On Fri, 2009-03-20 at 09:33 -0600, Scott Ribe wrote:
> > Hopefully we regard it as a missing feature rather than as a separate
> > definition. We could cluster the index, we just don't, yet.
>
> Wouldn't this require keeping around multiple versions of index pages for
> MVCC?
No, it wouldn't require that.
Just think one index tuple points to more than one heap row. We would
still need to check visibility on the rows returned to ensure MVCC.
Less index pointers, smaller index.
The trick is: How? But that's a secondary issue to getting it on the
TODO list, which is all I'm suggesting at present.
--
Simon Riggs www.2ndQuadrant.com
PostgreSQL Training, Services and Support
