Hi all,
I have a database with a number of tables that are partitioned monthly,
after that the tables are mostly read only (on rare occasions we may
delete from a table but normally we just drop the partitions). Recently
I've noticed that we have a lot of these tables are vacuumed around the
same time, after a little big of digging I've realised that postgres is
vacuuming them to stop xaction wrap around. So for example in a few
million xactions (later today) postgres is going to want to vacuum 37
tables for just this reason.
I know I can fiddle autovacuum_freeze_max_age and vacuum_freeze_min_age
to change how regularly the tables have this occur, and I can do this on
a per table basis in pg_autovacuum (yes this means I'm running an old
version, version 8.2) but what I'm wondering is how other people are
breaking this synchronisation?
Should I add a random value to the freeze_max_age for all the old tables
when I start a new month? Or do the same with the freeze_min_age?
Perhaps I should just force a vacuum on some of the tables the break it?
Cheers,
--
Michael Graham <mgraham@bloxx.com>