[DRILL-4376] Wrong results when doing a count(*) on part of directories with metadata cache

• Type: Bug
• Status: Closed
• Priority: Critical
• Resolution: Fixed
• Affects Version/s: 1.4.0
• Fix Version/s: 1.7.0
• Component/s: Metadata
• Assignee: Abdel Hakim Deneche
• Reporter: Abdel Hakim Deneche
• Reviewer: Rahul Kumar Challapalli
• Created: 09/Feb/16 00:15
• Updated: 18/May/16 20:19
• Resolved: 16/Mar/16 17:11

Description

First create some parquet tables in multiple subfolders:

create table dfs.tmp.`test/201501` as select employee_id, full_name from cp.`employee.json` limit 2;
create table dfs.tmp.`test/201502` as select employee_id, full_name from cp.`employee.json` limit 2;
create table dfs.tmp.`test/201601` as select employee_id, full_name from cp.`employee.json` limit 2;
create table dfs.tmp.`test/201602` as select employee_id, full_name from cp.`employee.json` limit 2;

Running the following query gives the expected count:

select count(*) from dfs.tmp.`test/20160*`;
+---------+
| EXPR$0  |
+---------+
| 4       |
+---------+

But once you create the metadata cache files, the query no longer returns the correct results:

refresh table metadata dfs.tmp.`test`;
select count(*) from dfs.tmp.`test/20160*`;
+---------+
| EXPR$0  |
+---------+
| 2       |
+---------+

Issue Links

• relates to: DRILL-4484 NPE when querying empty directory