[HIVE-24926] Correlated fields not detected if subquery has where and group clauses.

• Type: Bug
• Status: Open
• Priority: Major
• Resolution: Unresolved
• Component/s: HiveServer2
• Assignee: Unassigned
• Reporter: Steve Carlin
• Created: 23/Mar/21 15:18
• Updated: 23/Mar/21 15:18

Description

The following query is failing to compile:

 

SELECT id
FROM alltypestiny t1
WHERE EXISTS
  (SELECT 1
   FROM alltypestiny t2
   WHERE t1.id = t2.id
   GROUP BY t2.id
   HAVING count(1) = 1);

 

The reason it fails is because there is a group by clause and because t1.id is used in the where clause.

The failure can be traced to code within HiveFilter.traverseFilter, which doesn't find the filter because it returns when it sees the HiveAggregate.