To: "Spiegelberg, Greg" <gspiegelberg@cranel.com>
cc: pgsql-performance@postgresql.org
Subject: Re: DELETE with filter on ctid 
In-reply-to: <82E74D266CB9B44390D3CCE44A781ED90B6ADC@POSTOFFICE.cranel.local> 
References: <82E74D266CB9B44390D3CCE44A781ED90B6ADC@POSTOFFICE.cranel.local>
Comments: In-reply-to "Spiegelberg, Greg" <gspiegelberg@cranel.com>
	message dated "Mon, 09 Apr 2007 16:01:53 -0400"
Date: Mon, 09 Apr 2007 16:55:27 -0400
Message-ID: <10502.1176152127@sss.pgh.pa.us>
From: Tom Lane <tgl@sss.pgh.pa.us>

"Spiegelberg, Greg" <gspiegelberg@cranel.com> writes:
> We have a query which generates a small set of rows (~1,000) which are
> to be used in a DELETE on the same table.  The problem we have is that
> we need to join on 5 different columns and it takes far too long.  I
> have a solution but I'm not sure it's the right one.  Instead of joining
> on 5 columns in the DELETE the join uses the ctid column.

> BEGIN;
> CREATE INDEX gregs_table_ctid_idx ON gregs_table(ctid);
> DELETE FROM gregs_table gt
>    USING (SELECT ctid FROM gregs_table WHERE ...) as s
>    WHERE gt.ctid=s.ctid;
> DROP INDEX gregs_table_ctid_idx;
> COMMIT;

Forget the index, it's useless here (hint: ctid is a physical address).
I'm wondering though why you don't just transpose the subquery's WHERE
condition into the DELETE's WHERE?  Or is this example oversimplified?

			regards, tom lane