Re: Why the ERROR: duplicate key violates uniqueconstraint "master_pkey" is raised? - Is this a Bug?

Hi,

On further testing I found the same behaviour in Unique keys too, The following batch can reproduce the behaviour.

CREATE TABLE master ( m1 INT primary key , m2 int unique ) ;

INSERT INTO master VALUES ( 1, 1 ) ;

INSERT INTO master VALUES ( 2, 2) ;

UPDATE master SET m2 = m2 + 1;

I agree with the statement “Success will depend on the order in which postgresql decides to visit rows: if it visits all rows in descending order, you'll be fine,”

But since you can’t control the order, it is likely that the operation will sometimes and will succeed sometimes, with the same data.. Weird I feel .

The work around given by you will work, Does it mean that whenever we are updating a unique key we have to take this precaution ?

On Mon, 2007-09-24 at 12:50 +0530, Anoo Sivadasan Pillai wrote:

CREATE TABLE master ( m1 INT primary key , m2 VARCHAR(100)) ;
INSERT  INTO master VALUES  ( 1, 'm1' ) ;
INSERT  INTO master VALUES  ( 2, 'm2' ) ;
UPDATE  master SET m1 = m1 + 1;
Update fails with the message - ERROR: duplicate key violates unique constraint "master_pkey"

Primary key constraints are not deferred and are not deferrable (try: \x, then select * from pg_constraint where conname~'master_pkey'). This means that the constraint is checked immediately for each row updated. Success will depend on the order in which postgresql decides to visit rows: if it visits all rows in descending order, you'll be fine, but that's unlikely and uncontrollable (AFAIK).

Here's a sketch of an easy workaround. You might have to modify it for your particular range of m1.

begin;
update master set m1=-m1;
update master set m1=-m1+1;
commit;

You could just as easily add N to m1, then subtract (N-1) from m1. You'll need N>max(m1).

Good luck,
Reece

-- 

Reece Hart, http://harts.net/reece/, GPG:0x25EC91A0