On Sat, Feb 21, 2015 at 2:09 PM, Peter Geoghegan <pg@heroku.com> wrote:
> I can recreate the problem you see with text sort regressions.
> Abbreviation is aborted for the case in question, unsurprisingly, and
> fairly far in. With that many tuples, the idea of taking abbreviated
> cardinality as a proxy for full cardinality becomes less important,
> because either way you have to do at least 10 comparisons per item on
> average.
Actually, it's closer to 20 comparisons past 1 million, on average.
See my earlier 0002-* patch comments [1].
[1] http://www.postgresql.org/message-id/attachment/35861/0002-Estimate-total-number-of-rows-to-be-sorted.patch
(search for "20")
--
Peter Geoghegan