On Sun, 2011-03-13 at 19:40 -0400, Tom Lane wrote:
> It strikes me that it'd be possible to have btvacuumcleanup directly
> measure order correlation when it's processing a btree index, yielding a
> reliable answer for any btree index regardless of number of columns.
> We could do that by comparing the heap block numbers of adjacent
> index entries' TIDs and counting the number of up-transitions (block
> number larger than previous index entry) versus down-transitions (block
> number smaller than previous).
Link to previous discussion:
http://archives.postgresql.org/pgsql-hackers/2008-10/msg01279.php
Regards,Jeff Davis