Re: Partitions are not excluded in stored procedures

From: Marcelo Sena <marceloslacerda(at)gmail(dot)com>
To: pgsql-novice(at)postgresql(dot)org
Subject: Re: Partitions are not excluded in stored procedures
Date: 2012-06-09 18:21:34
Message-ID: CAPmRTtMRu5ORsYWECfVjt8UwiYRhS_iHWtb2Pmn6ydvm_Z+tvQ@mail.gmail.com
Views: Raw Message | Whole Thread | Download mbox | Resend email
Thread:
Lists: pgsql-novice

I might be misunderstanding but I think that what Tom meant is that the
EXPLAIN does not try to guess whether
the constraint will actually exclude some data, so it show you a
pessimistic estimate of the cost.

If I'm right, using EXPLAIN ANALYZE should show that the query considers
any constraint clauses in your query.

Cheers,
Marcelo Lacerda

On 06/06/2012 05:10 AM, Ioannis Anagnostopoulos wrote:

On 05/06/2012 17:55, Tom Lane wrote:

Ioannis Anagnostopoulos<ioannis(at)anatec(dot)com> <ioannis(at)anatec(dot)com> writes:

I came across the following problem. Say I have a select query that
given the appropriate where clauses it will only query the correct
partitioned tables and not all of them. That can be examined by using
the EXPLAIN. However the same select query used in a stored procedure
will transverse all the tables regardless of the where clause parameters
passed by the user as variables of the stored procedure . Is there any
work around other than composing an EXECUTE "select ....."?

Yeah, if the WHERE clauses that correspond to the partitioning
constraints contain parameters, the planner can't prove the exclusions
hold. So you have to use EXECUTE to get a one-shot plan. This will
be better in 9.2 (which, basically, will automatically recognize that
it needs to use one-shot plans).

regards, tom lane

When you say better in 9.2 you mean that we will be able to avoid the
EXECUTE, and the plan will be, more or less, calculated per call without
great performance hit?

Regards and thank you
Yiannis

In response to

Browse pgsql-novice by date

  From Date Subject
Next Message Marcelo Lacerda 2012-06-09 18:41:48 Re: EDTERROR: invalid page header in block 3271 of relation base/16395/16426
Previous Message Prima Chairunnanda 2012-06-09 01:30:47 Re: Re: Buffer Management: Can dirty pages be written before transaction commits?