On Thu, Jun 30, 2022 at 08:57:04AM -0400, Isaac Morland wrote: > I was going to point out that pg_database_owner is the same way, but it is > fundamentally different in that it has no special allowed access and is > meant to be the target of permission grants rather than being granted to > other roles. > > +1 to rename it to pg_checkpoint or to some similar name.
We are still in beta, so, FWIW, I am fine to adjust this name even if it means an extra catversion bump.
"checkpoint" is not a verb (right?), so would something like "pg_perform_checkpoint" rather than "pg_checkpoint" fit better in the larger picture?
I would argue it’s OK. In the Postgres context, I can imagine someone saying they’re going to checkpoint the database, and the actual command is just CHECKPOINT. Changing from checkpointer to checkpoint means that we’re describing the action rather than what a role member is.
If we are going to put a more standard verb in there, I would use execute rather than perform, because that is what the documentation says members of this role can do — “Allow executing the CHECKPOINT command”. Zooming out a little, I think we normally talk about executing commands rather than performing them, so this is consistent with those other uses; otherwise we should reconsider what the documentation itself says to match other commands that we talk about running.
OK, I think I’ve bikeshedded enough. I’m just happy to have all these new roles to avoid handing out full superuser access routinely.