david@lang.hm wrote:
> On Wed, 6 Jun 2007, Steinar H. Gunderson wrote:
>
>> On Tue, Jun 05, 2007 at 05:59:25PM -0400, Tom Lane wrote:
>>> I think the main argument for partitioning is when you are interested in
>>> being able to drop whole partitions cheaply.
>>
>> Wasn't there also talk about adding the ability to mark individual
>> partitions
>> as read-only, thus bypassing MVCC and allowing queries to be satisfied
>> using
>> indexes only?
>>
>> Not that I think I've seen it on the TODO... :-)
>
> now that's a very interesting idea, especially when combined with
> time-based data where the old times will never change.
That's been discussed, but it's controversial. IMHO a better way to
achieve that is to design the dead-space-map so that it can be used to
check which parts of a table are visible to everyone, and skip
visibility checks. That doesn't require any user action, and allows updates.
--
Heikki Linnakangas
EnterpriseDB http://www.enterprisedb.com