2012/1/17 Filip Rembiałkowski <plk.zuber@gmail.com>:
> postgres will still try to choose best execution plan. seq scan may simply be
> faster here. breaking point is somewhere near 50% selectivity.
The tipping point is usually far lower than that; in fact, it's more
often around 10%. Random IO is *very* expensive, as compared to
sequential IO (at least on spinning rust; SSDs are a different matter,
of course). It's usually vastly cheaper to read in an entire table
and filter the rows you want than to seek left and right (and then
left, left, and right again) to cherry-pick the pages you need.
rls
--
:wq