Date: Thu, 23 Feb 2006 17:30:55 +0100
To: mario.splivalo@mobart.hr,
	"Achilleus Mantzios" <achill@matrix.gatewaynet.com>
Subject: Re: 'locking' the SELECTs based on indices...
Cc: pgsql-sql@postgresql.org
References: <Pine.LNX.4.44.0602231349430.15710-100000@matrix.gatewaynet.com>
	<1140692542.5002.26.camel@localhost.localdomain>
From: PFC <lists@peufeu.com>
Content-Type: text/plain; format=flowed; delsp=yes; charset=utf-8
MIME-Version: 1.0
Content-Transfer-Encoding: 8bit
Message-ID: <op.s5fphtpwcigqcu@apollo13>
In-Reply-To: <1140692542.5002.26.camel@localhost.localdomain>
User-Agent: Opera M2/8.51 (Linux, build 1462)


> I lock just that particular row, which is no good. I need to have all
> the codes for the service 1 locked, so if it happens that two users send
> the very same code, one has to fail. Therefore, from within plpgsql I
> first do:

	I'm a bit tired tonight so I'll simplify your example :

CREATE TABLE stuff ( a INT, b INT );

	Basically you want to lock ALL rows with a certain value of a, in order  
to perform an operation on only one of them.
	You could do this :

CREATE TABLE all_as ( a INT PRIMARY KEY )
CREATE TABLE stuff ( a INT REFERENCES all_as(a), b INT );

	Now all the rows in "stuff" that have the same value of "a" reference the  
same row in "all_as".
	All you have to do is

	SELECT * FROM all_as WHERE a=the value FOR UPDATE

	and you lock all rows having that particular value of a in the big table.