Re: relational class vs partitioned table (was

On Mon, 2005-10-10 at 19:58 -0700, Trent Shipley wrote:
> Of course, there is no reason a relation in a relational class might not be 
> huge.  

Well, as a designer, I would make it so.

> Orthoganal partion rules would be created for the class.  The rules would be 
> applied to each member relation.  Finally, the rules would be applied to the 
> relevant unifying (presumably unique) indexes.
> 
> But inasmuch as Postgresql has implemented neither partitioning nor unique 
> constraints for relational classes we are getting somewhat ahead of 
> ourselves.

Maybe you aren't aware of the new constraint_exclusion feature in 8.1 ?

> Partitioning is obviously dominated by partitioning rules.  Oracle's SQL 
> dialect provides a negative example of how to elegantly incorporate 
> partitioning rules into SQL.  Ideally partitioning rules should be 
> first-class objects.  A database engineer or the poor DBA who inherits his 
> implementation should be able to query the meta-data to get a listing of all 
> partitioned relations.

The partitioning doesn't follow Oracle syntax at all. Partitions are
first class objects as you suggest.

Best Regards, Simon Riggs