On Thu, Jul 8, 2010 at 10:11 AM, Tom Lane <tgl@sss.pgh.pa.us> wrote:
> Martin Below <machtin.below@googlemail.com> writes:
>> test=# select count(*) total, count(distinct client_id) ids,
>> count(distinct expires_on) dates from ps;
>> total | ids | dates
>> --------+--------+--------
>> 213645 | 123366 | 213549
>
> That says the expires_on column is practically unique, which makes me
> think the planner is indeed making the right choice.
I think so too, but suppose we wanted to force the other plan anyways:
select * from ps where (client_id, expires_on) >=
('123', '24.11.2010'::timestamp) and (client_id, expires_on) < ('123',
null) order by client_id, expires_on;
I'd be curious to see explain analyze (not explain) comparisons for
the 'wrong' index vs above. I suspect the plan is 'correct' for
*most* of the data, or you cherry picked (or unluckily drew) a bad
value to get your 22 times speed difference.
merlin