On 11 December 2013 00:44, Greg Stark <stark@mit.edu> wrote:
> On Wed, Dec 11, 2013 at 12:40 AM, Simon Riggs <simon@2ndquadrant.com> wrote:
>> When we select a block we should read all rows on that block, to help
>> identify the extent of clustering within the data.
>
> So how do you interpret the results of the sample read that way that
> doesn't introduce bias?
Yes, it is not a perfect statistical sample. All sampling is subject
to an error that is data dependent.
I'm happy that we have an option to select this/or not and a default
that maintains current behaviour, since otherwise we might expect some
plan instability.
I would like to be able to
* allow ANALYZE to run faster in some cases
* increase/decrease sample size when it matters
* have the default sample size vary according to the size of the
table, i.e. a proportional sample
-- Simon Riggs http://www.2ndQuadrant.com/PostgreSQL Development, 24x7 Support, Training & Services