Em 25-01-2013 16:29, Jeff Janes escreveu:
> On Fri, Jan 25, 2013 at 7:34 AM, alexandre - aldeia digital
> <adaldeia@gmail.com> wrote:
>> Hi,
>>
>> Last weekend, we upgrade a PG from 8.4 to 9.2 version (full
>> pg_dump/restore/vacuum/analyze). After this, some simple join querys
>> became very slow, maybe because the use of nested loops. Bellow, an example
>> with nestedloop on and off:
>
> What happens if you bump up default_statistics_target by a factor of
> 10 or 100 and redo the analyze?
Before send the e-mail, the default_statistics_target was 500 and I
return to 100 (default). I will try to set 1000.
> Here it is finding 39 times more rows than expected:
>
> Index Scan using ad_pc13t3_modpadrao on pc13t3 t1 (cost=0.00..6.21
> rows=1 width=65) (actual time=0.090..0.252 rows=39 loops=1)
>
> It would interesting to know why that is.
This is a partial index:
"ad_pc13t3_modpadrao" btree (pc13emp08p, pc13anoped, pc13codped,
pc13codigo, pc13cor, pc13empins, pc13emp08, pc13tipin2, pc13insest)
WHERE pc13emp08p = 0 AND pc13anoped = 0 AND pc13codped = 0 AND pc13item = 0
I can't connect to databse now. I will retry tests in sunday.
Best regards,
Alexandre