Обсуждение: pgstattuple, vacuum and free_space

Hi:

  I did the following after installing the pgstattuple contrib code:

 select * from pgstattuple('wind');
-[ RECORD 1 ]------+----------
table_len          | 224854016
tuple_count        | 1492601
tuple_len          | 207535124
tuple_percent      | 92.3
dead_tuple_count   | 11569
dead_tuple_len     | 1361848
dead_tuple_percent | 0.61
free_space         | 3311416
free_percent       | 1.47


vacuum verbose wind;
INFO:  vacuuming "public.wind"
INFO:  index "wind_pkey" now contains 1492601 row versions in 4152 pages
DETAIL:  11569 index row versions were removed.
0 index pages have been deleted, 0 are currently reusable.
CPU 0.56s/1.77u sec elapsed 36.73 sec.
INFO:  index "wind_measurement_date_index" now contains 1492601 row
versions in 4156 pages
DETAIL:  11569 index row versions were removed.
0 index pages have been deleted, 0 are currently reusable.
CPU 0.43s/1.83u sec elapsed 35.90 sec.
INFO:  "wind": removed 11569 row versions in 371 pages
DETAIL:  CPU 0.03s/0.03u sec elapsed 2.38 sec.
INFO:  "wind": found 11569 removable, 1492601 nonremovable row versions
in 27448 pages
DETAIL:  0 dead row versions cannot be removed yet.
There were 0 unused item pointers.
0 pages are entirely empty.
CPU 2.01s/3.95u sec elapsed 79.39 sec.
VACUUM

select * from pgstattuple('wind');
-[ RECORD 1 ]------+----------
table_len          | 224854016
tuple_count        | 1492601
tuple_len          | 207535124
tuple_percent      | 92.3
dead_tuple_count   | 0
dead_tuple_len     | 0
dead_tuple_percent | 0
free_space         | 4673256
free_percent       | 2.08


My question: when you vacuum a table and generate 'free_space', who is
allowed to consume this 'free_space'? Is it released to the OS for
general use? Or is it reserved just for the database? If the latter, is
it reserved just for 'wind' (in this case)?

Thanks again!

On 11/24/05, Colton Smith <smith@skio.peachnet.edu> wrote:

as far as i know this space is mapped as free witinh data-files of table "wind".
thus effectivelyu - your OS doesnt get more free space, nor is (generally speaking) the database.
benefit of this "free space" is that next inserts to wind table will fit inside of this free space - thus stopping bloat of datafiles of this table.

if you would like to reclaim the disk-space, you should consider using vacuum full.

depesz

On Sat, Nov 26, 2005 at 11:39:22AM +0100, hubert depesz lubaczewski wrote:
> On 11/24/05, Colton Smith <smith@skio.peachnet.edu> wrote:
> >
> > My question: when you vacuum a table and generate 'free_space', who is
> > allowed to consume this 'free_space'? Is it released to the OS for
> > general use? Or is it reserved just for the database? If the latter, is
> > it reserved just for 'wind' (in this case)?
> >
>
>
> as far as i know this space is mapped as free witinh data-files of table
> "wind".
> thus effectivelyu - your OS doesnt get more free space, nor is (generally
> speaking) the database.
> benefit of this "free space" is that next inserts to wind table will fit
> inside of this free space - thus stopping bloat of datafiles of this table.

Keep in mind that the free space needs to be recorded in the Free Space
Map for it to be re-used. If your FSM is too small then some of the free
space will not be used until you vacuum again.

Also, it is possible over time for tables to shrink. Vacuum will try to
remove any pages from the end of the table that are empty. But it's
difficult to make it so that pages at the end of the table are empty,
and AFAIK indexes can never shrink, so ultimately something like a
vacuum full/reindex or a cluster are your best bets for reclaiming disk
space.
--
Jim C. Nasby, Sr. Engineering Consultant      jnasby@pervasive.com
Pervasive Software      http://pervasive.com    work: 512-231-6117
vcard: http://jim.nasby.net/pervasive.vcf       cell: 512-569-9461