Re: pgsql: Don't consider newly inserted tuples in nbtree VACUUM.

On Wed, Mar 10, 2021 at 4:27 PM Peter Geoghegan <pg(at)bowt(dot)ie> wrote:
> Don't consider newly inserted tuples in nbtree VACUUM.
>
> Remove the entire idea of "stale stats" within nbtree VACUUM (stop
> caring about stats involving the number of inserted tuples). Also
> remove the vacuum_cleanup_index_scale_factor GUC/param on the master
> branch (though just disable them on postgres 13).

It looks like there is an issue on buildfarm animal crake, with
XversionUpgrade-REL_11_STABLE-HEAD:

https://buildfarm.postgresql.org/cgi-bin/show_log.pl?nm=crake&dt=2021-03-11%2000%3A33%3A42

The v11 tests set the vacuum_cleanup_index_scale_factor storage
parameter in an index that is not dropped by the same test, to get
pg_dump coverage and so on (see btree_index.sql tests). The pg_upgrade
tests therefore go on to attempt to restore a DB with the parameter
set in the same index. This fails with Postgres 14/master due to my
having removed vacuum_cleanup_index_scale_factor on that branch.

Perhaps the best solution is also the simplest: revert the tests for
vacuum_cleanup_index_scale_factor storage on the backbranches
(Postgres 12 and 13) too. The tests in question don't test much.

Any input on how I should proceed here, Andrew?

--
Peter Geoghegan