Hi Simon,
Thanks for the insight. I might take that as a long term project. I have
to discuss that with my colleagues at Aster. It would certainly help to
put together a wiki page with the key insights on the design of such
implementation to be able to better scope the project and agree on what
is the right approach for this.
manu
> On Tue, 2009-09-22 at 18:16 -0400, Emmanuel Cecchet wrote:
>
>
>> If the partitioning implementation does not make progress (and does not
>> make it for 8.5)
>>
>
> Manu, not sure if you are referring to Kedar's patch I reviewed earlier
> in July, but that patch didn't implement anything like the right
> internals for this to work.
>
> As Tom says, what we need is a single internal data structure that
> describes all the partitions. That can then be used to automatically
> route inserts, perform the push down merge joins and speed-up constraint
> exclusion. We should build this once and store it in the relcache for
> the parent relation and so would need some changes in relcache
> invalidation as well to make sure child ddl changes invalidate that.
>
>
--
Emmanuel Cecchet
Aster Data Systems
Web: http://www.asterdata.com