On 8/11/21 2:29 PM, Peter Geoghegan wrote:
> On Wed, Aug 11, 2021 at 11:23 AM Jonathan S. Katz <jkatz@postgresql.org> wrote:
>> How about:
>>
>> * `pg_upgrade` now carries forward the old installation's `oldestXID`
>> value, which can improve things from a performance standpoint by no
>> longer forcing an anti-wraparound `VACUUM`.
>
> I don't think that framing this as a performance thing really makes
> sense.
I had grabbed the performance bit from the release notes (though the
comment was "[t]hat's not desirable from a performance standpoint.").
It certainly helps performance to not do something that's
> totally unnecessary, and only ever happened because of a bug in the
> implementation. But to me the point is that we're not doing these
> weird wholly unnecessary antiwraparound VACUUMs on upgrade now.
> Running pg_upgrade no longer affects when or how we VACUUM, which is
> exactly what you'd expect all along.
So perhaps:
"* `pg_upgrade` now carries forward the old installation's `oldestXID`
value and no longer forces an anti-wraparound `VACUUM`."
or maybe even:
"* `pg_upgrade` no longer forces an anti-wraparound `VACUUM`."
Jonathan
