[HIVE-6711] ORC maps uses getMapSize() from MapOI which is unreliable

• Type: Bug
• Status: Closed
• Priority: Major
• Resolution: Fixed
• Affects Version/s: 0.11.0, 0.12.0, 0.13.0
• Fix Version/s: 0.13.0, 0.14.0
• Labels: orcfile
• Assignee: Prasanth Jayachandran
• Reporter: Prasanth Jayachandran
• Created: 20/Mar/14 18:39
• Updated: 13/Nov/14 19:44
• Resolved: 24/Mar/14 00:23

Description

HIVE-6707 had issues with map size. getMapSize() of LazyMap and LazyBinaryMap does not deserialize the keys and count the number of unique keys. Since getMapSize() may return non-distinct count of keys, the length of maps stored using ORC's map tree writer will not be in sync with actual map size. As a result of this RLE reader will try to read beyond the disk range expecting more map entries and will throw exception.

Stack trace will look like:

Caused by: java.io.EOFException: Read past end of RLE integer from compressed stream Stream for column 2 kind DATA position: 22059699 length: 22059699 range: 0 offset: 22359014 limit: 22359014 range 0 = 0 to 22059699 uncompressed: 53370 to 53370
        at org.apache.hadoop.hive.ql.io.orc.RunLengthIntegerReaderV2.readValues(RunLengthIntegerReaderV2.java:54)
        at org.apache.hadoop.hive.ql.io.orc.RunLengthIntegerReaderV2.next(RunLengthIntegerReaderV2.java:301)
        at org.apache.hadoop.hive.ql.io.orc.RecordReaderImpl$StringDictionaryTreeReader.next(RecordReaderImpl.java:1572)
        at org.apache.hadoop.hive.ql.io.orc.RecordReaderImpl$StringTreeReader.next(RecordReaderImpl.java:1330)
        at org.apache.hadoop.hive.ql.io.orc.RecordReaderImpl$MapTreeReader.next(RecordReaderImpl.java:2041)
        at org.apache.hadoop.hive.ql.io.orc.RecordReaderImpl$StructTreeReader.next(RecordReaderImpl.java:1772)
        at org.apache.hadoop.hive.ql.io.orc.RecordReaderImpl.next(RecordReaderImpl.java:2963)
        at org.apache.hadoop.hive.ql.io.orc.OrcInputFormat$OrcRecordReader.next(OrcInputFormat.java:121)