Robert Haas <robertmhaas(at)gmail(dot)com> wrote:
> I'm not sure why we're so glibly rejecting Dan's original
> proposal. Sure, adjusting pg_upgrade when we whack around
> pg_statistic is work, but who ever said that a workable in-place
> upgrade facility would be maintenance-free? We're operating under
> a number of restrictions imposed by the need to be pg_upgrade-
> compatible, and this doesn't strike me as a particularly severe
> one by comparison -- we can always arrange to NOT migrate
> statistics between incompatible versions; that doesn't mean that
> we shouldn't migrate them when they ARE compatible. Also, unlike
> the alternatives thus far proposed, Dan's idea actually fixes the
> problem.
In case it got lost with my various timings, I agree with Robert on
all of the above. The three-minute downtime for pg_upgrade to
upgrade our multi-TB databases is *very* impressive; but I think we
lose bragging rights if we follow that up with -- oh, but the
database isn't really fully *usable* until you run a one-hour
analyze afterward.
-Kevin