Martin Lesser <ml-pgsql@bettercom.de> writes:
> the time needed for a daily VACUUM on a table with about 28 mio records
> increases from day to day.
My guess is that the original timings were artificially low because the
indexes were in nearly perfect physical order, and as that condition
degrades over time, it takes longer for VACUUM to scan them. If that's
the right theory, the runtime should level off soon, and maybe you don't
need to do anything. You could REINDEX periodically but I think the
time taken to do that would probably be more than you want to spend
(especially since REINDEX locks out writes where VACUUM does not).
You should check that your FSM settings are large enough, but given that
the table itself doesn't seem to be bloating, that's probably not the
issue.
regards, tom lane
