[SPARK-21269][Core][WIP] Fix FetchFailedException when enable maxReqSizeShuffleToMem and KryoSerializer

[wangyum]

What changes were proposed in this pull request?

Spark cluster can reproduce, local can't:

1. Start a spark context with spark.reducer.maxReqSizeShuffleToMem=1K and spark.serializer=org.apache.spark.serializer.KryoSerializer:

$ spark-shell --conf spark.reducer.maxReqSizeShuffleToMem=1K --conf spark.serializer=org.apache.spark.serializer.KryoSerializer

2. A shuffle:

scala> sc.parallelize(0 until 3000000, 10).repartition(2001).count()

The error messages:

org.apache.spark.shuffle.MetadataFetchFailedException: Missing an output location for shuffle 0
        at org.apache.spark.MapOutputTracker$$anonfun$convertMapStatuses$2.apply(MapOutputTracker.scala:808)
        at org.apache.spark.MapOutputTracker$$anonfun$convertMapStatuses$2.apply(MapOutputTracker.scala:804)
        at scala.collection.TraversableLike$WithFilter$$anonfun$foreach$1.apply(TraversableLike.scala:733)
        at scala.collection.IndexedSeqOptimized$class.foreach(IndexedSeqOptimized.scala:33)
        at scala.collection.mutable.ArrayOps$ofRef.foreach(ArrayOps.scala:186)
        at scala.collection.TraversableLike$WithFilter.foreach(TraversableLike.scala:732)
        at org.apache.spark.MapOutputTracker$.convertMapStatuses(MapOutputTracker.scala:804)
        at org.apache.spark.MapOutputTrackerWorker.getMapSizesByExecutorId(MapOutputTracker.scala:618)
        at org.apache.spark.shuffle.BlockStoreShuffleReader.read(BlockStoreShuffleReader.scala:49)
        at org.apache.spark.rdd.ShuffledRDD.compute(ShuffledRDD.scala:105)
        at org.apache.spark.rdd.RDD.computeOrReadCheckpoint(RDD.scala:323)
        at org.apache.spark.rdd.RDD.iterator(RDD.scala:287)
        at org.apache.spark.rdd.CoalescedRDD$$anonfun$compute$1.apply(CoalescedRDD.scala:100)
        at org.apache.spark.rdd.CoalescedRDD$$anonfun$compute$1.apply(CoalescedRDD.scala:99)
        at scala.collection.Iterator$$anon$12.nextCur(Iterator.scala:434)
        at scala.collection.Iterator$$anon$12.hasNext(Iterator.scala:440)
        at scala.collection.Iterator$$anon$11.hasNext(Iterator.scala:408)
        at org.apache.spark.util.Utils$.getIteratorSize(Utils.scala:1802)
        at org.apache.spark.rdd.RDD$$anonfun$count$1.apply(RDD.scala:1159)
        at org.apache.spark.rdd.RDD$$anonfun$count$1.apply(RDD.scala:1159)
        at org.apache.spark.SparkContext$$anonfun$runJob$5.apply(SparkContext.scala:2065)
        at org.apache.spark.SparkContext$$anonfun$runJob$5.apply(SparkContext.scala:2065)
        at org.apache.spark.scheduler.ResultTask.runTask(ResultTask.scala:87)
        at org.apache.spark.scheduler.Task.run(Task.scala:108)
        at org.apache.spark.executor.Executor$TaskRunner.run(Executor.scala:341)
        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 PR fix this issue.

How was this patch tested?

Add later

[SparkQA]

Test build #78987 has finished for PR 18490 at commit 1c837b6.

• This patch fails Spark unit tests.
• This patch merges cleanly.
• This patch adds no public classes.

[zsxwing]

I don't get it. Could you point out which place serializes MapStatus using Kyro? NVM

[zsxwing]

I cannot reproduce this issue. Could you provide a unit test to reproduce this?

Anyway, I suggest using kryo.register(classOf[HighlyCompressedMapStatus], new KryoJavaSerializer()) to force Kryo using Java serialization so that we don't need to worry about if the Kryo default serializer works or not.