On 22/1/2022 01:34, Tomas Vondra wrote:
> The other thing we could do is reduce the coefficient gradually - so
> it'd be 1.5 for the first pathkey, then 1.25 for the next one, and so
> on. But it seems somewhat arbitrary (I certainly don't have some sound
> theoretical justification ...).
I think, it hasn't a reason to increase complexity without any theory at
the bottom. Simple solution allows to realize problems much faster, if
they arise.
> ... I've skipped the path_save
> removal in planner.c, because that seems incorrect - if there are
> multiple pathkeys, we must start with the original path, not the
> modified one we built in the last iteration. Or am I missing something
You are right, I misunderstood the idea of path_save variable.
--
regards,
Andrey Lepikhov
Postgres Professional