Обсуждение: setting for maximum acceptable plan cost?

I was thinking that it might be nice to be able to tell postgres to
refuse to execute any plan with an estimated cost above some threshold.
For example, earlier today I produced this extremely bogus execution
plan with the following top line:

Nested Loop Left Join  (cost=13920.16..2257575559347.46 rows=3691992705807 width=128)

After a call to ANALYZE, the same query gave me:

Merge Left Join  (cost=16382.02..16853.87 rows=126768 width=59)

And runs in 5 seconds.  If I had been able to tell pg to reject any plan
with cost over, say 10E9, that would have saved my server from half an
hour of nested sequential scans.

Should I just use statement_timeout as a proxy for this?

-jwb

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

On Fri, 02 Nov 2007 13:49:27 -0700
"Jeffrey W. Baker" <jwbaker@acm.org> wrote:

> Nested Loop Left Join  (cost=13920.16..2257575559347.46
> rows=3691992705807 width=128)
> 
> After a call to ANALYZE, the same query gave me:
> 
> Merge Left Join  (cost=16382.02..16853.87 rows=126768 width=59)
> 
> And runs in 5 seconds.  If I had been able to tell pg to reject any
> plan with cost over, say 10E9, that would have saved my server from
> half an hour of nested sequential scans.

I am confused as to why you would want to do that... seems like a
band aid for lack of maintenance.

> 
> Should I just use statement_timeout as a proxy for this?
> 

That would yes but see my point about maintenance above.

Joshua D. Drake



- -- 

      === The PostgreSQL Company: Command Prompt, Inc. ===
Sales/Support: +1.503.667.4564   24x7/Emergency: +1.800.492.2240
PostgreSQL solutions since 1997  http://www.commandprompt.com/
            UNIQUE NOT NULL
Donate to the PostgreSQL Project: http://www.postgresql.org/about/donate
PostgreSQL Replication: http://www.commandprompt.com/products/

-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.6 (GNU/Linux)

iD8DBQFHK5peATb/zqfZUUQRAttyAJ9pHjQUuyY7e2cJXtkB2239vOqAxACfX2XW
AHVlhc4g/mzc7uesWpAGls0=
=i6n+
-----END PGP SIGNATURE-----

"Jeffrey W. Baker" <jwbaker@acm.org> writes:
> I was thinking that it might be nice to be able to tell postgres to
> refuse to execute any plan with an estimated cost above some threshold.

What you suggest has been suggested before, and I might think it was
a good idea if I trusted the planner's cost estimates more ;-)

There's always statement_timeout.

            regards, tom lane

On Fri, 2007-11-02 at 14:45 -0700, Joshua D. Drake wrote:
> -----BEGIN PGP SIGNED MESSAGE-----
> Hash: SHA1
>
> On Fri, 02 Nov 2007 13:49:27 -0700
> "Jeffrey W. Baker" <jwbaker@acm.org> wrote:
>
> > Nested Loop Left Join  (cost=13920.16..2257575559347.46
> > rows=3691992705807 width=128)
> >
> > After a call to ANALYZE, the same query gave me:
> >
> > Merge Left Join  (cost=16382.02..16853.87 rows=126768 width=59)
> >
> > And runs in 5 seconds.  If I had been able to tell pg to reject any
> > plan with cost over, say 10E9, that would have saved my server from
> > half an hour of nested sequential scans.
>
> I am confused as to why you would want to do that... seems like a
> band aid for lack of maintenance.

Well it's not "maintenance" really since all the inputs are temp tables,
but I do see your point.

Often I have wished for a language which is not SQL which would allow me
to simply specify the whole execution plan.  That would cut out a lot of
ambiguity.

Pie in the sky, I know.

-jwb