Details
-
Improvement
-
Status: Resolved
-
Major
-
Resolution: Incomplete
-
None
-
None
Description
For context, see the PR comments at https://github.com/apache/spark/pull/6423#discussion_r31370774
The current HashShuffleReader closes the input stream when all records have been read by using a NextIterator. However, we want the ShuffleBlockFetcherIterator to be more defensive about closing files and releasing memory to protect against other ShuffleReaders which may not.