On Wednesday, June 3, 2020, Laurenz Albe <
laurenz.albe@cybertec.at> wrote:
On Thu, 2020-06-04 at 16:41 +1200, Thomas Munro wrote:
> There's no doubt it's useful, and it's also part of the SQL spec,
> which says you can do catalog.schema.table. I would guess that we
> might get that as a byproduct of any project to make PostgreSQL
> multithreaded. That mountain moving operation will require us to get
> rid of all the global state that currently ties a whole process to one
> session and one database, and replace it with heap objects with names
> like Session and Database that can be passed around between worker
> threads.
I am -1 on cross-database queries.
I think it is a desirable feature to have databases isolated from
each other, so you don't have to worry about a permission you forgot
that allows somebody to access a different database.
I think this is particularly relevant since all databases share the
same users.
I understand that sometimes the opposite would be desirable, but
foreign data wrappers have alleviated that pain.
I agree with the conclusion but not so much with the premise. Even with global users you still need to grant permissions to individual databases and its debatable whether its “more safe” to prevent a user from directly accessing a database in the “catalog.schema” reference manner if they can do so with a direct login.
I agree with the general premise that modularity and isolation are generally positive qualities, especially as scale grows, and that expending considerable resources strictly for the goal of adding this capability to the system is not a direction that I would be in favor of. Now, if the prereqs for this feature also have other concrete benefits that are worth working toward, and in the end the sum of those makes cross-database queries a relatively simple matter, I would entertain putting in the last 10% of effort to become standard compliant.
David J.
