[HIVE-12435] SELECT COUNT(CASE WHEN...) GROUPBY returns 1 for 'NULL' in a case of ORC and vectorization is enabled.

• Type: Bug
• Status: Closed
• Priority: Critical
• Resolution: Fixed
• Affects Version/s: 2.0.0
• Fix Version/s: 2.0.0
• Component/s: Vectorization
• Target Version/s: 2.0.0
• Assignee: Matt McCline
• Reporter: Takahiko Saito
• Created: 17/Nov/15 16:51
• Updated: 02/Jun/16 10:17
• Resolved: 18/Dec/15 00:59

Description

Run the following query:

create table count_case_groupby (key string, bool boolean) STORED AS orc;
insert into table count_case_groupby values ('key1', true),('key2', false),('key3', NULL),('key4', false),('key5',NULL);

The table contains the following:

key1	true
key2	false
key3	NULL
key4	false
key5	NULL

The below query returns:

SELECT key, COUNT(CASE WHEN bool THEN 1 WHEN NOT bool THEN 0 ELSE NULL END) AS cnt_bool0_ok FROM count_case_groupby GROUP BY key;
key1	1
key2	1
key3	1
key4	1
key5	1

while it expects the following results:

key1	1
key2	1
key3	0
key4	1
key5	0

The query works with hive ver 1.2. Also it works when a table is not orc format.
Also even if it's an orc table, when vectorization is disabled, the query works.

Issue Links

• is related to: HIVE-12209 Vectorized simple CASE expressions with nulls