I wrote:
> Weird. I assume that your 'activa' field is 'bool'? I've been trying
> to duplicate this misbehavior here, and as near as I can tell the system
> handles selectivity estimates for boolean fields just fine. Whatever
> percentage of 't' values was seen by the last VACUUM ANALYZE is exactly
> what it uses.
On second thought: 6.5.* can get confused if the column contains more
NULLs than anything else. Dunno if you have a lot of nulls in activa,
but if so you might try changing them all to explicit 'f' and then
redoing the VACUUM ANALYZE. Next release will be smarter about keeping
stats in the presence of many nulls.
It'd be useful to double-check my theory that the system is
misestimating the selectivity of the WHERE (u.activa) clause.
You could try this:SELECT count(*) FROM usarios WHERE activa;EXPLAIN SELECT count(*) FROM usarios WHERE activa;
and see how far off the row count estimate in the EXPLAIN is
from reality.
regards, tom lane