Re: [HACKERS] Re: [PORTS] vacuum takes too long

On Thu, 7 Jan 1999, Bruce Momjian wrote:

> > With MVCC an occasional 'vacuum analyze' should only be noticed from the
> > performance improvements.  As far as I can tell most of the work done by
> > an analyze is in reading the table data.  If you make sure to write the
> > new information at the end of the transaction you only lock the indexes
> > for the amount of time it takes to write them.
> >
> > I see a 'vacuum analyze' being less of a problem than 'vacuum'.
> > Any of you experts can contradict my assumptions.
>
> The problem is that vacuum analyze does both vacuum and analyze.
> Analyze takes so long, we figured we might as well vacuum too.  Maybe we
> need to change that.

There is, IMHO, no problem with them being combined...in the past, the
problem was that the whole system was effectively locked up while a vacuum
analyze was being run because one of the 'statistics' tables was being
locked during the whole thing, instead of when required...

As DeJuan points out, though, this should no longer be a problem with
MVCC...

Marc G. Fournier
Systems Administrator @ hub.org
primary: scrappy@hub.org           secondary: scrappy@{freebsd|postgresql}.org