Scala spark, listbuffer is empty

问题

In this piece of code in comment 1 length of listbuffer items is shown correctly, but in the 2nd comment code never executes. Why it is occurs?

val conf = new SparkConf().setAppName("app").setMaster("local")
val sc = new SparkContext(conf)

var wktReader: WKTReader = new WKTReader(); 
val dataSet = sc.textFile("dataSet.txt")

val items = new ListBuffer[String]() 
dataSet.foreach { e =>
  items += e
  println("len = " + items.length) //1. here length is ok
}

println("!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!")
items.foreach { x => print(x)} //2. this code doesn't be executed

Logs are here:

16/11/20 01:16:52 INFO Utils: Successfully started service 'SparkUI' on port 4040.
    16/11/20 01:16:52 INFO SparkUI: Bound SparkUI to 0.0.0.0, and started at http://192.168.56.1:4040
    16/11/20 01:16:53 INFO Executor: Starting executor ID driver on host localhost
    16/11/20 01:16:53 INFO Utils: Successfully started service 'org.apache.spark.network.netty.NettyBlockTransferService' on port 58608.
    16/11/20 01:16:53 INFO NettyBlockTransferService: Server created on 192.168.56.1:58608
    16/11/20 01:16:53 INFO BlockManagerMaster: Registering BlockManager BlockManagerId(driver, 192.168.56.1, 58608)
    16/11/20 01:16:53 INFO BlockManagerMasterEndpoint: Registering block manager 192.168.56.1:58608 with 347.1 MB RAM, BlockManagerId(driver, 192.168.56.1, 58608)
    16/11/20 01:16:53 INFO BlockManagerMaster: Registered BlockManager BlockManagerId(driver, 192.168.56.1, 58608)
    Starting app
    16/11/20 01:16:57 INFO MemoryStore: Block broadcast_0 stored as values in memory (estimated size 139.6 KB, free 347.0 MB)
    16/11/20 01:16:58 INFO MemoryStore: Block broadcast_0_piece0 stored as bytes in memory (estimated size 15.9 KB, free 346.9 MB)
    16/11/20 01:16:58 INFO BlockManagerInfo: Added broadcast_0_piece0 in memory on 192.168.56.1:58608 (size: 15.9 KB, free: 347.1 MB)
    16/11/20 01:16:58 INFO SparkContext: Created broadcast 0 from textFile at main.scala:25
    16/11/20 01:16:58 INFO FileInputFormat: Total input paths to process : 1
    16/11/20 01:16:58 INFO SparkContext: Starting job: foreach at main.scala:28
    16/11/20 01:16:58 INFO DAGScheduler: Got job 0 (foreach at main.scala:28) with 1 output partitions
    16/11/20 01:16:58 INFO DAGScheduler: Final stage: ResultStage 0 (foreach at main.scala:28)
    16/11/20 01:16:58 INFO DAGScheduler: Parents of final stage: List()
    16/11/20 01:16:58 INFO DAGScheduler: Missing parents: List()
    16/11/20 01:16:58 INFO DAGScheduler: Submitting ResultStage 0 (dataSet.txt MapPartitionsRDD[1] at textFile at main.scala:25), which has no missing parents
    16/11/20 01:16:58 INFO MemoryStore: Block broadcast_1 stored as values in memory (estimated size 3.3 KB, free 346.9 MB)
    16/11/20 01:16:58 INFO MemoryStore: Block broadcast_1_piece0 stored as bytes in memory (estimated size 2034.0 B, free 346.9 MB)
    16/11/20 01:16:58 INFO BlockManagerInfo: Added broadcast_1_piece0 in memory on 192.168.56.1:58608 (size: 2034.0 B, free: 347.1 MB)
    16/11/20 01:16:58 INFO SparkContext: Created broadcast 1 from broadcast at DAGScheduler.scala:1012
    16/11/20 01:16:59 INFO DAGScheduler: Submitting 1 missing tasks from ResultStage 0 (dataSet.txt MapPartitionsRDD[1] at textFile at main.scala:25)
    16/11/20 01:16:59 INFO TaskSchedulerImpl: Adding task set 0.0 with 1 tasks
