Anton Gozhiy created DRILL-6856:
-----------------------------------
Summary: Wrong result returned if the query filters a boolean
column with both "is true" and "is null" conditions
Key: DRILL-6856
URL: https://issues.apache.org/jira/browse/DRILL-6856
Project: Apache Drill
Issue Type: Bug
Affects Versions: 1.15.0
Reporter: Anton Gozhiy
Attachments: 0_0_0.parquet
*Data:*
A parquet file with a boolean column that contains null values.
An example is attached.
*Query:*
{code:sql}
select bool_col from dfs.tmp.`Test_data` where bool_col is true or bool_col is
null
{code}
*Result:*
{noformat}
null
null
{noformat}
*Plan:*
{noformat}
00-00 Screen : rowType = RecordType(ANY bool_col): rowcount = 3.75,
cumulative cost = {37.875 rows, 97.875 cpu, 15.0 io, 0.0 network, 0.0 memory},
id = 1980
00-01 Project(bool_col=[$0]) : rowType = RecordType(ANY bool_col):
rowcount = 3.75, cumulative cost = {37.5 rows, 97.5 cpu, 15.0 io, 0.0 network,
0.0 memory}, id = 1979
00-02 SelectionVectorRemover : rowType = RecordType(ANY bool_col):
rowcount = 3.75, cumulative cost = {33.75 rows, 93.75 cpu, 15.0 io, 0.0
network, 0.0 memory}, id = 1978
00-03 Filter(condition=[IS NULL($0)]) : rowType = RecordType(ANY
bool_col): rowcount = 3.75, cumulative cost = {30.0 rows, 90.0 cpu, 15.0 io,
0.0 network, 0.0 memory}, id = 1977
00-04 Scan(table=[[dfs, tmp, Test_data]],
groupscan=[ParquetGroupScan [entries=[ReadEntryWithPath
[path=maprfs:///tmp/Test_data]], selectionRoot=maprfs:/tmp/Test_data,
numFiles=1, numRowGroups=1, usedMetadataFile=false, columns=[`bool_col`]]]) :
rowType = RecordType(ANY bool_col): rowcount = 15.0, cumulative cost = {15.0
rows, 15.0 cpu, 15.0 io, 0.0 network, 0.0 memory}, id = 1976
{noformat}
*Notes:*
- "true" values were not included in the result though they should have.
- Result is correct if use "bool_col = true" instead of "is true"
- In the plan you can see that "is true" condition is absent in the Filter
operator
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
