On Thu, 2020-07-09 at 12:47 -0400, Stephen Frost wrote:
> I realize this is likely to go over like a lead balloon, but the churn
> in pg_class from updating reltuples/relpages has never seemed all that
> great to me when just about everything else is so rarely changed, and
> only through some user DDL action- and I agree that it seems like those
> particular columns are more 'statistics' type of info and less info
> about the definition of the relation. Other columns that do get changed
> regularly are relfrozenxid and relminmxid. I wonder if it's possible to
> move all of those elsewhere- perhaps some to the statistics tables as
> you seem to be alluding to, and the others to $somewhereelse that is
> dedicated to tracking that information which VACUUM is primarily
> concerned with.
Perhaps we could create pg_class with a fillfactor less than 100
so we het HOT updates there.
That would be less invasive.
Yours,
Laurenz Albe