Re: WIP: Upper planner pathification

From: Kouhei Kaigai <kaigai(at)ak(dot)jp(dot)nec(dot)com>
To: Kouhei Kaigai <kaigai(at)ak(dot)jp(dot)nec(dot)com>, Tom Lane <tgl(at)sss(dot)pgh(dot)pa(dot)us>, "Robert Haas" <robertmhaas(at)gmail(dot)com>
Cc: Petr Jelinek <petr(at)2ndquadrant(dot)com>, David Rowley <david(dot)rowley(at)2ndquadrant(dot)com>, "pgsql-hackers(at)postgresql(dot)org" <pgsql-hackers(at)postgresql(dot)org>
Subject: Re: WIP: Upper planner pathification
Date: 2016-03-17 04:13:05
Message-ID: 9A28C8860F777E439AA12E8AEA7694F8011CE9CD@BPXM15GP.gisp.nec.co.jp
Views: Raw Message | Whole Thread | Download mbox | Resend email
Thread:
Lists: pgsql-hackers

> > Robert Haas <robertmhaas(at)gmail(dot)com> writes:
> > > On Mon, Mar 14, 2016 at 9:21 PM, Kouhei Kaigai <kaigai(at)ak(dot)jp(dot)nec(dot)com> wrote:
> > >> So, even though we don't need to define multiple hook declarations,
> > >> I think the hook invocation is needed just after create_xxxx_paths()
> > >> for each. It will need to inform extension the context of hook
> > >> invocation, the argument list will take UpperRelationKind.
> >
> > > That actually seems like a pretty good point. Otherwise you can't
> > > push anything from the upper rels down unless you are prepared to
> > > handle all of it.
> >
> > I'm not exactly convinced of the use-case for that. What external
> > thing is likely to handle window functions but not aggregation,
> > for example?
> >
> WindowPath usually takes a SortPath. Even though extension don't want to
> handle window function itself, it may want to add alternative sort logic
> than built-in.
> Unless it does not calculate expected cost, nobody knows whether WindowPath +
> SortPath is really cheaper than WindowPath + CustomPath("GpuSort").
>
> The supplied query may require to run group-by prior to window function,
> but extension may not be interested in group-by on the other hands, thus,
> extension needs to get control around the location where built-in logic
> also adds paths to fetch the cheapest path of the underlying paths.
>
If I would design the hook, I will put its entrypoint at:
- tail of create_grouping_paths(), just before set_cheapest()
- tail of create_window_paths(), just before set_cheapest()
- tail of create_distinct_paths(), just before set_cheapest()
- tail of create_ordered_paths(), just before set_cheapest()
- tail of grouping_planner(), after the loop of create_modifytable_path()

I'm not 100% certain whether the last one is the straightforward idea
to provide alternative writing stuff. For example, if an extension has
own special storage like columnar format, we may need more consideration
whether CustomPath and related stuff are suitable tool.

On the other hands, I believe the earlier 4 entrypoints are right location.

Thanks,
--
NEC Business Creation Division / PG-Strom Project
KaiGai Kohei <kaigai(at)ak(dot)jp(dot)nec(dot)com>

Attachment Content-Type Size
pgsql-v9.6-upper-custom-path.1.patch application/octet-stream 12.0 KB

In response to

Browse pgsql-hackers by date

  From Date Subject
Next Message Chapman Flack 2016-03-17 04:39:35 typmod is always -1
Previous Message Amit Kapila 2016-03-17 04:01:02 Re: RFC: replace pg_stat_activity.waiting with something more descriptive