Re: Postgres architecture for multiple instances

Il giorno dom, 22/02/2015 alle 14.53 -0500, Tom Lane ha scritto:
[...]
> That's because the above claim is nonsense. pg_largeobject is not shared
> across databases of a cluster.
>
> You could well have collisions against large objects in the same database,
> though, if you're adding more large objects to an existing database and
> expecting to preserve their OIDs.

The problem is that when you use large objects, you have a table that
contain the OIDs or the large objects, and you need to keep the same
link table->LOB when moving the database. So, when you export the
database using pg_dump, it create an sql script that restore the db
using the same OIDs. If you run that script on any cluster, you may
possibly have the OID already used, and the import process does not
work.

Basically, you cannot use pg_dump for moving databases (that use large
objects), because there is no guarantee that your import succeed.

I normally import such dumps in new clusters, and it works.

Bye,
Giuseppe