scala – 当按键分组时,Spark耗尽内存

我试图使用 this guide在EC2上使用Spark主机执行常见的爬网数据的简单转换,我的代码如下所示：

package ccminer

import org.apache.spark.SparkContext
import org.apache.spark.SparkContext._

object ccminer {
  val english = "english|en|eng"
  val spanish = "es|esp|spa|spanish|espanol"
  val turkish = "turkish|tr|tur|turc"
  val greek = "greek|el|ell"
  val italian = "italian|it|ita|italien"
  val all = (english :: spanish :: turkish :: greek :: italian :: Nil).mkString("|")

  Def LangIndep(s: String) = s.toLowerCase().replaceAll(all,"*")

  def main(args: Array[String]): Unit = {
    if (args.length != 3) {
      System.err.println("Bad command line")
      System.exit(-1)
    }

    val cluster = "spark://???"
    val sc = new SparkContext(cluster,"Common Crawl miner",System.getenv("SPARK_HOME"),Seq("/root/spark/ccminer/target/scala-2.10/cc-miner_2.10-1.0.jar"))

    sc.sequenceFile[String,String](args(0)).map {
      case (k,v) => (langIndep(k),v)
    }
    .groupByKey(args(2).toInt)
    .filter {
      case (_,vs) => vs.size > 1
    }
    .saveAsTextFile(args(1))
  }
}

我正在运行它的命令如下：

sbt/sbt "run-main ccminer.ccminer s3n://aws-publicdatasets/common-crawl/parse-output/segment/1341690165636/textData-* s3n://parallelcorpus/out/ 2000"

但是很快就失败了,错误如下

java.lang.OutOfMemoryError: Java heap space
at com.ning.compress.BufferRecycler.allocEncodingBuffer(BufferRecycler.java:59)
at com.ning.compress.lzf.ChunkEncoder.<init>(ChunkEncoder.java:93)
at com.ning.compress.lzf.impl.UnsafeChunkEncoder.<init>(UnsafeChunkEncoder.java:40)
at com.ning.compress.lzf.impl.UnsafeChunkEncoderLE.<init>(UnsafeChunkEncoderLE.java:13)
at com.ning.compress.lzf.impl.UnsafeChunkEncoders.createEncoder(UnsafeChunkEncoders.java:31)
at com.ning.compress.lzf.util.ChunkEncoderFactory.optimalInstance(ChunkEncoderFactory.java:44)
at com.ning.compress.lzf.LZFOutputStream.<init>(LZFOutputStream.java:61)
at org.apache.spark.io.LZFCompressionCodec.compressedOutputStream(CompressionCodec.scala:60)
at org.apache.spark.storage.BlockManager.wrapForCompression(BlockManager.scala:803)
at org.apache.spark.storage.BlockManager$$anonfun$5.apply(BlockManager.scala:471)
at org.apache.spark.storage.BlockManager$$anonfun$5.apply(BlockManager.scala:471)
at org.apache.spark.storage.diskBlockObjectWriter.open(BlockObjectWriter.scala:117)
at org.apache.spark.storage.diskBlockObjectWriter.write(BlockObjectWriter.scala:174)
at org.apache.spark.scheduler.ShuffleMapTask$$anonfun$runTask$1.apply(ShuffleMapTask.scala:164)
at org.apache.spark.scheduler.ShuffleMapTask$$anonfun$runTask$1.apply(ShuffleMapTask.scala:161)
at scala.collection.Iterator$class.foreach(Iterator.scala:727)
at scala.collection.AbstractIterator.foreach(Iterator.scala:1157)
at org.apache.spark.scheduler.ShuffleMapTask.runTask(ShuffleMapTask.scala:161)
at org.apache.spark.scheduler.ShuffleMapTask.runTask(ShuffleMapTask.scala:102)
at org.apache.spark.scheduler.Task.run(Task.scala:53)
at org.apache.spark.executor.Executor$TaskRunner$$anonfun$run$1.apply$mcV$sp(Executor.scala:213)
at org.apache.spark.deploy.SparkHadoopUtil.runAsUser(SparkHadoopUtil.scala:49)
at org.apache.spark.executor.Executor$TaskRunner.run(Executor.scala:178)
at java.util.concurrent.ThreadPoolExecutor.runWorker(ThreadPoolExecutor.java:1145)
at java.util.concurrent.ThreadPoolExecutor$Worker.run(ThreadPoolExecutor.java:615)
at java.lang.Thread.run(Thread.java:744)

所以我的基本问题是,需要写一个Spark任务,可以通过键分组几乎无限量的输入,而不会耗尽内存？

解决方法

随机播放任务中java.lang.OutOfMemoryError异常的最常见原因(如groupByKey,reduceByKey等)为 parallelism的低级别.

您可以通过在configuration中设置spark.default.parallelism属性来增加默认值.