Github user squito commented on a diff in the pull request:

    https://github.com/apache/spark/pull/6423#discussion_r32050912
  
    --- Diff: 
core/src/main/scala/org/apache/spark/shuffle/hash/HashShuffleReader.scala ---
    @@ -33,17 +34,52 @@ private[spark] class HashShuffleReader[K, C](
         "Hash shuffle currently only supports fetching one partition")
     
       private val dep = handle.dependency
    +  private val blockManager = SparkEnv.get.blockManager
     
       /** Read the combined key-values for this reduce task */
       override def read(): Iterator[Product2[K, C]] = {
    +    val blockStreams = BlockStoreShuffleFetcher.fetchBlockStreams(
    +      handle.shuffleId, startPartition, context)
    +
    +    // Wrap the streams for compression based on configuration
    +    val wrappedStreams = blockStreams.map { case (blockId, inputStream) =>
    +      blockManager.wrapForCompression(blockId, inputStream)
    +    }
    +
         val ser = Serializer.getSerializer(dep.serializer)
    -    val iter = BlockStoreShuffleFetcher.fetch(handle.shuffleId, 
startPartition, context, ser)
    +    val serializerInstance = ser.newInstance()
    +
    +    // Create a key/value iterator for each stream
    +    val recordIterator = wrappedStreams.flatMap { wrappedStream =>
    +      val kvIter = 
serializerInstance.deserializeStream(wrappedStream).asKeyValueIterator
    +      CompletionIterator[(Any, Any), Iterator[(Any, Any)]](kvIter, {
    +        // Close the stream once all the records have been read from it
    +        wrappedStream.close()
    --- End diff --
    
    and can you also expand the comment a bit here that the 
`CompletionIterator` necessary to free underlying ManagedBuffers as soon as 
possible (but there is already a `TaskCompletionListener` to guarantee the 
buffer is released even if there is an exception).


---
If your project is set up for it, you can reply to this email and have your
reply appear on GitHub as well. If your project does not have this feature
enabled and wishes so, or if the feature is enabled but not working, please
contact infrastructure at [email protected] or file a JIRA ticket
with INFRA.
---

---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]

Reply via email to