    16/11/20 01:16:59 INFO TaskSetManager: Starting task 0.0 in stage 0.0 (TID 0, localhost, partition 0, PROCESS_LOCAL, 5427 bytes)
    16/11/20 01:16:59 INFO Executor: Running task 0.0 in stage 0.0 (TID 0)
    16/11/20 01:16:59 INFO HadoopRDD: Input split: file:/D:/dataSet.txt:0+291
    16/11/20 01:16:59 INFO deprecation: mapred.tip.id is deprecated. Instead, use mapreduce.task.id
    16/11/20 01:16:59 INFO deprecation: mapred.task.id is deprecated. Instead, use mapreduce.task.attempt.id
    16/11/20 01:16:59 INFO deprecation: mapred.task.is.map is deprecated. Instead, use mapreduce.task.ismap
    16/11/20 01:16:59 INFO deprecation: mapred.task.partition is deprecated. Instead, use mapreduce.task.partition
    16/11/20 01:16:59 INFO deprecation: mapred.job.id is deprecated. Instead, use mapreduce.job.id
    len = 1
    len = 2
    len = 3
    len = 4
    len = 5
    len = 6
    len = 7
    16/11/20 01:16:59 INFO Executor: Finished task 0.0 in stage 0.0 (TID 0). 989 bytes result sent to driver
    16/11/20 01:16:59 INFO TaskSetManager: Finished task 0.0 in stage 0.0 (TID 0) in 417 ms on localhost (1/1)
    16/11/20 01:16:59 INFO TaskSchedulerImpl: Removed TaskSet 0.0, whose tasks have all completed, from pool 
    16/11/20 01:16:59 INFO DAGScheduler: ResultStage 0 (foreach at main.scala:28) finished in 0,456 s
    16/11/20 01:16:59 INFO DAGScheduler: Job 0 finished: foreach at main.scala:28, took 0,795126 s
    !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
    16/11/20 01:16:59 INFO SparkContext: Invoking stop() from shutdown hook
    16/11/20 01:16:59 INFO SparkUI: Stopped Spark web UI at http://192.168.56.1:4040
    16/11/20 01:16:59 INFO MapOutputTrackerMasterEndpoint: MapOutputTrackerMasterEndpoint stopped!
    16/11/20 01:16:59 INFO MemoryStore: MemoryStore cleared
    16/11/20 01:16:59 INFO BlockManager: BlockManager stopped
    16/11/20 01:16:59 INFO BlockManagerMaster: BlockManagerMaster stopped
    16/11/20 01:16:59 INFO OutputCommitCoordinator$OutputCommitCoordinatorEndpoint: OutputCommitCoordinator stopped!
    16/11/20 01:16:59 INFO SparkContext: Successfully stopped SparkContext
    16/11/20 01:16:59 INFO ShutdownHookManager: Shutdown hook called
    16/11/20 01:16:59 INFO ShutdownHookManager: Deleting directory

回答1:

Apache Spark doesn't provide shared memory therefore here:

dataSet.foreach { e =>
  items += e
  println("len = " + items.length) //1. here length is ok
}

you modify a local copy of items on a respective exectuor. The original items list defined on the driver is not modified. As a result this:

items.foreach { x => print(x) }

executes, but there is nothing to print.

Please check Understanding closures

While it would be recommended here, you could replace items with an accumulator

val acc = sc.collectionAccumulator[String]("Items")
dataSet.foreach(e => acc.add(e))

回答2:

Spark runs in executers and returns the results. The above code doesn't work as intended. If you need to add the elements from foreach then need to collect the data in the driver and add to the current_set. But collecting the data is a bad idea when you have large data.

val items = new ListBuffer[String]()

val rdd = spark.sparkContext.parallelize(1 to 10, 4)
rdd.collect().foreach(data => items += data.toString())
println(items)

Output:

ListBuffer(1, 2, 3, 4, 5, 6, 7, 8, 9, 10)

来源：https://stackoverflow.com/questions/52645867/scala-not-save-modify-sequence-when-exit-to-foreach

标签

scala

apache-spark

listbuffer