Details
-
Bug
-
Status: Resolved
-
Major
-
Resolution: Fixed
-
1.2.0
-
None
-
None
Description
There is a bug in the org.apache.hadoop.hive.ql.io.orc.RecordReaderImpl class which caused the bloom filter index saved in the ORC file not being used. The root cause is the bloomFilterIndices variable defined in the SargApplier class superseded the one defined in its parent class. Therefore, in the ReaderImpl.pickRowGroups()
protected boolean[] pickRowGroups() throws IOException { // if we don't have a sarg or indexes, we read everything if (sargApp == null) { return null; } readRowIndex(currentStripe, included, sargApp.sargColumns); return sargApp.pickRowGroups(stripes.get(currentStripe), indexes); }
The bloomFilterIndices populated by readRowIndex() is not picked up by sargApp object. One solution is to make SargApplier.bloomFilterIndices a reference to its parent counterpart.
18:46 $ diff src/java/org/apache/hadoop/hive/ql/io/orc/RecordReaderImpl.java src/java/org/apache/hadoop/hive/ql/io/orc/RecordReaderImpl.java.original
174d173
< bloomFilterIndices = new OrcProto.BloomFilterIndex[types.size()];
178c177
< sarg, options.getColumnNames(), strideRate, types, included.length, bloomFilterIndices);
---
> sarg, options.getColumnNames(), strideRate, types, included.length);
204a204
> bloomFilterIndices = new OrcProto.BloomFilterIndex[types.size()];
673c673
< List<OrcProto.Type> types, int includedCount, OrcProto.BloomFilterIndex[] bloomFilterIndices) {
---
> List<OrcProto.Type> types, int includedCount) {
677c677
< this.bloomFilterIndices = bloomFilterIndices;
---
> bloomFilterIndices = new OrcProto.BloomFilterIndex[types.size()];
Attachments
Attachments
Issue Links
- is related to
-
HIVE-9188 BloomFilter support in ORC
-
- Closed
-