Re: Table space not returned to the OS ?

On Mon, Jun 27, 2022 at 11:30 AM Florents Tselai <florents(dot)tselai(at)gmail(dot)com>
wrote:

> Hi,
>
> A few months back (October) I had upgraded a Postgres instance from v12 —>
> 14.
>
> The database disk size under /var/lib/postgresql/12 was around 800GB+ back
> then.
> Note, that IIRC I had used hard-linking during the upgrade.
>
> In the database itself, lots of things have changed since.
> In fact, that database itself has been dropped at some point and restored
> from a backup.
>
> As I was running out of disk space, I started investigating and found out
> that
>
> /var/lib/postgresql/12/main/base/16385 —> 886GB+
> /var/lib/postgresql/14 —> 400GB
>
> The last modification date on that directory (../12/) appears to be around
> a month ago,
> When the table with relied 16385 was in fact dropped.
>
> Now, In my update scripts (I use this db as an OLAP) I occasionally run
> VACUUM.
>
> Is it weird that the 886GB space hasn’t been returned to the OS yet?
>
> What’s the safest way to return it to the OS manually?
>
>
When you use hardlinks in the upgrade all the files remain in the old
directory when they are removed from the new one such as when you drop a
relation. it is there for emergency recoveries. It's only the contents of
the files that's "mirrored", not the existance.

It looks like you didn't actually delete the old cluster, which you are
supposed to do once you have verified that the new one works. This looks
like a debian/ubuntu system, which means you probably forgot to run
"pg_dropcluster 12 main"? Or if it's not a debian cluster, the equivalent
of that which results in removing the data directory for 12 along with any
configuration files it has elsewhere.

--
Magnus Hagander
Me: https://www.hagander.net/ <http://www.hagander.net/>
Work: https://www.redpill-linpro.com/ <http://www.redpill-linpro.com/>