Re: suggestion to improve planer

On Thu, 2009-09-03 at 10:35 +0200, Ľubomír Varga wrote:
> Hi.
>
> I hope, that this is right mailing list.
>
> SELECT date, value FROM t_event
> WHERE t_event.id in (SELECT id FROM t_event
> WHERE date < '2009-08-25'
> ORDER BY date DESC LIMIT 1)
> ORDER BY date;
> cost 6.4
>
> SELECT date, value FROM t_event
> WHERE t_event.id = (SELECT id FROM t_event
> WHERE date < '2009-08-25'
> ORDER BY date DESC LIMIT 1)
> ORDER BY date;
> cost 6.36..6.37
>
>
> Why that two query dont have equal cost? If it is not problem, try add some
> planer code to recognize that sublesect HAVE TO return just one row (limit 1)
> and in plan could be used filter/index scan instead of hash aggregate.

Well, there is always a tradeoff between more planner analysis and more
complicated and slow planning. Seeing that the cost estimates are close
enough for practical purposes, it doesn't seem worthwhile to fix
anything here.

> I have
> also some complex query examples where cost difference is more visible.

Having real examples where a change might actually improve runtime is
always more interesting than an academic exercise like the above.