Re: Query planner skipping index depending on DISTINCT parameter order (2)

On 09/25/2017 12:43 PM, Greg Stark wrote:
> On 17 September 2017 at 18:15, Tom Lane <tgl(at)sss(dot)pgh(dot)pa(dot)us> wrote:
>
>> For plain DISTINCT, yeah we could consider other orderings ... but
>> we're rather unlikely to find an index that matches all the output
>> columns, regardless of what order they're in. So it's just not that
>> exciting.
>
> I don't follow this part. Are you saying a simple "select distinct a,b
> from table" is unlikely to find a matching index on <b,a>? Don't we
> already do exactly this for "select a,b from table group by a,b"?

FWIW we don't do that:

create table t (a int, b int);
insert into t select mod(i, 1000), mod(i, 1000)
from generate_series(1,1000000) S(i);
create index on t (a,b);
vacuum analyze t;

explain (costs off) select a, b, count(*) from t group by a , b;
QUERY PLAN
--------------------------------------------
GroupAggregate
Group Key: a, b
-> Index Only Scan using t_a_b_idx on t
(3 rows)

explain (costs off) select a, b, count(*) from t group by b , a;
QUERY PLAN
---------------------------
GroupAggregate
Group Key: b, a
-> Sort
Sort Key: b, a
-> Seq Scan on t
(5 rows)

But yeah, fixing this is probably more worthwhile than the original
DISTINCT ON issue ...

regards

--
Tomas Vondra http://www.2ndQuadrant.com
PostgreSQL Development, 24x7 Support, Remote DBA, Training & Services