[DRILL-7260] Error on clause that should not be evaluated

• Type: Bug
• Status: Open
• Priority: Major
• Resolution: Unresolved
• Affects Version/s: 1.16.0
• Component/s: Query Planning & Optimization
• Assignee: Unassigned
• Reporter: benj
• Created: 15/May/19 14:58
• Updated: 15/May/19 14:58

Description

Some requests failed when evaluate condition which should not be evaluated.

Example:

SELECT * FROM (
 SELECT * FROM (
  SELECT '1d28' a ) 
 WHERE regexp_matches(a,'^[0-9]*$')) /* Keep only row on INT format */
WHERE CAST(a AS INT) < 2000;         /* This CAST should't failed due */ 
                                     /* to filtering in the inner queries */
=>
Error: SYSTEM ERROR: NumberFormatException: 1d28

Note that the problem doesn't appear in all situation.

Examples :

/* Just adding another row and it's work ! */
SELECT * FROM (
 SELECT * FROM (
  SELECT '1d28' a UNION SELECT '2a2')
 WHERE regexp_matches(a,'^[0-9]*$'))
WHERE CAST(a AS INT) < 2000;
=>
+---+
| a |
+---+
+---+

// MyFile.csvh
a,b
1,2
a1a,3
// !MyFile.csvh 

SELECT * FROM (
 SELECT * FROM ....`MyFile.csvh ` WHERE regexp_matches(a,'^[0-9]*$')) AS x WHERE CAST(a AS INT) < 200;
=>
+---+---+
| a | b |
+---+---+
| 1 | 2 |
+---+---+

In this case, there is no error - the cast is never evaluated (as expected) for the row with a='a1a'.