[jira] [Created] (HIVE-19557) stats: filters for dates are taking advantage of min/max values

["Zoltan Haindrich (JIRA)" <ji...@apache.org>]

Zoltan Haindrich created HIVE-19557:
---------------------------------------

             Summary: stats: filters for dates are taking advantage of min/max values
                 Key: HIVE-19557
                 URL: https://issues.apache.org/jira/browse/HIVE-19557
             Project: Hive
          Issue Type: Bug
          Components: Query Planning
            Reporter: Zoltan Haindrich
            Assignee: Zoltan Haindrich


in StatsRulesProcFactory 
[https://github.com/apache/hive/blob/ab189f54047bbf6beeeaf8d0dcfd5fbe92e465fb/ql/src/java/org/apache/hadoop/hive/ql/optimizer/stats/annotation/StatsRulesProcFactory.java#L754|dates are assumed to be an integer]; however this is currently not true - and the resulting exception is handled as a default case... for N/3


{code}
set hive.explain.user=true;

create table d1(d date);
--  tblproperties('transactional'='false');

insert into d1 values
        ('2010-10-01'),
        ('2010-10-02'),
        ('2010-10-03'),
        ('2010-10-04'),
        ('2010-10-05'),
        ('2010-10-06'),
        ('2010-10-07'),
        ('2010-10-08'),
        ('2010-10-09'),
        ('2010-10-10');

analyze table d1 compute statistics for columns;

desc formatted d1;
desc formatted d1 d;

explain
select 'stats: FIL ~0 read',count(1) from d1 where d < '2010-03-01';

explain
select 'stats: FIL estimate some read',count(1) from d1 where d < '2010-10-03';

explain
select 'stats: FIL estimate all read',count(1) from d1 where d < '2010-11-03';
{code}



--
This message was sent by Atlassian JIRA
(v7.6.3#76005)