Re: GLOBAL vs LOCAL temp tables

On Tue, Apr 15, 2003 at 11:17:59PM -0400, Bruce Momjian wrote:
> Alvaro Herrera wrote:

> > This is fixed in 7.4 already.  It wasn't a problem with temp tables, but
> > with btree indexes.
> 
> Yes, it is fixed partly, but I want to point out that the fix somewhat
> asymetric.
> 
> For example, it would be nice to tell people that they should either use
> VACUUM several times a day _or_ run VACUUM FULL nightly.  The problem
> with this simplification is indexes --- VACUUM records free indx pages,
> while VACUUM FULL doesn't do anything with empty index pages.
> 
> Is there anything we can do to improve this situation?  Should VACUUM
> FULL record free index pages?

Maybe VACUUM FULL could reorder index pages and truncate to recover disk
space.  But this is very costly, so there could be a "VACUUM REALLY
FULL" or something :-)

I think VACUUM FULL should at least record free index pages just like
VACUUM.  I thought it did that.

Anyway I think the "right" solution would be to integrate the vacuum
daemon in the backend.  Isn't this being worked on?  Maybe this daemon
could not only vacuum and analyze automatically when needed, but also
reorder btree pages and truncate the index file when the server load is
low.

-- 
Alvaro Herrera (<alvherre[a]dcc.uchile.cl>)
"Entristecido, Wutra
echa a Freyr a rodar
y a nosotros al mar" (cancion de Las Barreras)