Jeff Janes wrote on 26.08.2012 20:45:
> The seq scan is estimated to use sequential reads, while the
> index-only scan is estimated to use random reads (because the index is
> scanned in logical order, not physical order).
>
> If you set random_page_cost equal to seq_page_cost, that would
> artificially favor the index only scan.
>
> Also, your filler is highly compressible, which means the table is
> much smaller than you might think.
I tried it also with 750000 rows filled with 3 text columns of random string (between 20 and 15000 characters long).
But also with that bigger data I just don't get an index scan.
Seems that the prerequisites for an index only scan to happen are quite narrow.
But given the fact that it's a brand new feature I guess it will improve over time ;)
Regards
Thomas