Re: planner or statistical bug on 8.5

Il 12/01/2010 08:55, Pavel Stehule ha scritto:
> I checked query and I was surprised with very strange plan:
>
> postgres=# explain select a, b from a,b,c;
> QUERY PLAN
> -----------------------------------------------------------------------
> Nested Loop (cost=0.00..276595350.00 rows=13824000000 width=8)
> -> Nested Loop (cost=0.00..115292.00 rows=5760000 width=8)
> -> Seq Scan on a (cost=0.00..34.00 rows=2400 width=4)
> -> Materialize (cost=0.00..82.00 rows=2400 width=4)
> -> Seq Scan on b (cost=0.00..34.00 rows=2400 width=4)
> -> Materialize (cost=0.00..82.00 rows=2400 width=0)
> -> Seq Scan on c (cost=0.00..34.00 rows=2400 width=0)
> (7 rows)

It doesn't surprise me. Tables are empty, thus get a default non-0 row
estimate, which happens to be 2400:

test=# create table a (a int);
CREATE TABLE
test=# ANALYZE a;
ANALYZE
test=# EXPLAIN SELECT * from a;
QUERY PLAN
-----------------------------------------------------
Seq Scan on a (cost=0.00..14.80 rows=2400 width=4)
(1 row)

That said, 2400^3 (cross join of 3 tables) == 13824000000

Cheers
--
Matteo Beccati

Development & Consulting - http://www.beccati.com/