"scott.marlowe" <scott.marlowe@ihs.com> writes:
> Oh, yeah I have no doubt of that. I was thinking more along the lines of
> when a transaction ends it throws a background "vacuum table1;vacuum
> table2;vacuum tablen" command into some kind of vacuuming hopper.
Actually, the plans I liked best for driving auto-vacuum were
essentially an indirect version of that: the FSM module would keep track
of committed deletes + aborted inserts for each active table, and then
the autovacuum scheduler could use that info to decide which tables are
highest-priority to vacuum.
(Or possibly the runtime stats module would be a better place to track
it than FSM.)
regards, tom lane