Frederic Jolliton kirjutas N, 24.04.2003 kell 19:33:
> > Frederic Jolliton <fred-pg(at)jolliton(dot)com> writes:
> >>> To "emulate" a parametred view, I created a function as follow:
> >>> CREATE FUNCTION get_info (integer) RETURNS SETOF type_get_info
> >>> AS '...' <- here the query show below, where 'LIMIT $1' is used instead of 'LIMIT 10'
> >>> LANGUAGE sql;
> >> So, the query in the function is not using index but the exact same
> >> query alone does !
> Tom Lane <tgl(at)sss(dot)pgh(dot)pa(dot)us> writes:
> > But it's not the same query, is it? With "LIMIT $1" the planner can't
> > know what the limit value is exactly, so it has to generate a plan that
> > won't be too unreasonable for either a small or a large limit.
> Ok. So the query is optimized once and not each time.. I understand
> But, since I "know" better that PostgreSQL that query must use index
> in most of case, can I force in some manner the function when
> declaring it to take this in account ?
You could define two functions - one for small sets with constant LIMITs
(maybe 50) in UNION parts, and another with $1. Then use accordingly.
> I suppose (not tested) that
> setting enable_seqscan just before will probably do it, but what about
> dump/restore of the database when recreating the function and keep
> this "fix" automatically ?
In response to
pgsql-performance by date
|Next:||From: Jim C. Nasby||Date: 2003-04-25 14:38:01|
|Subject: Re: More tablescanning fun|
|Previous:||From: Tom Lane||Date: 2003-04-25 05:23:10|
|Subject: Re: More tablescanning fun |