Date: Mon, 6 Feb 2006 10:35:24 -0700
From: Michael Fuhr <mike@fuhr.org>
To: Markus Schaber <schabi@logix-tt.com>
Cc: Alvaro Herrera <alvherre@commandprompt.com>,
	PostgreSQL SQL List <pgsql-sql@postgresql.org>
Subject: Re: no notnull values, invalid stats?
Message-ID: <20060206173523.GA16445@winnie.fuhr.org>
References: <43E739CF.6030908@logix-tt.com> <20060206140227.GA5765@surnet.cl>
	<43E75E00.8050105@logix-tt.com>
Mime-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <43E75E00.8050105@logix-tt.com>

On Mon, Feb 06, 2006 at 03:32:32PM +0100, Markus Schaber wrote:
> Alvaro Herrera wrote:
> >>The following message occasionally appears in my postgresql log - from
> >>temporal corellation, it might be caused by autovacuum.
> >>
> >>NOTICE:   no notnull values, invalid stats
> > 
> > I see no such message in 8.1 sources.  Is this verbatim or did you
> > translate it?
> 
> It is verbatim from /var/log/postgresql/postgresql-8.1-main.log.
> 
> But I have PostGIS installed in some of the databases, so it might be
> from there.

That is indeed a PostGIS message; it's in compute_geometry_stats()
in lwgeom/lwgeom_estimate.c.  I think it means that during an
analyze, all of the sampled rows had NULL values in their geometry
columns (i.e., no not-NULL values were found); that would explain
the correlation with autovacuum.  Here's an example that elicits
the notice:

postgis=> CREATE TABLE foo ();
CREATE TABLE
postgis=> SELECT AddGeometryColumn('foo', 'geom', -1, 'GEOMETRY', 2);
               addgeometrycolumn                
------------------------------------------------
 public.foo.geom SRID:-1 TYPE:GEOMETRY DIMS:2
 
(1 row)

postgis=> INSERT INTO foo VALUES (NULL);
INSERT 0 1
postgis=> ANALYZE foo;
NOTICE:   no notnull values, invalid stats
ANALYZE
postgis=> UPDATE foo SET geom = GeomFromText('POINT(0 0)');
UPDATE 1
postgis=> ANALYZE foo;
ANALYZE

-- 
Michael Fuhr