Francisco Reyes <lists@natserv.com> writes:
> With explain analyze I have with sequential scan on
> Sort (cost=382.01..382.15 rows=56 width=196)
> (actual time=64.346..64.469 rows=24 loops=1)
> And with seqscan off
> Sort (cost=490.82..490.96 rows=56 width=196)
> (actual time=56.668..56.789 rows=24 loops=1)
> So I guess that for this particular query I am better off setting the
> seqscan off.
For that kind of margin, you'd be a fool to do any such thing.
You might want to look at making some adjustment to random_page_cost
to bring the estimated costs in line with reality (though I'd counsel
taking more than one example into account while you tweak it). But
setting seqscan off as a production setting is just a recipe for
shooting yourself in the foot.
regards, tom lane
