To: mario.splivalo@mobart.hr
cc: pgsql-sql@postgresql.org
Subject: Re: 'locking' the SELECTs based on indices... 
In-reply-to: <1140631350.14229.8.camel@localhost.localdomain> 
References: <1140631350.14229.8.camel@localhost.localdomain>
Comments: In-reply-to Mario Splivalo <mario.splivalo@mobart.hr>
	message dated "Wed, 22 Feb 2006 19:02:30 +0100"
Date: Wed, 22 Feb 2006 13:58:24 -0500
Message-ID: <14024.1140634704@sss.pgh.pa.us>
From: Tom Lane <tgl@sss.pgh.pa.us>

Mario Splivalo <mario.splivalo@mobart.hr> writes:
> Now, when I do this from one connection (psql shell, for instance):

> [A]BEGIN TRANSACTION;
> [A]SELECT * FROM bla WHERE code_id = 1 FOR UPDATE;

> and then, from another psql i do:
> [B]SELECT * FROM bla WHERE code_id = 1 FOR UPDATE

> the second SELECT will wait untill I rollback or commit first
> transaction. That is cool. 

> But, if I do second SELECT like this:

> [C]SELECT * FROM bla WHERE code_id = 2 FOR UPDATE

> I will get the rows.

Well, of course.  Why would you want something different?  Why do you
think the table's indexes should have anything to do with it?

If you want a full-table lock then some form of LOCK TABLE seems like
the proper answer.  SELECT FOR UPDATE is designed to lock the specified
rows, no more.

> If I erase the index bla_idx1, then [C] select will wait, same as [B]
> select will wait.

I don't think so.  If it does, it's a bug; please provide a reproducible
test case.

			regards, tom lane