[DRILL-1335] Merge Join incorrectly return unqualified rows when join key is null.

• Type: Bug
• Status: Resolved
• Priority: Major
• Resolution: Fixed
• Fix Version/s: 0.5.0
• Assignee: DrillCommitter
• Reporter: Jinfeng Ni
• Created: 25/Aug/14 19:51
• Updated: 09/Dec/14 17:50
• Resolved: 29/Aug/14 17:55

Description

Drill's merge join operator incorrectly returns unqualified rows, when join key is null value. For example,

alter session set `planner.enable_hashjoin` = false;

// Q1
select * from cp.`region.json` t1, cp.`region.json` t2 where t1.non_exist = t2.non_exist2 ;

Q1 should return 0 rows, since both left and right side of join comparison are NULL values. But Drill will return 12100 rows.

The cause of this problem is MergeJoin operator thinks a comparison between NULL and NULL is a qualified condition, and returns that pair of joining row.