Обсуждение: Vacuum questions

Question on vacuuming.

When you do a vacuum <table>, are the "freed" tuples available only the
table, or to the entire db, or to the entire cluster?

The reason I'm asking is that we are getting ready to preform a major
upgrade to our application that involves adding some new fields to
almost every table and populating them.  This is causing our test
conversion db's to double in size (we are assuming that every update is
causing a delete and re-insert behind the scenes due to the populating
of the new columns).  Anyway, we are working on trying to find the
fastest way to recover the space.

We have also had one test occurrence where after the upgrade, a vacuum
full would not recover the space.  However, when we bounced the
postmaster, and then performed a vacuum full, the space was recovered.
Any ideas on what might cause this?  This happened on a newly restored
db that was converted and then vacuumed full.  There would have been no
connections to the db after the conversion.

PG 7.3.4
RH 2.1

Thanks,

Chris

"Chris Hoover" <revoohc@sermonaudio.com> writes:
> We have also had one test occurrence where after the upgrade, a vacuum
> full would not recover the space.  However, when we bounced the
> postmaster, and then performed a vacuum full, the space was recovered.
> Any ideas on what might cause this?

Most likely, you had an open transaction lurking that was old enough
that it could still "see" the deleted data, and so VACUUM couldn't
safely reclaim the data.

VACUUM's decisions about this are cluster-wide; so even though you'd
recently created the database in question, a long-running transaction
in another database under the same postmaster could still cause the
problem.

            regards, tom lane

Tom,

Thanks so much for the possible reason the vacuum fulls are not running.

But, how about the first question.  When you run a vacuum <table>, who
has access to the freed tuples?  (only the table, the db, or the entire
cluster)???

Thanks,

Chris

On Wed, 2005-03-16 at 11:55, Chris Hoover wrote:
> Tom,
>
> Thanks so much for the possible reason the vacuum fulls are not running.
>
> But, how about the first question.  When you run a vacuum <table>, who
> has access to the freed tuples?  (only the table, the db, or the entire
> cluster)???

the table