On Thu, Apr 11, 2019 at 6:13 AM Michael Paquier <michael@paquier.xyz> wrote:
Hi all,
Recent commit bfc80683 has added some documentation in pg_rewind about the fact that it is possible to do the operation with a non-superuser, assuming that this role has sufficient grant rights to execute the functions used by pg_rewind.
Attached is a patch which switches all the TAP tests of pg_rewind to do that. As of now, the tests depend on a superuser for everything, and it seems to me that it makes little sense to make the tests more pluggable by being able to switch the roles used on-the-fly (the invocation of pg_rewind is stuck into RewindTest.pm) as a superuser has no restrictions.
Any thoughts?
+1.
I definitely think having tests for this is good, otherwise we'll just end up making a change at some point that then suddenly breaks it and we won't notice.
If we haven't already (and knowing you it wouldn't surprise me if you had :P), we should probably look through the rest of the tests to see if we have other similar cases. In general I think any case where "can be run by non-superuser with specific permissions or a superuser" is the case, we should be testing it with the "non-superuser with permissions". Because, well, superusers will never have permission problems (and they will both test the functionality).
I do think it's perfectly reasonable to have that hardcoded in the RewindTest.pm module. It doesn't have to be pluggable.