pg.ddx.io pgsql-performance@postgresql.org mailing list archive
help / color / mirror / Atom feedmultiple table scan performance
5+ messages / 4 participants
[nested] [flat]
* multiple table scan performance
@ 2011-03-29 22:16 Samuel Gendler <sgendler@ideasculptor.com>
2011-03-29 22:28 ` Re: multiple table scan performance Claudio Freire <klaussfreire@gmail.com>
2011-03-29 23:31 ` Re: multiple table scan performance Craig James <craig_james@emolecules.com>
2011-03-30 00:05 ` Re: multiple table scan performance Marti Raudsepp <marti@juffo.org>
0 siblings, 3 replies; 5+ messages in thread
From: Samuel Gendler @ 2011-03-29 22:16 UTC (permalink / raw)
To: pgsql-performance
I've got some functionality that necessarily must scan a relatively large
table. Even worse, the total workload is actually 3 similar, but different
queries, each of which requires a table scan. They all have a resultset
that has the same structure, and all get inserted into a temp table. Is
there any performance benefit to revamping the workload such that it issues
a single:
insert into (...) select ... UNION select ... UNION select
as opposed to 3 separate "insert into () select ..." statements.
I could figure it out empirically, but the queries are really slow on my dev
laptop and I don't have access to the staging system at the moment. Also,
it requires revamping a fair bit of code, so I figured it never hurts to
ask. I don't have a sense of whether postgres is able to parallelize
multiple subqueries via a single scan
^ permalink raw reply [nested|flat] 5+ messages in thread
* Re: multiple table scan performance
2011-03-29 22:16 multiple table scan performance Samuel Gendler <sgendler@ideasculptor.com>
@ 2011-03-29 22:28 ` Claudio Freire <klaussfreire@gmail.com>
2 siblings, 0 replies; 5+ messages in thread
From: Claudio Freire @ 2011-03-29 22:28 UTC (permalink / raw)
To: Samuel Gendler <sgendler@ideasculptor.com>; +Cc: pgsql-performance
On Tue, Mar 29, 2011 at 7:16 PM, Samuel Gendler
<sgendler@ideasculptor.com> wrote:
> Is there any performance benefit to revamping the workload such that it issues
> a single:
> insert into (...) select ... UNION select ... UNION select
> as opposed to 3 separate "insert into () select ..." statements.
I wouldn't expect any difference - if you used UNION ALL (union will
be equivalent to insert into () select DISTINCT ...)
^ permalink raw reply [nested|flat] 5+ messages in thread
* Re: multiple table scan performance
2011-03-29 22:16 multiple table scan performance Samuel Gendler <sgendler@ideasculptor.com>
@ 2011-03-29 23:31 ` Craig James <craig_james@emolecules.com>
2 siblings, 0 replies; 5+ messages in thread
From: Craig James @ 2011-03-29 23:31 UTC (permalink / raw)
To: pgsql-performance
On 3/29/11 3:16 PM, Samuel Gendler wrote:
> I've got some functionality that necessarily must scan a relatively large table. Even worse, the total workload is actually 3 similar, but different queries, each of which requires a table scan. They all have a resultset that has the same structure, and all get inserted into a temp table. Is there any performance benefit to revamping the workload such that it issues a single:
>
> insert into (...) select ... UNION select ... UNION select
>
> as opposed to 3 separate "insert into () select ..." statements.
>
> I could figure it out empirically, but the queries are really slow on my dev laptop and I don't have access to the staging system at the moment. Also, it requires revamping a fair bit of code, so I figured it never hurts to ask. I don't have a sense of whether postgres is able to parallelize multiple subqueries via a single scan
You don't indicate how complex your queries are. If it's just a single table and the conditions are relatively simple, could you do something like this?
insert into (...) select ... where (...) OR (...) OR (...)
