On Mon, Apr 16, 2012 at 4:58 PM, Noah Misch <noah(at)leadboat(dot)com> wrote:
> The size hint I chose is fairly arbitrary. Any suggestions for principled
> alternatives?
Based on your test results, it doesn't seem like it matters very much
what you put in there, so I'm inclined to think that num_mcelem is
fine. I thought about maybe allowing for a little slop, like
num_mcelem * 10, but maybe the way you did it is better. It's
possible that people will set ridiculously overblown stats targets on
some columns, and that's certainly going to cost something no matter
what we do, but there's no point in making that worse than it has to
be without some clear reason for doing so.
--
Robert Haas
EnterpriseDB: http://www.enterprisedb.com
The Enterprise PostgreSQL Company