Date: Sat, 15 Dec 2007 21:21:47 -0800
From: andrew@pillette.com
Message-Id: <200712160521.lBG5Llf25219@pillette.com>
Subject: Re: update 600000 rows
To: okparanoid@free.fr
Cc: pgsql-performance@postgresql.org
MIME-Version: 1.0
Content-Type: text/plain
Content-Transfer-Encoding: 7bit

Loīc Marteau <okparanoid@free.fr> wrote ..
> Steve Crawford wrote:
> > If this
> > is correct, I'd first investigate simply loading the csv data into a
> > temporary table, creating appropriate indexes, and running a single 
> > query to update your other table.

My experience is that this is MUCH faster. My predecessor in my current position was doing an update from a csv file line by line with perl. That is one reason he is my predecessor. Performance did not justify continuing his contract.
 
> i can try this. The problem is that i have to make an insert if the 
> update don't have affect a rows (the rows don't exist yet). The number
> of rows affected by insert is minor regards to the numbers of updated 
> rows and was 0 when i test my script). I can do with a temporary table
> : update all the possible rows and then insert the rows that are in 
> temporary table and not in the production table with a 'not in' 
> statement. is this a correct way ?

That's what I did at first, but later I found better performance with a TRIGGER on the permanent table that deletes the target of an UPDATE, if any, before the UPDATE. That's what PG does anyway, and now I can do the entire UPDATE in one command.