Heikki Linnakangas <heikki(dot)linnakangas(at)enterprisedb(dot)com> wrote:
> Implementing index-only scans requires a few changes:
I'm happy to see this work! Now that the EXISTS predicates have been
optimized to consider semi-join and anti-join techniques, I believe
that these index issues (evaluating quals before heap access and
skipping heap access when all columns of interest are in an index) are
the last significant optimizations missing from PostgreSQL to allow it
to outperform our previous DBMS on *all* queries, rather than just the
vast majority of them. :-)
As far as our production queries go, based on our experience with
several other products against this schema, this is the area where the
biggest performance gains remain to be realized.
-Kevin