Craig
^ permalink raw reply [nested|flat] 5+ messages in thread
* Re: multiple table scan performance
2011-03-29 22:16 multiple table scan performance Samuel Gendler <sgendler@ideasculptor.com>
@ 2011-03-30 00:05 ` Marti Raudsepp <marti@juffo.org>
2011-03-30 00:12 ` Re: multiple table scan performance Samuel Gendler <sgendler@ideasculptor.com>
2 siblings, 1 reply; 5+ messages in thread
From: Marti Raudsepp @ 2011-03-30 00:05 UTC (permalink / raw)
To: Samuel Gendler <sgendler@ideasculptor.com>; +Cc: pgsql-performance
On Wed, Mar 30, 2011 at 01:16, Samuel Gendler <sgendler@ideasculptor.com> wrote:
> I've got some functionality that necessarily must scan a relatively large table
> Is there any performance benefit to revamping the workload such that it issues
> a single:
> insert into (...) select ... UNION select ... UNION select
> as opposed to 3 separate "insert into () select ..." statements.
Apparently not, as explained by Claudio Freire. This seems like missed
opportunity for the planner, however. If it scanned all three UNION
subqueries in parallel, the synchronized seqscans feature would kick
in and the physical table would only be read once, instead of 3 times.
(I'm assuming that seqscan disk access is your bottleneck)
You can trick Postgres (8.3.x and newer) into doing it in parallel
anyway: open 3 separate database connections and issue each of these
'INSERT INTO ... SELECT' parts separately. This way all the queries
should execute in about 1/3 the time, compared to running them in one
session or with UNION ALL.
Regards,
Marti
^ permalink raw reply [nested|flat] 5+ messages in thread
* Re: multiple table scan performance
2011-03-29 22:16 multiple table scan performance Samuel Gendler <sgendler@ideasculptor.com>
2011-03-30 00:05 ` Re: multiple table scan performance Marti Raudsepp <marti@juffo.org>
@ 2011-03-30 00:12 ` Samuel Gendler <sgendler@ideasculptor.com>
0 siblings, 0 replies; 5+ messages in thread
From: Samuel Gendler @ 2011-03-30 00:12 UTC (permalink / raw)
To: Marti Raudsepp <marti@juffo.org>; +Cc: pgsql-performance
On Tue, Mar 29, 2011 at 5:05 PM, Marti Raudsepp <marti@juffo.org> wrote:
> On Wed, Mar 30, 2011 at 01:16, Samuel Gendler <sgendler@ideasculptor.com>
> wrote:
>
> You can trick Postgres (8.3.x and newer) into doing it in parallel
> anyway: open 3 separate database connections and issue each of these
> 'INSERT INTO ... SELECT' parts separately. This way all the queries
> should execute in about 1/3 the time, compared to running them in one
> session or with UNION ALL.
>
That's a good idea, but forces a lot of infrastructural change on me. I'm
inserting into a temp table, then deleting everything from another table
before copying over. I could insert into an ordinary table, but then I've
got to deal with ensuring that everything is properly cleaned up, etc.
Since nothing is actually blocked, waiting for the queries to return, I
think I'll just let them churn for now. It won't make much difference in
production, where the whole table will fit easily into cache. I just wanted
things to be faster in my dev environment.
>
> Regards,
> Marti
>
^ permalink raw reply [nested|flat] 5+ messages in thread
end of thread, other threads:[~2011-03-30 00:12 UTC | newest]
Thread overview: 5+ messages (download: mbox mbox.gz follow: Atom feed)
-- links below jump to the message on this page --
2011-03-29 22:16 multiple table scan performance Samuel Gendler <sgendler@ideasculptor.com>
2011-03-29 22:28 ` Claudio Freire <klaussfreire@gmail.com>
2011-03-29 23:31 ` Craig James <craig_james@emolecules.com>
2011-03-30 00:05 ` Marti Raudsepp <marti@juffo.org>
2011-03-30 00:12 ` Samuel Gendler <sgendler@ideasculptor.com>
This inbox is served by DDX for PostgreSQL; see mirroring instructions
for how to clone and mirror all data and code used for this inbox