Vineet Garg created HIVE-21387:
----------------------------------
Summary: Wrong result for UNION query with GROUP BY consisting of
PK columns
Key: HIVE-21387
URL: https://issues.apache.org/jira/browse/HIVE-21387
Project: Hive
Issue Type: Bug
Components: Query Planning
Affects Versions: 4.0.0
Reporter: Vineet Garg
Assignee: Vineet Garg
*Reproducer*
{code:sql}
create table t1(i int primary key disable rely, j int);
insert into t1 values(1,100),(2,200);
create table t2(i int primary key disable rely, j int);
insert into t2 values(2,1000),(4,500);
select i from (select i, j from t1 union all select i,j from t2) subq group by
i,j;
{code}
*Expected Result*
{noformat}
2
2
4
1
{noformat}
*Actual Result*
{noformat}
1
2
4
{noformat}
*CBO Plan*
{code:sql}
HiveAggregate(group=[{0}])
HiveProject(i=[$0], j=[$1])
HiveUnion(all=[true])
HiveProject(i=[$0], j=[$1])
HiveTableScan(table=[[default, t1]], table:alias=[t1])
HiveProject(i=[$0], j=[$1])
HiveTableScan(table=[[default, t2]], table:alias=[t2])
{code}
This is due to Group by reduction logic reducing keys when it shouldn't.
Because of UNION relative cardinality of the group by keys are changed (they
are not PK/UNIQUE anymore). Therefore we shouldn't be trying to reduce group by
keys at all.
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
