On Mon, Nov 10, 2008 at 12:54 PM, Bryce Nesbitt <bryce2(at)obviously(dot)com> wrote:
> I've got a table for which "CLUSTER tablename USING index" makes an order of
> magnitude difference.
>
> Are there ways to determine how "unclustered" this table becomes over time,
> so I can schedule downtime to recluster? I could use the pg_stat tables,
> but this seems awkward.
You should be able to run analyze then select correlation from
pg_stats where schemaname='yourschename' and
tablename='yourtablename';
the closer you are to 1.0 the better the clustering.