[DRILL-3537] Empty Json file can potentially result into wrong results

• Type: Bug
• Status: Closed
• Priority: Critical
• Resolution: Fixed
• Fix Version/s: 1.2.0
• Component/s: Execution - Relational Operators, Storage - JSON
• Assignee: Chun Chang
• Reporter: Sean Hsuan-Yi Chu
• Created: 21/Jul/15 23:49
• Updated: 16/Sep/15 00:25
• Resolved: 27/Jul/15 22:17

Description

In the directory, we have two files. One has some data and the other one is empty. A query as below:

select * from dfs.`directory`;

will produce different results according to the order of the files being read (The default order is in the alphabetic order of the filenames). To give a more concrete example, the non-empty json has data:

{
      a:1
}

By naming the files, you can control the orders. If the empty file is read in firstly, the result is

+-------+----+
|   *   | a  |
+-------+----+
| null  | 1  |
+-------+----+

If the opposite order takes place, the result is

+----+
| a  |
+----+
| 1  |
| 2  |
+----+