A green backup job is reassuring, but I want to know what happens when somebody actually needs the data back. Can we restore it somewhere clean? Can the application use it? How long does that take? Those are much more useful questions than whether a file arrived in a bucket last night.
For PostgreSQL, the SQL dump documentation makes an easily missed distinction: a dump of one database does not include cluster-wide roles and tablespaces. Those need separate handling. A database can be backed up successfully while the instructions for recreating its surroundings are still incomplete.
I would start a restore exercise with an isolated destination and the access somebody would actually have during recovery. Restoring into a long-running development environment can hide missing setup because the required role, extension or configuration is already there. Keep the restored application away from real email, payment and background-job integrations as well. Recovery practice should not send yesterday’s notifications again.
Then check behaviour. Pick a few important records and relationships, run the relevant application reads and try a representative write in that isolated environment. A server accepting connections is a useful first check, but it does not tell you whether an attachment still exists in object storage or whether the application can decrypt a value it needs. The recovery scope has to include those dependencies.
Measure the whole process, including finding the backup, getting access, transferring it, restoring it and checking the result. If the business expects an hour and the transfer alone takes three, that is a planning problem worth discovering on a quiet afternoon. Check the age of the recovered data too. Getting the service running again and recovering recent work are separate requirements.
The useful output is a short recovery procedure somebody else can follow, with a recorded restore time and any gaps made explicit. Repeat the exercise when the storage, encryption or deployment setup changes enough to affect it. I would rather find one missing permission during practice than discover it while everyone is waiting for the service to come back.