Re: Duplicate key violation - Mailing list pgsql-general

From Tom Lane
Subject Re: Duplicate key violation
Date
Msg-id 15029.1169770847@sss.pgh.pa.us
Whole thread Raw
In response to Duplicate key violation  (Brian Wipf <brian@clickspace.com>)
Responses Re: Duplicate key violation
List pgsql-general
Brian Wipf <brian@clickspace.com> writes:
> I got a duplicate key violation when the following query was performed:
> INSERT INTO category_product_visible (category_id, product_id)
>         SELECT     cp.category_id, cp.product_id
>         FROM     category_product cp
>         WHERE     cp.product_id = $1 AND
>             not exists (
>                 select     'x'
>                 from     category_product_visible cpv
>                 where     cpv.product_id = cp.product_id and
>                     cpv.category_id = cp.category_id
>             );

> This is despite the fact the insert is written to only insert rows
> that do not already exist. The second time the same query was run it
> went through okay. This makes me think there is some kind of race
> condition, which I didn't think was possible with PostgreSQL's MVCC
> implementation.

If you're doing more than one of these concurrently, then of course
there's a race condition: the NOT EXISTS is testing for nonexistence
as of the query snapshot.  If two sessions do this concurrently then
they'll try to insert the same rows and one of them is gonna fail.

            regards, tom lane

pgsql-general by date:

Previous
From: Ron Johnson
Date:
Subject: Re: Controlling Database Growth
Next
From: gustavo halperin
Date:
Subject: triggers vs b-tree