Ron Arts <ron.arts@neonova.nl> writes:
> # select count(*) from cdr where billed = false;
> count
> -------
> 3
> (1 row)
>> =# explain select cdr.* from cdr where billed = false order by calldate;
>>
>> QUERY PLAN
>> -------------------------------------------------------------------
>> Sort (cost=37448.75..37526.94 rows=31273 width=465)
>> Sort Key: calldate
>> -> Seq Scan on cdr (cost=0.00..20323.81 rows=31273 width=465)
^^^^^^^^^^
>> Filter: (billed = false)
You haven't ANALYZEd this table (lately, or perhaps ever). The planner
is thus working with a default selectivity estimate (which I think is
50% for a bool column --- that seems to match your numbers anyway), and
that leads it to the conclusion that a seqscan is the right thing.
Which it would indeed be, if half the table has to be retrieved.
regards, tom lane