On Wed, Mar 18, 2020 at 1:40 PM Michael Lewis <mle...@entrata.com> wrote: > > Do you have default fillfactor set on this table? If not, I would wonder if > reducing it to 50% or even 20% would allow many more HOT updates that would > reduce bloat.
I don't believe we have a default fillfactor, but I'm still trying to understand why autovacs would completely stop -- that seems like a bug. Especially since there was no change between PG10 and PG12 and this problem never existed there. > Also, is there any period of lower activity on your system that you could > schedule a vacuum freeze for daily or weekly? I believe having frozen pages > would also mean all the autovacuums would be able to skip more pages and > therefore be faster. > > >> autovacuum_vacuum_cost_delay = 20 > > This was changed to 2ms in PG12. You should reduce that most likely. Actually, we set that back from 2ms > 20ms to replicate what PG10 was doing just in case it had an impact (spoiler: it did not).
