Noted that to use autovacuum we need to turn on stats_row_level (along
with stats_start_collector that is on by default). Since
stats_row_level is off by default, I wonder what sort of overhead is
incurred since it sounds like it could add up if it's storing additional
stats information on every row update, though it's not clear to me just
what it means to have row level stats.
We currently use cron to run vacuum and analyze daily in the early
morning. It's never been an issue, but the databases are getting
bigger and we wonder if autovacuum is a good addition to our operations.
Are the default values for autovacuum generally good enough for most
deployments? The various parameters are a bit complicated to
understand, so tweaking them is something we're naturally concerned
about doing.
Will autovacuum running allow regular vacuum and analyze commands to run
faster? Can it replace them entirely, or do we still need to run them
from time to time?
Can autovacuum be configured to run in a backup server that is in
"recovery" mode handling pg_standby WAL file updates?
Thanks,
David