Обсуждение: BUG #5322: Time to perform vacuums

The following bug has been logged online:

Bug reference:      5322
Logged by:          Eric Pailleau
Email address:      eric@numlog.fr
PostgreSQL version: 8.2.3
Operating system:   linux debian
Description:        Time to perform vacuums
Details:

Hello,
I really don't know if it can be a bug or not,
but when I do a 'VACCUM FULL ANALYZE VERBOSE',
it can take a very long time (expecially on large tables), while doing this
sequence of 3 commands is
quite quicker (on my system at least).

VACCUM VERBOSE
then
VACCUM FULL VERBOSE
then
VACCUM FULL ANALYZE VERBOSE

I mean adding the 'three commands' times is less than the time for the
direct 'VACCUM FULL ANALYZE VERBOSE'.

Does the fact of doing a 'simple' VACUUM first, make
other VACCUM more quick ? Does it finally does the same ? Is it only
coincidence due to system load ?

Thanks for your comments about this...

On Thu, Feb 11, 2010 at 11:25 AM, Robert Haas <robertmhaas@gmail.com> wrote:
> I can see VACUUM making VACUUM FULL faster. =A0I don't think VACUUM FULL
> should make VACUUM FULL ANALYZE faster.

Err, let me correct myself.  A second VACUUM FULL should be faster
than the first one.  But the two together I wouldn't expect to be
faster than doing it all in one shot.

...Robert

On Wed, Feb 10, 2010 at 8:56 AM, Eric Pailleau <eric@numlog.fr> wrote:
>
> The following bug has been logged online:
>
> Bug reference: =A0 =A0 =A05322
> Logged by: =A0 =A0 =A0 =A0 =A0Eric Pailleau
> Email address: =A0 =A0 =A0eric@numlog.fr
> PostgreSQL version: 8.2.3
> Operating system: =A0 linux debian
> Description: =A0 =A0 =A0 =A0Time to perform vacuums
> Details:
>
> Hello,
> I really don't know if it can be a bug or not,
> but when I do a 'VACCUM FULL ANALYZE VERBOSE',
> it can take a very long time (expecially on large tables), while doing th=
is
> sequence of 3 commands is
> quite quicker (on my system at least).
>
> VACCUM VERBOSE
> then
> VACCUM FULL VERBOSE
> then
> VACCUM FULL ANALYZE VERBOSE
>
> I mean adding the 'three commands' times is less than the time for the
> direct 'VACCUM FULL ANALYZE VERBOSE'.
>
> Does the fact of doing a 'simple' VACUUM first, make
> other VACCUM more quick ? Does it finally does the same ? Is it only
> coincidence due to system load ?
>
> Thanks for your comments about this...

I can see VACUUM making VACUUM FULL faster.  I don't think VACUUM FULL
should make VACUUM FULL ANALYZE faster.

It's a known problem that VACUUM FULL is really slow.  CLUSTER is
usually a better alternative; and in the next major release of
PostgreSQL VACUUM FULL will switch over to using approximately the
same method that CLUSTER now does.

...Robert

Robert Haas a =E9crit :
> On Thu, Feb 11, 2010 at 11:25 AM, Robert Haas <robertmhaas@gmail.com> wro=
te:
>> I can see VACUUM making VACUUM FULL faster.  I don't think VACUUM FULL
>> should make VACUUM FULL ANALYZE faster.
>=20
> Err, let me correct myself.  A second VACUUM FULL should be faster
> than the first one.  But the two together I wouldn't expect to be
> faster than doing it all in one shot.
>=20
> ...Robert
Thank you Robert for your comments.

What seems to be a bug, could be a feature ?
May be performing a simple VACUUM first before a real VACUUM FULL
can be a solution to increase VACUUM FULL performances ?

(I mean VACUUM FULL could be done in 2 sequences , first a simple VACUUM th=
en a real VACUUM FULL).

The time difference is sometime 1 for 10 ! But it should be confirmed on ot=
her platforms and OS ...

Thanks anyway, I will stay with my three commands sequence, waiting for the=
 new CLUSTER command...

Postgreqlement .

--=20
Salutations - Best regards - mit freundlichen Gr=FCssen