On 18-10-2016 10:13, Andreas Joseph Krogh wrote:
> From time to time pg_largeobject comes up as an issue with being
> implemented as a system-catalog.
>
Did you read the archives [1]?
Yes..
> As I see it, there are 2 relevant use-cases for improving the situation:
> 1. Being able to pg_dump *without* any LOs (think of it as
> without the contents of pg_largeobject). This is very handy
> for testing/troubleshooting.
>
It could be an option (--no-blobs). The -b option has a limited use case.
Yes, it definitely should be an option to pg_dump. I guess because of pg_largeobject being a system-catalog it adds additional difficulties implementing it?
> 2. Being able to move pg_largeobject to a different tablespace
> *without* turning on system_table_mods. This is important for
> people storing LOTS of large-objects on separate
> disks (non-SSD) and hence in a different tablespace.
> Anyone willing to discuss this?
>
This was proposed a few years ago but no one cared to draft a patch.
So that why I'm re-raising the issue:-)
Having "everything in the database" adds lots of benefits, conceptually (follows tx-semantics, consistent backups etc.), however it's currently not so easy in practice.
Thanks.
--
Andreas Joseph Krogh
CTO / Partner - Visena AS
Mobile: +47 909 56 963