> As I understand it, partitioning will certainly lead to some significant
> changes/enhancements to the planner. Do you think it is realistic to get
> that for 8.5?
I don't think that waiting for our plans for a more robust partitioning
implementation is a reason to put off improvements to the implementation
we have. If Simon or Pavel or someone was going all-out on getting the
new partitioning ready, that would be one thing. But to date, nobody
has volunteered to work on it; we just know we need it.
Of course, that completely leaves aside Tom's critique of the
implementation, which sounds like it needs some work. Trying to fit the
target table into a range partitioning mold would break with a lot of
real partitionings; for example I have several client DBs which are
partitioned active/inactive-by-date.
What about simply eliminating joins between partitioned tables by
checking which columns' constraints match exactly or are subsets? A lot
of partitioned DBs partition everything by month, and joining two tables
which were partitioned by month would be useful by itself.
--
Josh Berkus
PostgreSQL Experts Inc.
www.pgexperts.com
