Message-ID: <5422D171.5060801@encs.concordia.ca>
Date: Wed, 24 Sep 2014 10:13:05 -0400
From: Emi Lu <emilu@encs.concordia.ca>
Reply-To: emilu@encs.concordia.ca
User-Agent: Mozilla/5.0 (X11; Linux i686 on x86_64;
 rv:24.0) Gecko/20100101 Thunderbird/24.8.0
MIME-Version: 1.0
To: Heikki Linnakangas <hlinnakangas@vmware.com>,
 "pgsql-performance@postgresql.org" <pgsql-performance@postgresql.org>
Subject: Re: Which update action quicker?
References: <5421D9FB.2030008@encs.concordia.ca> <5422CBB4.3000301@vmware.com>
In-Reply-To: <5422CBB4.3000301@vmware.com>
Content-Type: text/plain; charset=windows-1252; format=flowed
Content-Transfer-Encoding: 7bit
Precedence: bulk
Sender: pgsql-performance-owner@postgresql.org

Hello,

> For a big table with more than 10 Million records, may I know which update is
> quicker please?
> (1) update t1
>         set c1 = a.c1
>         from a
>         where pk and
>                    t1.c1       <> a.c1;
>    ......
>         update t1
>         set c_N = a.c_N
>         from a
>         where pk and
>                    t1.c_N       <> a.c_N;
>
>
> (2)  update t1
>         set c1 = a.c1 ,
>               c2  = a.c2,
>               ...
>               c_N = a.c_N
>        from a
>        where pk AND
>        (t1.c1, c2...c_N) <> (a.c1, c2... c_N)

Probably (2). <> is not indexable, so each update will have to perform a
sequential scan of the table. With (2), you only need to scan it once,
with (1) you have to scan it N times. Also, method (1) will update the
same row multiple times, if it needs to have more than one column updated.

> Or other quicker way for update action?

If a large percentage of the table needs to be updated, it can be faster
to create a new table, insert all the rows with the right values, drop
the old table and rename the new one in its place. All in one transaction.

The situation is:
(t1.c1, c2, ... c_N) <> (a.c1, c2...c_N) won't return too many diff records. So, the calculation will only be query most of the case.

But if truncate/delete and copy will cause definitely write all more than 10 million data.

If for situation like this, will it still be quicker to delete/insert quicker?
Thank you
Emi


-- 
Sent via pgsql-performance mailing list (pgsql-performance@postgresql.org)
To make changes to your subscription:
http://www.postgresql.org/mailpref/pgsql-performance