Thread: deferrable foreign keys

deferrable foreign keys

From
Morus Walter
Date:
Hi,

are there downsides of making foreign keys deferrable (but initially
immediate) for updates, when the transaction does not set the
constraint behaviour to deferred?

I'd expect that to have the same behaviour as non deferrable foreign
keys.
What I don't understand is, why is non deferrable the default, then.

So maybe I miss something.

regards
    Morus

Re: deferrable foreign keys

From
Grzegorz Jaśkiewicz
Date:


On Wed, Dec 2, 2009 at 2:29 PM, Morus Walter <morus.walter.ml@googlemail.com> wrote:
Hi,

are there downsides of making foreign keys deferrable (but initially
immediate) for updates, when the transaction does not set the
constraint behaviour to deferred?

I'd expect that to have the same behaviour as non deferrable foreign
keys.
What I don't understand is, why is non deferrable the default, then.

it is just sometimes desired to not check the constraints, until comit. For instance, if you run bit of code that is old, and you don't want to mess around with keys.
Or you have some strange way of putting information together.
Basically it is all about order of operation within transaction. Sometimes it cannot be guaranteed, and hence an option to defer the constraint check.



--
GJ

Re: deferrable foreign keys

From
Tom Lane
Date:
Morus Walter <morus.walter.ml@googlemail.com> writes:
> are there downsides of making foreign keys deferrable (but initially
> immediate) for updates, when the transaction does not set the
> constraint behaviour to deferred?

> I'd expect that to have the same behaviour as non deferrable foreign
> keys.
> What I don't understand is, why is non deferrable the default, then.

Because the SQL standard says so.  I don't believe there is any actual
penalty for deferrable within the PG implementation, but perhaps there
is in other systems' implementations.

            regards, tom lane

Re: deferrable foreign keys

From
Morus Walter
Date:
Hallo Tom,

> Morus Walter <morus.walter.ml@googlemail.com> writes:
> > are there downsides of making foreign keys deferrable (but initially
> > immediate) for updates, when the transaction does not set the
> > constraint behaviour to deferred?
>
> > I'd expect that to have the same behaviour as non deferrable foreign
> > keys.
> > What I don't understand is, why is non deferrable the default, then.
>
> Because the SQL standard says so.

Ok. Understood.

> I don't believe there is any actual
> penalty for deferrable within the PG implementation, but perhaps there
> is in other systems' implementations.
>

Thanks a lot for your help.

Morus