On 2009-05-22, Brandon Metcalf <brandon@geronimoalloys.com> wrote:
> g == gryzman@gmail.com writes:
>
> g> you should use it, whenever you need db to keep its own key internally.
> g> Advantage of sequence is also the fact, that you can have the sequence
> g> value used on different columns/tables .
>
> g> My rule of thumb is , in that case: as long as it is a short type (not
> g> of toastable, or/and variable length), and as long as it won't change,
> g> and is unique - I can use it. Otherwise, I use sequence to connect
> g> rows internally for database.
> g> First rule, is because of index access, and the way btree works.
> g> Second is, because update of value will update other rows too - and
> g> HOT won't help you here, so that's not efficient. And also, forcing it
> g> to be unique is harder than.
>
> g> Hth.
>
> That does help. So, in my example of a table consisting of rows for
> each periodic table element, the atomic number would suffice as a
> unique key since, well, it's unique and not going to change. Right?
Well, until some fool* wants to insert a row for deuterium into your
table, and finds that spot taken by hydrogen.
If you can guarantee that you chosen natural key is sufficient and i
s not going to give trouble in the case of marriages, isomerism, twin
birth, isotopes, centegenarians, or some other condition that makes a
mockery of your chosen key then go for it.
*assuming you want it to be fool-proof.