Jasen Betts wrote:
> That means storing a long list of numbers and doing queries similar to
> the following to get ne next value for the sequence.
>
> select id from idtable
> order by id
> limit 1
> offset random(0, (select count (*) from idtable)
>
> a ramdom-looking 1:1 mapping is potentially much more efficient.
You'd probably be better off generating it with something like:
CREATE TABLE shuffled AS (n integer, s integer)
AS SELECT n, NULL FROM generate_series(0, max_value) AS n;
SELECT shuffle(); -- sets `s' for each `n'
... then querying it with:
SELECT s FROM shuffled WHERE n = <value-wanted>;
... but you still have to generate, shuffle, and store a huge collection
of values.
--
Craig Ringer
