Re: BUG #2075: Strange choice of bitmap-index-scan

Hi Tom,

The "zichtbaar" as false is indeed a very rare case and appearantly
isn't occuring right now. There are indeed 46631 rows in total, and all
46631 have the "zichtbaar" as true. Which reminds me to adjust the index
anyway ;-)

It appears to be happening if the fraction of zichtbaar's is small
enough. With 1 and 8 as false, it happens, with 27 as false its not
happening.

Best regards,

Arjen

Tom Lane wrote:
> "Arjen" <acmmailing@tweakers.net> writes:
>>    ->  BitmapAnd  (cost=5.62..5.62 rows=1 width=0)
>>          ->  Bitmap Index Scan on pwprodukten_cat2_popuindex
>> (cost=0.00..2.50 rows=144 width=0)
>>                Index Cond: (cat2 = 51)
>>          ->  Bitmap Index Scan on pwprodukten_cat2_zichtbaar
>> (cost=0.00..2.86 rows=144 width=0)
>>                Index Cond: ((cat2 = 51) AND (zichtbaar = true))
>
> Hmmm ... I can reproduce that if *all* the rows in the table have
> zichtbaar = true (or at least the ANALYZE stats say so) ... is that
> the case in your data?
>
>             regards, tom lane
>