Re: get_actual_variable_range vs idx_scan/idx_tup_fetch, again

On Mon, Mar 5, 2018 at 5:08 PM, Tom Lane <tgl(at)sss(dot)pgh(dot)pa(dot)us> wrote:

> Marko Tiikkaja <marko(at)joh(dot)to> writes:
> > So I'm in the same pickle again. According to pg_stat_user_indexes an
> > index is being used all the time. However, it's only being used by
> > mergejoinscansel() to compare these two plans:
>
> If it's not being used otherwise, could you drop it?
>

Yes. I want to drop it, as I think it's useless, but it's hard to be 100%
sure.

> > I think it would be really important to have a way to turn off
> > get_actual_variable_range() for a specific index during runtime. Would
> a C
> > level hook be acceptable for this?
>
> You haven't really made a case for why you (or anyone else) should care.
> As long as the planner makes the right choice, having investigated a wrong
> choice doesn't seem like a bug to me.
>

Because I'm certain the planner would make the right choice even without
the index, and I want it gone.

.m