Tom Lane wrote:
>
> Exactly. VACUUM sucks (ahem) in all ways but one: it pushes the
> maintenance costs associated with MVCC out of the foreground query code
> paths and into an asynchronous cleanup task. AFAIK we are the only DBMS
> that does it that way. Personally I believe it's a fundamentally
> superior approach --- because when you are under peak load you can defer
> the cleanup work --- but you do need to pay attention to make sure that
> the async cleanup isn't postponed too long. We're still fooling around
> with autovacuum and related tuning issues to make it work painlessly...
>
Should this paragraph be added to the FAQ here?
http://www.postgresql.org/docs/faqs.FAQ.html