[DRILL-4852] COUNT(*) query against a large JSON table slower by 2x

• Type: Bug
• Status: Closed
• Priority: Critical
• Resolution: Fixed
• Affects Version/s: 1.8.0
• Fix Version/s: 1.8.0
• Component/s: Execution - Flow

Environment

4 node cluster CentOS

• Assignee: Arina Ielchiieva
• Reporter: Khurram Faraaz
• Created: 18/Aug/16 13:22
• Updated: 25/Aug/16 09:25
• Resolved: 25/Aug/16 09:08

Description

We have this manual test where it does a COUNT over 26 million JSON keys. From the results it looks like we have regressed and are slower by 2x on current 1.8.0 master 1.8.0-SNAPSHOT git commit ID : 57dc9f43
Query takes over 30 seconds to execute consistently over several runs. Note that since this is a single large JSON file there is just one fragment doing all the work.

0: jdbc:drill:schema=dfs.tmp> select count(*) from `twoKeyJsn.json`;
+-----------+
|  EXPR$0   |
+-----------+
| 26212355  |
+-----------+
1 row selected (29.001 seconds)

On Drill 1.2.0 the above query took 13.949 seconds

Issue Links

• links to: GitHub Pull Request #576