Re: Linked data from upgrade after VACUUM FULL not deleted.

On Sat, May 11, 2019 at 07:04:32PM +0200, Josef Šimánek wrote:
> Hello.
>
> Few months ago we did successful pg_upgrade --link from 9.6 to 10. I did a
> VACUUM FULL of all database this weekend and data from 9.6 directory were not
> released.
>
> I have tablespace with only one database.
>
> PostgreSQL 10 folder has similar size to actual database size (by https://
> wiki.postgresql.org/wiki/Disk_Usage it is 660GB):
> 681G    ./postgres_raid_tablespace/PG_10_201707211
>
> PostgreSQL 9.6 folder still has a lot of data in there:
> 547G    ./postgres_raid_tablespace/PG_9.6_201608131
>
> I did a really naive check if any links are still present:
>
> ls -la . | grep "\->"

They are _hard_ links and do not show up as ->, but as the _link_ count,
which is the second column of ls -la.

> And nothing was found.
>
> I have checked also which files are waiting to be deleted via lsof -u postgres,
> but nothing was found in those directories.
>
> I'm wondering if I understand this well and this is expected state where old
> folder is not touched anymore and that's the reason why VACUUM FULL is not
> releasing space in there.
>
> Is there any way how to check if anything in that folder is really not used
> anymore and consider that safe to delete?
>
> I'll appreciate any suggestions.

They is an output line of pg_upgrade which says:

Running this script will delete the old cluster's data files:

In summary, you can delete the old data directory.

--
Bruce Momjian <bruce(at)momjian(dot)us> http://momjian.us
EnterpriseDB http://enterprisedb.com

+ As you are, so once was I. As I am, so you will be. +
+ Ancient Roman grave inscription +