Re: PG planning randomly ?

"Laurent Raufaste" <analogue(at)glop(dot)org> writes:
> 2008/2/26, Tom Lane <tgl(at)sss(dot)pgh(dot)pa(dot)us>:
>> ... I'm wondering if you have a definition of operator <@
>> that doesn't specify the new selectivity estimator. Please try a
>> pg_dump -s and see what it shows as the definition of <@.

> Here's the first definition of the <@ operator in my dump:

> CREATE OPERATOR <@ (
> PROCEDURE = ltree_risparent,
> LEFTARG = ltree,
> RIGHTARG = ltree,
> COMMUTATOR = @>,
> RESTRICT = ltreeparentsel,
> JOIN = contjoinsel
> );

That's the right RESTRICT function, but what exactly did you mean by
"first definition"? Are there more?

It may be that it's just not possible for the estimator to come up with
accurate rowcount estimates given the amount of info it has available.
The query you are complaining about confuses the issue quite a lot by
involving other issues. Would you try just "explain analyze select 1
from _commment where path <@ '....';" for various typical path values,
and see if it's coming up with estimated rowcounts that are in the right
ballpark compared to the actual ones?

regards, tom lane