Achilleus,
> My systems are (rather usual) linux/freebsd and the costs defined (by
> default) in postgresql.conf worked well for all queries except
> a cursed query on a cursed table.
> So i start to believe its an estimation selectivity
> problem.
We can probably fix the problem by re-writing the query then; see my previous
example this weekend about overdetermining criteria in order to force the use
of an index.
How about posting the query and the EXPLAIN ANALYZE results?
--
Josh Berkus
Aglio Database Solutions
San Francisco