[DRILL-4707] Conflicting columns names under case-insensitive policy lead to either memory leak or incorrect result

• Type: Bug
• Status: Closed
• Priority: Critical
• Resolution: Fixed
• Fix Version/s: 1.8.0
• Assignee: Jinfeng Ni
• Reporter: Jinfeng Ni
• Reviewer: Robert Hou
• Created: 03/Jun/16 18:55
• Updated: 27/Jul/16 22:09
• Resolved: 30/Jun/16 01:06

Description

On latest master branch:

select version, commit_id, commit_message from sys.version;
+-----------------+-------------------------------------------+---------------------------------------------------------------------------------+
|     version     |                 commit_id                 |                                 commit_message                                  |
+-----------------+-------------------------------------------+---------------------------------------------------------------------------------+
| 1.7.0-SNAPSHOT  | 3186217e5abe3c6c2c7e504cdb695567ff577e4c  | DRILL-4607: Add a split function that allows to separate string by a delimiter  |
+-----------------+-------------------------------------------+---------------------------------------------------------------------------------+

If a query has two conflicting column names under case-insensitive policy, Drill will either hit memory leak, or incorrect issue.

Q1.

select r_regionkey as XYZ, r_name as xyz FROM cp.`tpch/region.parquet`;
Error: SYSTEM ERROR: IllegalStateException: Memory was leaked by query. Memory leaked: (131072)
Allocator(op:0:0:1:Project) 1000000/131072/2490368/10000000000 (res/actual/peak/limit)


Fragment 0:0

Q2: return only one column in the result.

select n_nationkey as XYZ, n_regionkey as xyz FROM cp.`tpch/nation.parquet`;
+------+
| XYZ  |
+------+
| 0    |
| 1    |
| 1    |
| 1    |
| 4    |
| 0    |
| 3    |

The cause of the problem seems to be that the Project thinks the two incoming columns as identical (since Drill adopts case-insensitive for column names in execution).

The planner should make sure that the conflicting columns are resolved, since execution is name-based.

Issue Links

• links to: GitHub Pull Request #515