Robert Haas <robertmhaas@gmail.com> writes:
> ... I guess I'm just saying I'd think really, really hard
> before abandoning the idea of periodic sampling. You have to get an
> awful lot of benefit out of those cross-column stats to make it worth
> paying a run-time cost for them.
I've been trying to not discourage Tomas from blue-sky speculation,
but I have to say I agree with Robert that the chance of any usable
result from this approach is very very small. Feel free to do some
experimentation to prove us wrong --- but don't put a lot of effort
into it before that.
regards, tom lane