Thread: update vs unique index

Consider such table:

CREATE TABLE test (idx integer);

populated by following statements:

INSERT INTO test VALUES (1);
INSERT INTO test VALUES (2);
INSERT INTO test VALUES (3);

since idx schould be unique, we need an index

CREATE UNIQUE INDEX i_test ON test(idx);

Following SQL command fails:

UPDATE test SET idx = idx + 1;

I can imagine why it fails. Update operates on first row, making 2 out
of 1 and that collides with second row (which has 2 as its value
already). However, when you look at the update efect as a whole
uniqueness is preserved, so index schould not veto update.

My question is: is there a chance to bypass this behaviour? Something
like controlling the order in which rows go into update. If update
would start from last row, it would be successful for sure.

regards,
-- 
Jacek Prucia

On Mon, Jul 25, 2005 at 05:58:43AM -0700, jacekp@poczta.wprost.pl wrote:

> I can imagine why it fails. Update operates on first row, making 2 out
> of 1 and that collides with second row (which has 2 as its value
> already). However, when you look at the update efect as a whole
> uniqueness is preserved, so index schould not veto update.
> 
> My question is: is there a chance to bypass this behaviour? Something
> like controlling the order in which rows go into update. If update
> would start from last row, it would be successful for sure.

Yeah, this is a known limitation.  Usual workaround is issue two
updates instead of one,

update foo set a = -a where <condition>;
update foo set a = -a + 1 where <condition>;

The point is to move all unique keys to an unused interval and then
move them back, changed all at a time.

It'll eventually be fixed, but don't hold your breath.

-- 
Alvaro Herrera (<alvherre[a]alvh.no-ip.org>)
"No hay ausente sin culpa ni presente sin disculpa" (Prov. francés)