Quoting Tom Lane <tgl@sss.pgh.pa.us>:
> "Mindaugas Riauba" <mind@bi.lt> writes:
> > ... So contents of database changes very fast. Problem is that
> when
> > pg_autovacuum does vacuum those changes slows down too much.
>
> The "vacuum cost" parameters can be adjusted to make vacuums fired
> by pg_autovacuum less of a burden. I haven't got any specific
> numbers
> to suggest, but perhaps someone else does.
I solved one problem by cranking sleep scaling to -S 20.
It made pg_autovacuum back off longer during extended periods of heavy
disk-intensive query activity. Our update activity is near-constant
insert rate, then once or twice a day, massive deletes.
--
Dreams come true, not free.