Re: [HACKERS] mdnblocks is an amazing time sink in huge relations

Tom Lane wrote:
> 
> I think the main problem is that relpages and reltuples shouldn't
> be kept in pg_class columns at all, because they need to have
> very different update behavior from the other pg_class columns.

Yes, but is this reason to move them somewhere else?
Let's update them differently (i.e. update in-place) 
but keep in pg_class.
Should we provide read consistency for these internal-use columns?
I'm not sure.

> If we want to take reltuples seriously and try to maintain it
> on-the-fly, then I think it needs still a third behavior.  Clearly

...snip...
I agreed that there is no way to get accurate estimation for
# of rows to be seen by a query...
Well, let's keep up-to-date # of rows present in relation:
in any case a query will have to read them and this is what
we need to estimate cost of simple scans, as for costs of
joins - now way, currently(?) -:(

But please remember that there is another SCC goal -
faster catalog access...

Vadim