[DRILL-1362] Count(nullable-column) is incorrectly pushed into group scan operator

• Type: Bug
• Status: Resolved
• Priority: Major
• Resolution: Fixed
• Affects Version/s: 0.4.0
• Fix Version/s: 0.7.0
• Component/s: Execution - Relational Operators
• Assignee: Unassigned
• Reporter: Aman Sinha
• Created: 29/Aug/14 17:08
• Updated: 17/Nov/14 01:11
• Resolved: 17/Nov/14 01:11

Description

The following query on TPC-DS table web_returns produces wrong result because the aggregate count(wr_return_quantity) gets pushed into the parquet group scan operator even though wr_return_quantity is nullable and apparently the parquet metadata does not have stats on nullable column.

0: jdbc:drill:zk=local> select count(wr_return_quantity) from web_returns;
------------

EXPR$0

------------

71763

------------

0: jdbc:drill:zk=local> explain plan for select count(wr_return_quantity) from web_returns;

+------------+------------+
|    text    |    json    |
+------------+------------+
| 00-00    Screen
00-01      Project(EXPR$0=[$0])
00-02        Scan(groupscan=[org.apache.drill.exec.store.pojo.PojoRecordReader@e4acaad])

For reference, here are the correct results:
tpcds=# select count(wr_return_quantity) from web_returns;
count
-------
68616
(1 row)

tpcds=# select count from web_returns;
count
-------
71763
(1 row)

Issue Links

• duplicates: DRILL-1434 count of a nullable column in tpcds gives incorrect results