[DRILL-4513] Query returns null when star in project and filter over non null values

• Type: Bug
• Status: Open
• Priority: Critical
• Resolution: Unresolved
• Affects Version/s: 1.4.0
• Component/s: Storage - JSON

Environment

4 node cluster CentOS

• Assignee: Unassigned
• Reporter: Khurram Faraaz
• Created: 16/Mar/16 07:21
• Updated: 16/Mar/16 11:09

Description

When there is a star in project and there are many nulls followed by a non null value in the column c1 in JSON file, query does not return correct results.

json file with 4096 records that have a field

{c1: null}

followed by the 4097th record that has

{c1: "Hello World"}

Drill version
http://yum.qa.lab/drill-opensource.ebf/mapr-drill-1.4.0.201603151147-1.noarch.rpm

0: jdbc:drill:schema=dfs.tmp> select * from sys.version;
+---------+-----------+----------------+-------------+-------------+------------+
| version | commit_id | commit_message | commit_time | build_email | build_time |
+---------+-----------+----------------+-------------+-------------+------------+
| 1.4.0 | 99ec15919cd73f5e7b976f6193af45c0e46ffd7a | DRILL-4479: For empty fields under all_text_mode enabled (a) use varchar for the default columns and (b) ensure we create fields corresponding to all columns. | 14.03.2016 @ 20:52:13 UTC | Unknown | 15.03.2016 @ 18:47:50 UTC |
+---------+-----------+----------------+-------------+-------------+------------+
1 row selected (0.375 seconds)
0: jdbc:drill:schema=dfs.tmp> alter system set `store.json.all_text_mode`=true;
+-------+------------------------------------+
|  ok   |              summary               |
+-------+------------------------------------+
| true  | store.json.all_text_mode updated.  |
+-------+------------------------------------+
1 row selected (0.136 seconds)
0: jdbc:drill:schema=dfs.tmp> select * from `tooManyNulls.json` t where t.c1 = 'Hello World';
Error: SYSTEM ERROR: NumberFormatException: Hello World

Fragment 0:0

[Error Id: f9022f22-12cd-46c7-b471-470a50506ba2 on centos-04.qa.lab:31010] (state=,code=0)
0: jdbc:drill:schema=dfs.tmp> select * from `tooManyNulls.json` t where t.c1 IS NOT NULL;
+-------+
|   *   |
+-------+
| null  |
+-------+
1 row selected (0.239 seconds)
0: jdbc:drill:schema=dfs.tmp> select t.c1 from `tooManyNulls.json` t where t.c1 = 'Hello World';
+--------------+
|      c1      |
+--------------+
| Hello World  |
+--------------+
1 row selected (0.282 seconds)
0: jdbc:drill:schema=dfs.tmp> select t.c1 from `tooManyNulls.json` t where t.c1 IS NOT NULL;
+--------------+
|      c1      |
+--------------+
| Hello World  |
+--------------+
1 row selected (0.259 seconds)

Another interesting observation to note, the below query returns null. However, when the same query is used as a sub-query, we see the query returns correct results.

0: jdbc:drill:schema=dfs.tmp> select * from `tooManyNulls.json` t where t.c1 IS NOT NULL;
+-------+
|   *   |
+-------+
| null  |
+-------+
1 row selected (0.203 seconds)
0: jdbc:drill:schema=dfs.tmp> select tmp.c1 from (select * from `tooManyNulls.json` t where t.c1 IS NOT NULL) tmp;
+--------------+
|      c1      |
+--------------+
| Hello World  |
+--------------+
1 row selected (0.25 seconds)