On Wed, Aug 09, 2023 at 09:20 AM +0800, Michael Paquier wrote:
> This does not really explain the reason why in-place tablespaces need
> to be skipped (in short they don't need a separate creation or check
> like the others in create_script_for_old_cluster_deletion because they
> are part of the data folder). Anyway, the more I think about it, the
> less excited I get about the need to support pg_upgrade with in-place
> tablespaces, especially regarding the fact that the patch blindly
> enforces allows_in_place_tablespaces, assuming that it is OK to do so.
> So what about the case where one would want to be warned if these are
> still laying around when doing upgrades? And what's the actual use
> case for supporting that? There is something else that we could do
> here: add a pre-run check to make pg_upgrade fail gracefully if we
> find in-place tablespaces in the old cluster.
I have implemented the changes you suggested in our previous discussion. I have added the necessary code to ensure that pg_upgrade fails gracefully with in-place tablespaces and reports a hint to let the check pass.
Thank you for your guidance and support. Please review my latest patch.