Re: reducing random_page_cost from 4 to 2 to force index scan

> Well, all of that stuff sounds impractically expensive to me... but I
> just work here.

I'll point out that the simple version, which just checks for hot tables
and indexes, would improve estimates greatly and be a LOT less
complicated than these proposals.  Certainly having some form of
block-based or range-based stats would be better, but it also sounds
hard enough to maybe never get done.

Having user-accessible "hot" stats would also be useful to DBAs.

--
Josh Berkus
PostgreSQL Experts Inc.
http://pgexperts.com