On Mon, 2004-03-22 at 04:23, Karel Zak wrote:
> All. It's important do it as backend process. Because libpq has very,
> very limited and slow resources for work with backend stuff.
Agreed.
> The base should be the standard backend with different "main loop" that
> will instead socket checks some shared information about tables and
> calls directly vacuum stuff. In this case you can omit work with
> connections, parser etc.
So am I to understand I can start up a postmaster subprocess and then be
able to monitor the activity of all the databases? I guess that makes
sense since I would be talking to the stats collector directly via a
socket. But that doesn't solve the problem of issuing the vacuum to
different databases, I would still create a new backend for every
database that needs a vacuum or analyze issues.
Also, we don't want to launch multiple simultaneous vacuums so we want
the commands to be serialized (I know some people want to be able to do
this if databases are located on different disks, but for now i'm
keeping things simple). I have an idea for this that I just mentioned
in another message to the list.
> I thought about it in last days and I found perfect Tom's idea about
> FSM tables usage:
There has been lots of discusion of incorporating FSM data into the
auto_vacuum decision process. I am interested in exploring this, but
since I'm already biting off more than I can easily chew, I am going to
try and leave the decision making process the same for now.
BTW I think we need to use both tools (stats and FSM) since not all
tables will be in the FSM, an insert only table still needs to be
analyzed periodically and a lightly updated table will eventually need
to be vacuumed.