Обсуждение: converting FK's to "DEFERRABLE"

In order to try to reduce lock contention on my FK's, I need to
convert them to DEFERRABLE.  The straightforward way is to drop and
recreate the modified FK.  However, on a table with 65M rows, this is
taking quite some time.  I'm afraid how long it will take to update
both FK's on my 170M+ row table...

Anyhow, is there some trickier way to make an FK deferrable?  Mucking
with the system tables, perhaps?

I see that pg_restore has a way to turn off triggers during the data
load.  If I can guarantee no updates to the table in question, can I
use that same code to disable triggers, drop+add the FK, then
re-enable triggers?  Or will that not avoid the check when I create
the new FK?

I'd like to avoid a few hours of downtime while updating these
triggers.

Thanks.

--
=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=
Vivek Khera, Ph.D.                Khera Communications, Inc.
Internet: khera@kciLink.com       Rockville, MD  +1-301-869-4449 x806
AIM: vivekkhera Y!: vivek_khera   http://www.khera.org/~vivek/

Vivek Khera <khera@kcilink.com> writes:
> Anyhow, is there some trickier way to make an FK deferrable?

Hack its pg_constraint.condeferrable and pg_constraint.condeferred
fields (the latter is the INITIALLY DEFERRED flag).  You will also
need to find the triggers that implement the constraint and update
their pg_trigger.tgdeferrable and pg_trigger.tginitdeferred copies
of these values.  Then start fresh backend sessions and I think
you're there.

AFAIK the most reliable way to find the triggers is to follow the
linking entries in pg_depend.

            regards, tom lane

On Sep 17, 2004, at 3:27 PM, Tom Lane wrote:

> Vivek Khera <khera@kcilink.com> writes:
>> Anyhow, is there some trickier way to make an FK deferrable?
>
> Hack its pg_constraint.condeferrable and pg_constraint.condeferred
> fields (the latter is the INITIALLY DEFERRED flag).  You will also
> need to find the triggers that implement the constraint and update
> their pg_trigger.tgdeferrable and pg_trigger.tginitdeferred copies
> of these values.  Then start fresh backend sessions and I think
> you're there.

Thanks a bunch.  This worked flawlessly.  Basically I did this:

begin;
select pg_constraint.oid from pg_constraint,pg_class where
pg_constraint.conrelid=pg_class.oid and relname='mytable' and
conname='$1';
X=oid number
update pg_constraint set condeferrable='t' where oid=X;
update pg_trigger set tgdeferrable='t' where oid in (select objid from
pg_depend where refobjid=X);
commit;



=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=
Vivek Khera, Ph.D.                MailerMailer, LLC     Rockville, MD
http://www.MailerMailer.com/                     +1-301-869-4449 x806