Re: BRIN indexes for MAX, MIN, ORDER BY?

Alvaro Herrera <alvherre(at)2ndquadrant(dot)com> writes:
> Gavin Wahl wrote:
>> It seems trivial to accelerate a MAX or MIN query with a BRIN index. You
>> just find the page range with the largest/smallest value, and then only
>> scan that one. Would that be hard to implement? I'm interested in working
>> on it if someone can give me some pointers.

> I think this means you need to represent this operation as a specific
> Path in some way. See build_minmax_path() and its callers in planagg;
> you probably need to tweak preprocess_minmax_aggregates() to consider
> this.

> This doesn't look like a simple project to me, mind.

>> Somewhat harder but still possible would be using BRIN indexes to
>> accelerate ORDER BY. This would require a sorting algorithm that can take
>> advantage of mostly-sorted inputs. You would sort the page ranges by their
>> minimum or maximum value, then feed the sorting algorithm in that order.

> I wouldn't know where to start for this. Maybe once Tom is done with
> planner rejiggering it would make sense to consider looking at how to do
> it.

Yeah. I would urgently recommend that people *not* try to build new
things like planagg.c right now. A large part of the point of upper
planner path-ification is to have a less grotty way of dealing with
things like specialized aggregate implementations.

(And yes, I am working on that. Honest.)

regards, tom lane