Spark：GraphX 无法在边数少、路径长的图中找到连通分量

Question

我是 Spark 和 GraphX 的新手，用它的算法做了一些实验来找到连接的组件。我注意到图形的结构似乎对性能有很大影响。

它能够计算具有数百万个顶点和边的图，但是对于特定的一组图，算法没有及时完成，但最终失败并显示OutOfMemoryError: GC overhead limit exceeded。

该算法似乎对包含长路径的图形有问题。例如，对于此图 { (i,i+1) | i <- {1..200} }，计算失败。但是，当我添加传递边时，计算立即完成：

{ (i,j) | i <- {1..200}, j <- {i+1,200} }

像这样的图表也没有问题：

{ (i,1) | i <- {1..200} }

这是重现问题的最小示例：

import org.apache.spark._
import org.apache.spark.graphx._
import org.apache.spark.graphx.lib._
import org.apache.spark.storage.StorageLevel

import scala.collection.mutable

object Matching extends Logging {

  def main(args: Array[String]): Unit = {
    val fname = "input.graph"
    val optionsList = args.drop(1).map { arg =>
      arg.dropWhile(_ == '-').split('=') match {
        case Array(opt, v) => opt -> v
        case _ => throw new IllegalArgumentException("Invalid argument: " + arg)
      }
    }
    val options = mutable.Map(optionsList: _*)

    val conf = new SparkConf()
    GraphXUtils.registerKryoClasses(conf)

    val partitionStrategy: Option[PartitionStrategy] = options.remove("partStrategy")
      .map(PartitionStrategy.fromString(_))
    val edgeStorageLevel = options.remove("edgeStorageLevel")
      .map(StorageLevel.fromString(_)).getOrElse(StorageLevel.MEMORY_ONLY)
    val vertexStorageLevel = options.remove("vertexStorageLevel")
      .map(StorageLevel.fromString(_)).getOrElse(StorageLevel.MEMORY_ONLY)

    val sc = new SparkContext(conf.setAppName("ConnectedComponents(" + fname + ")"))
    val unpartitionedGraph = GraphLoader.edgeListFile(sc, fname,
      edgeStorageLevel = edgeStorageLevel,
      vertexStorageLevel = vertexStorageLevel).cache()
    log.info("Loading graph...")
    val graph = partitionStrategy.foldLeft(unpartitionedGraph)(_.partitionBy(_))
    log.info("Loading graph...done")

    log.info("Computing connected components...")
    val cc = ConnectedComponents.run(graph)
    log.info("Computed connected components...done")

    sc.stop()
  }
}

input.graph文件可以这样看（10个节点，9条边连接它们）：

失败时，挂起在 ConnectedComponents.run(graph)。错误信息是：

Exception in thread "dag-scheduler-event-loop" java.lang.OutOfMemoryError: GC overhead limit exceeded
    at java.util.regex.Pattern.compile(Pattern.java:1054)
    at java.lang.String.replace(String.java:2239)
    at org.apache.spark.util.Utils$.getFormattedClassName(Utils.scala:1632)
    at org.apache.spark.storage.RDDInfo$$anonfun.apply(RDDInfo.scala:58)
    at org.apache.spark.storage.RDDInfo$$anonfun.apply(RDDInfo.scala:58)
    at scala.Option.getOrElse(Option.scala:121)
    at org.apache.spark.storage.RDDInfo$.fromRdd(RDDInfo.scala:58)
    at org.apache.spark.scheduler.StageInfo$$anonfun.apply(StageInfo.scala:80)
    at org.apache.spark.scheduler.StageInfo$$anonfun.apply(StageInfo.scala:80)
    at scala.collection.TraversableLike$$anonfun$map.apply(TraversableLike.scala:245)
    at scala.collection.TraversableLike$$anonfun$map.apply(TraversableLike.scala:245)
    at scala.collection.mutable.ResizableArray$class.foreach(ResizableArray.scala:59)
    at scala.collection.mutable.ArrayBuffer.foreach(ArrayBuffer.scala:48)
    at scala.collection.TraversableLike$class.map(TraversableLike.scala:245)
    at scala.collection.AbstractTraversable.map(Traversable.scala:104)
    at org.apache.spark.scheduler.StageInfo$.fromStage(StageInfo.scala:80)
    at org.apache.spark.scheduler.Stage.<init>(Stage.scala:99)
    at org.apache.spark.scheduler.ShuffleMapStage.<init>(ShuffleMapStage.scala:44)
    at org.apache.spark.scheduler.DAGScheduler.newShuffleMapStage(DAGScheduler.scala:317)
    at org.apache.spark.scheduler.DAGScheduler.org$apache$spark$scheduler$DAGScheduler$$newOrUsedShuffleStage(DAGScheduler.scala:352)
    at org.apache.spark.scheduler.DAGScheduler$$anonfun$org$apache$spark$scheduler$DAGScheduler$$getShuffleMapStage.apply(DAGScheduler.scala:286)
    at org.apache.spark.scheduler.DAGScheduler$$anonfun$org$apache$spark$scheduler$DAGScheduler$$getShuffleMapStage.apply(DAGScheduler.scala:285)
    at scala.collection.Iterator$class.foreach(Iterator.scala:742)
    at scala.collection.AbstractIterator.foreach(Iterator.scala:1194)
    at scala.collection.IterableLike$class.foreach(IterableLike.scala:72)
    at scala.collection.mutable.Stack.foreach(Stack.scala:170)
    at org.apache.spark.scheduler.DAGScheduler.org$apache$spark$scheduler$DAGScheduler$$getShuffleMapStage(DAGScheduler.scala:285)
    at org.apache.spark.scheduler.DAGScheduler$$anonfun$visit.apply(DAGScheduler.scala:389)
    at org.apache.spark.scheduler.DAGScheduler$$anonfun$visit.apply(DAGScheduler.scala:386)
    at scala.collection.immutable.List.foreach(List.scala:381)
    at org.apache.spark.scheduler.DAGScheduler.visit(DAGScheduler.scala:386)
    at org.apache.spark.scheduler.DAGScheduler.getParentStages(DAGScheduler.scala:398)

我是运行本地 Spark 节点并使用以下选项启动 JVM：

-Dspark.master=local -Dspark.local.dir=/home/phil/tmp/spark-tmp -Xms8g -Xmx8g

你能帮我理解为什么这个玩具图（201 个节点和 200 条边）有问题，但另一方面可以在大约 80 秒内解决具有数百万条边的真实图吗？（在这两个示例中，我使用相同的设置和配置。）

更新：

也可以在spark中复现-shell:

import org.apache.spark.graphx._
import org.apache.spark.graphx.lib._

val graph = GraphLoader.edgeListFile(sc, "input.graph").cache()
ConnectedComponents.run(graph)

我创建了一个错误报告：SPARK-15042

Answer 1

根据SPARK-15042，问题仍然存在于 2.1.0-snapshot。

可以在 SPARK-5484.

中看到修复错误的进度

Spark：GraphX 无法在边数少、路径长的图中找到连通分量

Spark: GraphX fails to find connected components in graphs with few edges and long paths

apache-spark

spark-graphx

connected-components