Re: PassDownLimitBound for ForeignScan/CustomScan [take-2]

From: Kouhei Kaigai <kaigai(at)ak(dot)jp(dot)nec(dot)com>
To: Robert Haas <robertmhaas(at)gmail(dot)com>
Cc: "pgsql-hackers(at)postgresql(dot)org" <pgsql-hackers(at)postgresql(dot)org>, "Jeevan Chalke" <jeevan(dot)chalke(at)enterprisedb(dot)com>, Etsuro Fujita <fujita(dot)etsuro(at)lab(dot)ntt(dot)co(dot)jp>, Andres Freund <andres(at)anarazel(dot)de>
Subject: Re: PassDownLimitBound for ForeignScan/CustomScan [take-2]
Date: 2016-11-09 23:49:44
Views: Raw Message | Whole Thread | Download mbox | Resend email
Lists: pgsql-hackers

> -----Original Message-----
> From: pgsql-hackers-owner(at)postgresql(dot)org
> [mailto:pgsql-hackers-owner(at)postgresql(dot)org] On Behalf Of Robert Haas
> Sent: Thursday, November 10, 2016 3:08 AM
> To: Kaigai Kouhei(海外 浩平)
> Cc: pgsql-hackers(at)postgresql(dot)org; Jeevan Chalke; Etsuro Fujita; Andres Freund
> Subject: Re: [HACKERS] PassDownLimitBound for ForeignScan/CustomScan [take-2]
> On Mon, Oct 31, 2016 at 10:20 AM, Kouhei Kaigai <kaigai(at)ak(dot)jp(dot)nec(dot)com> wrote:
> > As an example, I enhanced postgres_fdw to understand the ps_numTuples
> > if it is set. If and when remote ORDER BY is pushed down, the latest
> > code tries to sort the entire remote table because it does not know
> > how many rows to be returned. Thus, it took larger execution time.
> > On the other hands, the patched one runs the remote query with LIMIT
> > clause according to the ps_numTuples; which is informed by the Limit
> > node on top of the ForeignScan node.
> So there are two cases here. If the user says LIMIT 12, we could in
> theory know that at planner time and optimize accordingly. If the
> user says LIMIT twelve(), however, we will need to wait until
> execution time unless twelve() happens to be capable of being
> simplified to a constant by the planner.
> Therefore, it's possible to imagine having two mechanisms here. In the
> simple case where the LIMIT and OFFSET values are constants, we could
> implement a system to get hold of that information during planning and
> use it for whatever we like. In addition, we can have an
> execution-time system that optimizes based on values available at
> execution (regardless of whether those values were also available
> during planning). Those are, basically, two separate things, and this
> patch has enough to do just focusing on one of them.
OK, we need to have a private value of postgres_fdw to indicate whether
LIMIT and OFFSET were supplied at the planner stage. If any, it has to
be matched with the ps_numTuples informed at the executor stage.

I'll revise the patch.
NEC OSS Promotion Center / PG-Strom Project
KaiGai Kohei <kaigai(at)ak(dot)jp(dot)nec(dot)com>

In response to

Browse pgsql-hackers by date

  From Date Subject
Next Message Kouhei Kaigai 2016-11-09 23:59:02 Re: PassDownLimitBound for ForeignScan/CustomScan [take-2]
Previous Message Clifford Hammerschmidt 2016-11-09 23:18:05 Re: C based plugins, clocks, locks, and configuration variables