Re: "xmin" system column

On Thu, Jan 26, 2006 at 04:19:34PM -0500, Eric B. Ridge wrote:
> Outside of "VACUUM FREEZE", is there any way the "xmin" column in a
> relation can change, assuming of course the tuple is never updated
> again?  I'm considering using this as a way to identify all tuples
> modified in the same transaction (in an effort to group them
> together), and am wondering if there's any way tuples from different
> transactions could end up with the same xmin value.

I don't know about tuples from different transactions having the
same xmin (aside from 1/BootstrapXID and 2/FrozenXID), but tuples
from the same outer transaction could have different xmin values
due to savepoints.

test=> CREATE TABLE foo (x integer);
test=> BEGIN;
test=> INSERT INTO foo VALUES (1);
test=> SAVEPOINT s;
test=> INSERT INTO foo VALUES (2);
test=> RELEASE SAVEPOINT s;
test=> INSERT INTO foo VALUES (3);
test=> COMMIT;
test=> SELECT xmin, * FROM foo;
  xmin  | x
--------+---
 424584 | 1
 424585 | 2
 424584 | 3
(3 rows)

Explicit savepoints aren't the only way to get this effect; you'll
also see it if the savepoint is implicit, as when trapping errors
in a function.

--
Michael Fuhr