Hey Tom,
Thanks for the response. I agree that this implementation looks shady, and I would rather have postgres to disallow role drops in such situations. Currently, the same behaviour can be observed even without the "proxy" role that I used in my example. I think that would be consistent with the way SQLServer/Oracle deals with it - not allowing the context to be removed while in use.
Kirill.
PG Bug reporting form <noreply@postgresql.org> writes:
> We're using temporary credentials to access the DB. The temporary
> credentials are granted a role and automatically become that role on
> connect. As soon as temporary credential is removed while the connection is
> still alive, any regular query would continue to work, however, a parallel
> query would cause an error:
> ERROR: role with OID XXXXXX does not exist
> CONTEXT: parallel worker
I don't think this is particularly a bug. You are relying on an
undocumented and un-guaranteed implementation artifact that it's
possible to drop a role at all while it's still in use in some
session. If we were to make any effort in this area, the end result
would certainly be that the DROP ROLE would be rejected, not that the
parallel query case would start to work. One reason why we'd want
to go in that direction is that there are almost certainly many other
cases where such an "orphaned" session misbehaves.
(Doing that has in fact been discussed, but nothing's been done
about it so far AFAIK.)
regards, tom lane
