Description
CompressedRDFReader now buffers ZipEntrys in java heap to read in zipped file. Instead, it should stream the reading process, which is desired and preferrable. CompressedAggregateReader and CompressedDelimitedTextReader has done the same.
In an attempt to do the above job, I found that there is some class in CompressedRDFReader that closes the ZipInputStream in middle of a job. It is related to a java bug JDK-6539065. For CompressedAggregateReader, because I know it's XMLStreamReader that closed the ZipInputStream in middle of a job, we can bypass the issue by overload ZipInputStream to disable the behavior. However, for CompressedRDFReader, I didn't figure out who closed ZipInputStream.
You can refer to bug:39688 and bug:25918 for more info.
Please investigate. Thanks.