Re: agg/order-by question

On Sat, Jul 12, 2003 at 00:39:06 -0700,
Sailesh Krishnamurthy <sailesh(at)cs(dot)berkeley(dot)edu> wrote:
>
> Consider the explain for the following queries ..
>
> sample=# explain select a, count(*) from foo group by a order by a;
> QUERY PLAN
> -------------------------------------------------------------------------
> Aggregate (cost=69.83..77.33 rows=100 width=4)
> -> Group (cost=69.83..74.83 rows=1000 width=4)
> -> Sort (cost=69.83..72.33 rows=1000 width=4)
> Sort Key: a
> -> Seq Scan on foo (cost=0.00..20.00 rows=1000 width=4)
> (5 rows)
>
> sample=# explain select a, count(*) from foo group by a order by a desc;
> QUERY PLAN
> -------------------------------------------------------------------------------
> Sort (cost=80.65..80.90 rows=100 width=4)
> Sort Key: a
> -> Aggregate (cost=69.83..77.33 rows=100 width=4)
> -> Group (cost=69.83..74.83 rows=1000 width=4)
> -> Sort (cost=69.83..72.33 rows=1000 width=4)
> Sort Key: a
> -> Seq Scan on foo (cost=0.00..20.00 rows=1000 width=4)
> (7 rows)
>
>
> In the first case pgsql doesn't have a Sort on top because the Sort
> below the Group produces the right "interesting order" (using the
> System-R term). In the second case however, since the order-by clause
> demands "desc" there is a Sort tagged on on top.
>
> Now, instead of doing this, isn't it better to just have a similar
> plan as in the first case, but just change the lower Sort to be
> descending ? It doesn't affect the Group and the Agg in any way ..

You might try this in 7.4. I am pretty sure a change was made a couple
of weeks ago to let group by work with either sort order. Also hash
aggragates have been available for quite a while in 7.4. This is a better
plan when there are only a small number of distinct values.