[SPARK-28927] Improve error for ArrayIndexOutOfBoundsException and Not-stable AUC metrics in ALS for datasets with 12 billion instances

• Type: Improvement
• Status: Resolved
• Priority: Minor
• Resolution: Fixed
• Affects Version/s: 2.2.1
• Fix Version/s: 3.0.0
• Component/s: ML
• Assignee: L. C. Hsieh
• Reporter: Qiang Wang
• Created: 30/Aug/19 07:59
• Updated: 18/Sep/19 14:22
• Resolved: 18/Sep/19 14:22

Description

The stack trace is below:

19/08/28 07:00:40 WARN Executor task launch worker for task 325074 BlockManager: Block rdd_10916_493 could not be removed as it was not found on disk or in memory 19/08/28 07:00:41 ERROR Executor task launch worker for task 325074 Executor: Exception in task 3.0 in stage 347.1 (TID 325074) java.lang.ArrayIndexOutOfBoundsException: 6741 at org.apache.spark.dpshade.recommendation.ALS$$anonfun$org$apache$spark$ml$recommendation$ALS$$computeFactors$1.apply(ALS.scala:1460) at org.apache.spark.dpshade.recommendation.ALS$$anonfun$org$apache$spark$ml$recommendation$ALS$$computeFactors$1.apply(ALS.scala:1440) at org.apache.spark.rdd.PairRDDFunctions$$anonfun$mapValues$1$$anonfun$apply$40$$anonfun$apply$41.apply(PairRDDFunctions.scala:760) at org.apache.spark.rdd.PairRDDFunctions$$anonfun$mapValues$1$$anonfun$apply$40$$anonfun$apply$41.apply(PairRDDFunctions.scala:760) at scala.collection.Iterator$$anon$11.next(Iterator.scala:409) at org.apache.spark.storage.memory.MemoryStore.putIteratorAsValues(MemoryStore.scala:216) at org.apache.spark.storage.BlockManager$$anonfun$doPutIterator$1.apply(BlockManager.scala:1041) at org.apache.spark.storage.BlockManager$$anonfun$doPutIterator$1.apply(BlockManager.scala:1032) at org.apache.spark.storage.BlockManager.doPut(BlockManager.scala:972) at org.apache.spark.storage.BlockManager.doPutIterator(BlockManager.scala:1032) at org.apache.spark.storage.BlockManager.getOrElseUpdate(BlockManager.scala:763) at org.apache.spark.rdd.RDD.getOrCompute(RDD.scala:334) at org.apache.spark.rdd.RDD.iterator(RDD.scala:285) at org.apache.spark.rdd.CoGroupedRDD$$anonfun$compute$2.apply(CoGroupedRDD.scala:141) at org.apache.spark.rdd.CoGroupedRDD$$anonfun$compute$2.apply(CoGroupedRDD.scala:137) at scala.collection.TraversableLike$WithFilter$$anonfun$foreach$1.apply(TraversableLike.scala:733) at scala.collection.immutable.List.foreach(List.scala:381) at scala.collection.TraversableLike$WithFilter.foreach(TraversableLike.scala:732) at org.apache.spark.rdd.CoGroupedRDD.compute(CoGroupedRDD.scala:137) at org.apache.spark.rdd.RDD.computeOrReadCheckpoint(RDD.scala:323) at org.apache.spark.rdd.RDD.iterator(RDD.scala:287) at org.apache.spark.rdd.MapPartitionsRDD.compute(MapPartitionsRDD.scala:38) at org.apache.spark.rdd.RDD.computeOrReadCheckpoint(RDD.scala:323) at org.apache.spark.rdd.RDD.iterator(RDD.scala:287) at org.apache.spark.rdd.MapPartitionsRDD.compute(MapPartitionsRDD.scala:38) at org.apache.spark.rdd.RDD.computeOrReadCheckpoint(RDD.scala:323) at org.apache.spark.rdd.RDD.iterator(RDD.scala:287) at org.apache.spark.rdd.MapPartitionsRDD.compute(MapPartitionsRDD.scala:38) at org.apache.spark.rdd.RDD.computeOrReadCheckpoint(RDD.scala:323) at org.apache.spark.rdd.RDD.iterator(RDD.scala:287) at org.apache.spark.scheduler.ShuffleMapTask.runTask(ShuffleMapTask.scala:96) at org.apache.spark.scheduler.ShuffleMapTask.runTask(ShuffleMapTask.scala:53) at org.apache.spark.scheduler.Task.run(Task.scala:108) at org.apache.spark.executor.Executor$TaskRunner.run(Executor.scala:358) at java.util.concurrent.ThreadPoolExecutor.runWorker(ThreadPoolExecutor.java:1142) at java.util.concurrent.ThreadPoolExecutor$Worker.run(ThreadPoolExecutor.java:617) at java.lang.Thread.run(Thread.java:745)

This exception happened sometimes.  And we also found that the AUC metric was not stable when evaluating the inner product of the user factors and the item factors with the same dataset and configuration. AUC varied from 0.60 to 0.67 which was not stable for production environment. 

Dataset capacity: ~12 billion ratings
Here is the our code:

val hivedata = sc.sql(sqltext).select("id", "dpid", "score", "tag")
            .repartition(6000).persist(StorageLevel.MEMORY_AND_DISK_SER)
val zeroValueArrItem = ArrayBuffer[(String, Int)]()
val predataItem = hivedata.
  map(r => (r.getString(0), (r.getString(1), r.getInt(2)))).rdd.
  aggregateByKey(zeroValueArrItem, 6000)((a, b) => a += b, (a, b) => a ++ b).
  zipWithIndex().
  setName(predataItemName).
  persist(StorageLevel.MEMORY_AND_DISK_SER)
val zeroValueArr = ArrayBuffer[(Int, Int)]()
val predataUser = predataItem.
  flatMap(r => r._1._2.map(y => (y._1, (r._2.toInt, y._2)))).
  aggregateByKey(zeroValueArr, 6000)((a, b) => a += b, (a, b) => a ++ b).
  zipWithIndex().setName(predataUserName).persist(StorageLevel.MEMORY_AND_DISK_SER)
val trainData = predataUser.flatMap(x => x._1._2.map(y => (x._2.toInt, y._1, y._2.toFloat)))
 .setName(trainDataName).persist(StorageLevel.MEMORY_AND_DISK_SER)case class ALSData(user:Int, item:Int, rating:Float) extends Serializable
 val ratingData = trainData.map(x => ALSData(x._1, x._2, x._3)).toDF()
 val als = new ALS
 val paramMap = ParamMap(als.alpha -> 25000).
 put(als.checkpointInterval, 5).
 put(als.implicitPrefs, true).
 put(als.itemCol, "item").
 put(als.maxIter, 60).
 put(als.nonnegative, false).
 put(als.numItemBlocks, 600).
 put(als.numUserBlocks, 600).
 put(als.regParam, 4.5).
 put(als.rank, 25).
 put(als.userCol, "user")
 als.fit(ratingData, paramMap)

Issue Links

• links to: GitHub Pull Request #25789