User-Agent: Microsoft-MacOutlook/14.4.1.140326
Date: Tue, 13 May 2014 21:08:09 +0100
Subject: Adaptive query execution
From: Tim Kane <tim.kane@gmail.com>
To: "pgsql-performance@postgresql.org" <pgsql-performance@postgresql.org>
Message-ID: <CF983A39.7EF6F%tim.kane@gmail.com>
Thread-Topic: Adaptive query execution
Mime-version: 1.0
Content-type: multipart/alternative;
	boundary="B_3482860098_69660349"
Precedence: bulk
Sender: pgsql-performance-owner@postgresql.org

> This message is in MIME format. Since your mail reader does not understand
this format, some or all of this message may not be legible.

--B_3482860098_69660349
Content-type: text/plain;
	charset="UTF-8"
Content-transfer-encoding: quoted-printable


Hi all,

So I was thinking about the following, after experimenting with constraint
exclusion.

I thought I would see what happens when I do this:

  SELECT * FROM ONLY table_a UNION SELECT * FROM table_b;


I noticed that despite table_a still having no data in it, the planner has
already decided that it needs to insert a chain of  =E2=80=98append->sort->unique=
=E2=80=99
nodes into the plan.

That=E2=80=99s fairly reasonable.
While I understand that we can=E2=80=99t readily know about wether a given node w=
ill
return anything or not - would it be possible to have the execution engine
branch off in the event that a given node returns nothing at all?

I guess there are probably a lot of considerations, and I suspect it would
considerably increase planning time, though maybe it also presents an
opportunity for some interesting approaches to adaptive query execution.

I don=E2=80=99t know so much about this, though I=E2=80=99m sure there are all kinds of
research papers discussing it.
Is this something that has been considered before?

Tim


--B_3482860098_69660349
Content-type: text/html;
	charset="UTF-8"
Content-transfer-encoding: quoted-printable

<html><head></head><body style=3D"word-wrap: break-word; -webkit-nbsp-mode: s=
pace; -webkit-line-break: after-white-space; color: rgb(0, 0, 0); font-size:=
 14px; font-family: Calibri, sans-serif;"><div><br></div><div><div>Hi all,</=
div><div><br></div><div>So I was thinking about the following, after experim=
enting with constraint exclusion.</div><div><br></div><div><div>I thought I =
would see what happens when I do this:</div><div><br></div><div><font face=3D"=
Consolas">&nbsp; SELECT * FROM ONLY table_a UNION SELECT * FROM table_b;</fo=
nt></div><div><br></div><div><br></div><div>I noticed that despite&nbsp;<i>t=
able_a</i>&nbsp;still having no data in it, the planner has already decided =
that it needs to insert a chain of &nbsp;&#8216;append-&gt;sort-&gt;unique&#=
8217; &nbsp;nodes into the plan.</div><div><br></div><div>That&#8217;s fairl=
y reasonable.</div><div>While I understand that we can&#8217;t readily know =
about wether a given node will return anything or not - would it be possible=
 to have the execution engine branch off in the event that a given node retu=
rns nothing at all?</div><div><br></div><div>I guess there are probably a lo=
t of considerations, and I suspect it would considerably increase planning t=
ime, though maybe it also presents an opportunity for some interesting appro=
aches to adaptive query execution.</div><div><br></div></div></div><div>I do=
n&#8217;t know so much about this, though I&#8217;m sure there are all kinds=
 of research papers discussing it.</div><div>Is this something that has been=
 considered before?</div><div><br></div><div>Tim</div><div><br></div></body>=
</html>

--B_3482860098_69660349--