On Mon, Mar 26, 2018 at 05:33:19PM +0300, Artem Tomyuk wrote: > Can't, it generates huge IO spikes. > > But.... > > Few hours ago i manually started vacuum verbose on pg_attribute, now its > finished and i have some outputs: > > INFO: "pg_attribute": found 554728466 removable, 212058 nonremovable row > versions in 44550921 out of 49326696 pages DETAIL: 178215 dead row versions > cannot be removed yet. There were 53479 unused item pointers. 0 pages are > entirely empty. CPU 1097.53s/1949.50u sec elapsed 6337.86 sec. Query > returned successfully with no result in 01:47:3626 hours. > > what do you think? > > select count(*) on pg_attribute returns: > 158340 rows > > So as i understand vacuum full will create new pg_attribute and will wrote > those amount of "valid" rows, but still it will scan 300GB old table? > So estimate will be even ~same compering with regular vacuum?
more or less, yes.
the thing is - find and fix whatever is causing this insane churn of tables/attributes.
