Re: [PATCH] random_normal function - Mailing list pgsql-hackers

I wrote:
> Also, I tried running the new random.sql regression cases over
> and over, and found that the "not all duplicates" test fails about
> one time in 100000 or so.  We could probably tolerate that given
> that the random test is marked "ignore" in parallel_schedule, but
> I thought it best to add one more iteration so we could knock the
> odds down.

Hmm ... it occurred to me to try the same check on the existing
random() tests (attached), and darn if they don't fail even more
often, usually within 50K iterations.  So maybe we should rethink
that whole thing.

            regards, tom lane

\timing on

create table if not exists random_tbl (random bigint);

do $$
begin
for i in 1..1000000 loop
TRUNCATE random_tbl;

INSERT INTO RANDOM_TBL (random)
  SELECT count(*) AS random
  FROM onek WHERE random() < 1.0/10;

-- select again, the count should be different
INSERT INTO RANDOM_TBL (random)
  SELECT count(*)
  FROM onek WHERE random() < 1.0/10;

-- select again, the count should be different
INSERT INTO RANDOM_TBL (random)
  SELECT count(*)
  FROM onek WHERE random() < 1.0/10;

-- select again, the count should be different
INSERT INTO RANDOM_TBL (random)
  SELECT count(*)
  FROM onek WHERE random() < 1.0/10;

-- now test that they are different counts
if (select true FROM RANDOM_TBL
  GROUP BY random HAVING count(random) > 3) then
raise notice 'duplicates at iteration %', i;
exit;
end if;

-- approximately check expected distribution
if (select true FROM RANDOM_TBL
  HAVING AVG(random) NOT BETWEEN 80 AND 120) then
raise notice 'range at iteration %', i;
exit;
end if;

end loop;
end $$;