Excerpts from Noah Misch's message of vie ene 20 22:33:30 -0300 2012:
> On Fri, Jan 20, 2012 at 07:03:22PM -0500, Jaime Casanova wrote:
> > On Wed, Jan 18, 2012 at 7:01 PM, Noah Misch <noah@leadboat.com> wrote:
> > > On Wed, Jan 18, 2012 at 09:46:20AM -0500, Jaime Casanova wrote:
> > >>
> > >> ignoring all non-leaf pages still gives a considerable difference
> > >> between pgstattuple and relation_free_space()
> > >
> > > pgstattuple() counts the single B-tree meta page as always-full, while
> > > relation_free_space() skips it for all purposes. ?For tiny indexes, that can
> > > shift the percentage dramatically.
> > >
> >
> > ok, i will reformulate the question. why is fine ignoring non-leaf
> > pages but is not fine to ignore the meta page?
>
> pgstattuple() figures the free_percent by adding up all space available to
> hold tuples and dividing that by the simple size of the relation. Non-leaf
> pages and the meta page get identical treatment: both never hold tuples, so
> they do not contribute to the free space.
Hm. Leaf pages hold as much tuples as non-leaf pages, no? I mean
for each page element there's a value and a CTID. In non-leaf those
CTIDs point to other index pages, one level down the tree; in leaf pages
they point to the heap.
The metapage is special in that it is not used to store any user data,
just a pointer to the root page.
--
Álvaro Herrera <alvherre@commandprompt.com>
The PostgreSQL Company - Command Prompt, Inc.
PostgreSQL Replication, Consulting, Custom Development, 24x7 support