Re: Ordered Append Node

Gregory Stark wrote:
> Ideally we would also be able to do this in the planner. If the planner could
> determine from the constraints that all the key values from each partition are
> non-overlapping and order them properly then it could generate a regular
> append node with a path order without the overhead of the run-time
> comparisons.
> 
> But that requires a) dealing with the problem of the parent table which has no
> constraints and ...

It would help even if you could only prove it for some partitions. You 
could use a regular append node for the partitions you know not to 
overlap, and merge the output of that with the new kind of ordered 
append node. We'd see that the parent table is empty on the first 
invocation, and after that the ordered append node would just pass 
through the tuples.

--   Heikki Linnakangas  EnterpriseDB   http://www.enterprisedb.com