On Tue, Feb 22, 2022 at 04:12:15PM -0500, Greg Stark wrote:
> I've been playing with an idea I had a while back. Basically that it
> would be useful to have some "noop" operators that are used purely to
> influence the planner.
>
> For context I've suggested in the past that there are two categories of hints:
>
> 1 Hints that override the planner's decisions with explicit planning
> decisions. These we don't like for a variety of reasons, notably
> because users don't have very good knowledge of all the plan types
> available and new plan types come out in the future.
>
> 2 Hints that just influence the estimates that the planner uses to
> make its decisions. These seem more acceptable as it amounts to
> admitting the users know the distribution of their data and the
> behaviour of their functions better than the statistics. Also, there
> are plenty of cases where the statistics are really just wild guesses.
> They still allow the planner to make decisions about what join orders
> or types and so on given the updated data.
#2 is an interesting distinction.
--
Bruce Momjian <bruce@momjian.us> https://momjian.us
EDB https://enterprisedb.com
If only the physical world exists, free will is an illusion.
