID 3)
14/11/28 00:55:16 INFO NewHadoopRDD: Input split: file:/Users/ssimanta/spark/test.txt:0+123
14/11/28 00:55:16 INFO Executor: Finished task 0.0 in stage 3.0 (TID 3). 1717 bytes result sent to driver
14/11/28 00:55:16 INFO TaskSetManager: Finished task 0.0 in stage 3.0 (TID 3) in 16 ms on localhost (1/1)
14/11/28 00:55:16 INFO DAGScheduler: Stage 3 (collect at <console>:25) finished in 0.017 s
14/11/28 00:55:16 INFO TaskSchedulerImpl: Removed TaskSet 3.0, whose tasks have all completed, from pool 
14/11/28 00:55:16 INFO DAGScheduler: Job 3 finished: collect at <console>:25, took 0.021439 s
res6: Array[String] = Array(MMMM_|NNNNN_|OOOOO\ |)

scala>@transient val conf=新配置
conf:org.apache.hadoop.conf.Configuration=Configuration:core-default.xml、core-site.xml、mapred-default.xml、mapred-site.xml
scala>val raw_data=sc.newAPIHadoopFile（../test.txt），classOf[TextInputFormat]，classOf[LongWritable]，classOf[Text]，conf）.map（u.\u 2.toString）
14/11/28 00:54:03信息内存存储：使用curMem=70594、maxMem=278302556调用EnsureRefreeSpace（32937）
14/11/28 00:54:03信息内存存储：块广播_4作为值存储在内存中（估计大小32.2 KB，可用容量265.3 MB）
原始数据：org.apache.spark.rdd.rdd[String]=MappedRDD[5]位于map at:18
scala>val data=raw_data.filter{x=>x.split（FIELD_SEP.size>=3}
数据：org.apache.spark.rdd.rdd[String]=filtereddd[6]位于：22处的过滤器
scala>data.count
14/11/28 00:54:16信息文件InputFormat:要处理的总输入路径：1
14/11/28 00:54:16信息SparkContext:开始作业：计数时间：25
14/11/28 00:54:16信息调度程序：获得了带有1个输出分区的作业2（计数：25）（allowLocal=false）
14/11/28 00:54:16信息调度程序：最后阶段：第二阶段（计数：25）
14/11/28 00:54:16信息调度程序：最终阶段的父级：列表（）
14/11/28 00:54:16信息调度程序：缺少的家长：列表（）
14/11/28 00:54:16信息调度程序：正在提交第2阶段（filter at:22处的FilteredRDD[6]），该阶段没有丢失的父级
14/11/28 00:54:16信息内存存储：使用curMem=103531、maxMem=278302556调用EnsureRefreeSpace（4488）
14/11/28 00:54:16信息内存存储：块广播_5作为值存储在内存中（估计大小4.4 KB，可用容量265.3 MB）
14/11/28 00:54:16信息调度程序：从阶段2提交1个缺少的任务（FilteredRDD[6]位于：22处的筛选器）
14/11/28 00:54:16信息TaskSchedulerImpl:添加任务集2.0和1个任务
14/11/28 00:54:16信息任务集管理器：在阶段2.0中启动任务0.0（TID 2，本地主机，进程\本地，1223字节）
14/11/28 00:54:16信息执行者：在阶段2.0（TID 2）中运行任务0.0
14/11/28 00:54:16信息NewHadoopRDD:输入拆分：文件：/Users/ssimanta/spark/test.txt:0+123
14/11/28 00:54:16信息执行者：已完成阶段2.0（TID 2）中的任务0.0。1731字节结果发送到驱动程序
14/11/28 00:54:16信息任务集管理器：在本地主机（1/1）上以19毫秒的时间完成阶段2.0（TID 2）中的任务0.0
14/11/28 00:54:16信息调度程序：第2阶段（计数：25）在0.019秒内完成
14/11/28 00:54:16信息TaskSchedulerImpl:已从池中删除任务集2.0，其任务已全部完成
14/11/28 00:54:16信息调度程序：作业2已完成：计数时间：25，耗时0.025300秒
res5:Long=1
scala>data.collect
14/11/28 00:55:16信息SparkContext:开始作业：收取时间：25
14/11/28 00:55:16信息调度程序：获得作业3（收集时间：25）wi