On 2013-06-03 11:00:38 -0700, Josh Berkus wrote:
>
> >> Also, locking while it does its work.
> >
> > Eh?
>
> Even if we're doing lazy vacuum, we have to lock a few pages at a time
> of each table. This does result in response time delays on the current
> workload, which can be quite bad if it's a highly contended table already.
We don't really lock more pages at a time than normal DML does. 1 heap
page at a time, possibly several index pages at once.
There's something related which can cause problems which is that we
require cleanup locks on the page to be able to repair fragmentation
which makes *vacuum* wait for all clients to release their page pins.
Greetings,
Andres Freund
-- Andres Freund http://www.2ndQuadrant.com/PostgreSQL Development, 24x7 Support, Training &
Services
