CODEWITHSUNDEEP
scalaapache-sparkakka
Renkai
Renkai

Reputation: 2121

Why spark broadcast doesn't work well when I use extends App?

The first code throws null pointer exception.

object TryBroadcast extends App{
  val conf = new SparkConf().setAppName("o_o")
  val sc = new SparkContext(conf)
  val sample = sc.parallelize(1 to 1024)
  val bro = sc.broadcast(6666)
  val broSample = sample.map(x => x.toString + bro.value)
  broSample.collect().foreach(println)
}

The second works well.

object TryBroadcast {
  def main(args: Array[String]) {
    val conf = new SparkConf().setAppName("o_o")
    val sc = new SparkContext(conf)
    val sample = sc.parallelize(1 to 1024)
    val bro = sc.broadcast(6666)
    val broSample = sample.map(x => x.toString + bro.value)
    broSample.collect().foreach(println)
  }
}

It seems spark broadcast has something conflict with scala.App

scala version: 2.10.5 spark version: 1.4.0 stackTrace:

lang.NullPointerException
    at TryBroadcast$$anonfun$1.apply(TryBroadcast.scala:11)
    at TryBroadcast$$anonfun$1.apply(TryBroadcast.scala:11)
    at scala.collection.Iterator$$anon$11.next(Iterator.scala:328)
    at scala.collection.Iterator$class.foreach(Iterator.scala:727)
    at scala.collection.AbstractIterator.foreach(Iterator.scala:1157)
    at scala.collection.generic.Growable$class.$plus$plus$eq(Growable.scala:48)
    at scala.collection.mutable.ArrayBuffer.$plus$plus$eq(ArrayBuffer.scala:103)
    at scala.collection.mutable.ArrayBuffer.$plus$plus$eq(ArrayBuffer.scala:47)
    at scala.collection.TraversableOnce$class.to(TraversableOnce.scala:273)
    at scala.collection.AbstractIterator.to(Iterator.scala:1157)
    at scala.collection.TraversableOnce$class.toBuffer(TraversableOnce.scala:265)
    at scala.collection.AbstractIterator.toBuffer(Iterator.scala:1157)
    at scala.collection.TraversableOnce$class.toArray(TraversableOnce.scala:252)
    at scala.collection.AbstractIterator.toArray(Iterator.scala:1157)
    at org.apache.spark.rdd.RDD$$anonfun$collect$1$$anonfun$12.apply(RDD.scala:885)
    at org.apache.spark.rdd.RDD$$anonfun$collect$1$$anonfun$12.apply(RDD.scala:885)
    at org.apache.spark.SparkContext$$anonfun$runJob$5.apply(SparkContext.scala:1765)
    at org.apache.spark.SparkContext$$anonfun$runJob$5.apply(SparkContext.scala:1765)
    at org.apache.spark.scheduler.ResultTask.runTask(ResultTask.scala:63)
    at org.apache.spark.scheduler.Task.run(Task.scala:70)
    at org.apache.spark.executor.Executor$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)

Upvotes: 3

Views: 1022

Answers (2)

Pierre Mage
Pierre Mage

Reputation: 2296

It is not very well documented but it is recommended to use def main(args: Array[String]): Unit = ??? instead of extends App.

See https://issues.apache.org/jira/browse/SPARK-4170 and https://github.com/apache/spark/pull/3497

Upvotes: 1

Daniel Darabos
Daniel Darabos

Reputation: 27455

bro in the two cases is quite different. In the first one it's a field on a singleton class instance (TryBroadcast). In the second one it is a local variable.

I the local variable gets captured, serialized and sent over to the executors. In the first case the reference is to a field, so the singleton would get captured and sent. I'm not sure how a Scala singleton is built and how it is captured. Apparently in this case it ends up uninitialized when it is accessed on the executor.

You could make bro a local variable like this:

object TryBroadcast extends App {
  val conf = new SparkConf().setAppName("o_o")
  val sc = new SparkContext(conf)
  val sample = sc.parallelize(1 to 1024)
  val broSample = {
    val bro = sc.broadcast(6666)
    sample.map(x => x.toString + bro.value)
  }
  broSample.collect().foreach(println)
}

Upvotes: 2

Related Questions