As a workaround, can you set the number of partitions higher in the
sc.textFile method ?
Cheers
On Mon, Oct 5, 2015 at 3:31 PM, Jegan <jega...@gmail.com> wrote:
> Hi All,
>
> I am facing the below exception when the size of the file being read in a
> partition is above 2GB. This is apparently because Java's limitation on
> memory mapped files. It supports mapping only 2GB files.
>
> Caused by: java.lang.IllegalArgumentException: Size exceeds
> Integer.MAX_VALUE
> at sun.nio.ch.FileChannelImpl.map(FileChannelImpl.java:836)
> at
> org.apache.spark.storage.DiskStore$$anonfun$getBytes$2.apply(DiskStore.scala:125)
> at
> org.apache.spark.storage.DiskStore$$anonfun$getBytes$2.apply(DiskStore.scala:113)
> at org.apache.spark.util.Utils$.tryWithSafeFinally(Utils.scala:1207)
> at org.apache.spark.storage.DiskStore.getBytes(DiskStore.scala:127)
> at org.apache.spark.storage.DiskStore.getBytes(DiskStore.scala:134)
> at org.apache.spark.storage.DiskStore.putIterator(DiskStore.scala:102)
> at org.apache.spark.storage.BlockManager.doPut(BlockManager.scala:791)
> at
> org.apache.spark.storage.BlockManager.putIterator(BlockManager.scala:638)
> at
> org.apache.spark.CacheManager.putInBlockManager(CacheManager.scala:153)
> at org.apache.spark.CacheManager.getOrCompute(CacheManager.scala:78)
> at org.apache.spark.rdd.RDD.iterator(RDD.scala:262)
> at
> org.apache.spark.rdd.MapPartitionsRDD.compute(MapPartitionsRDD.scala:38)
> at org.apache.spark.rdd.RDD.computeOrReadCheckpoint(RDD.scala:297)
> at org.apache.spark.rdd.RDD.iterator(RDD.scala:264)
> at
> org.apache.spark.rdd.MapPartitionsRDD.compute(MapPartitionsRDD.scala:38)
> at org.apache.spark.rdd.RDD.computeOrReadCheckpoint(RDD.scala:297)
> at org.apache.spark.rdd.RDD.iterator(RDD.scala:264)
> at org.apache.spark.scheduler.ResultTask.runTask(ResultTask.scala:66)
> at org.apache.spark.scheduler.Task.run(Task.scala:88)
> at
> org.apache.spark.executor.Executor$TaskRunner.run(Executor.scala:214)
> at
> java.util.concurrent.ThreadPoolExecutor.runWorker(ThreadPoolExecutor.java:1142)
> at
> java.util.concurrent.ThreadPoolExecutor$Worker.run(ThreadPoolExecutor.java:617)
> at java.lang.Thread.run(Thread.java:745)
>
> My use case is to read the files from S3 and do some processing. I am
> caching the data like below in order to avoid SocketTimeoutExceptions from
> another library I am using for the processing.
>
> val rdd1 = sc.textFile("*******").coalesce(1000)
> rdd1.persist(DISK_ONLY_2) // replication factor 2
> rdd1.foreachPartition { iter => } // one pass over the data to download
>
> The 3rd line fails with the above error when a partition contains a file
> of size more than 2GB file.
>
> Do you think this needs to be fixed in Spark? One idea may be is to use a
> wrapper class (something called BigByteBuffer) which keeps an array of
> ByteBuffers and keeps the index of the current buffer being read etc. Below
> is the modified DiskStore.scala.
>
> private def getBytes(file: File, offset: Long, length: Long):
> Option[ByteBuffer] = {
> val channel = new RandomAccessFile(file, "r").getChannel
> Utils.tryWithSafeFinally {
> // For small files, directly read rather than memory map
> if (length < minMemoryMapBytes) {
> // Map small file in Memory
> } else {
> // TODO Create a BigByteBuffer
>
> }
> } {
> channel.close()
> }
> }
>
> class BigByteBuffer extends ByteBuffer {
> val buffers: Array[ByteBuffer]
> var currentIndex = 0
>
> ... // Other methods
> }
>
> Please let me know if there is any other work-around for the same. Thanks for
> your time.
>
> Regards,
> Jegan
>
