Changing the configuration of a production server is dangerous.
A safety measure might involve taking a full system image snapshot prior to the change, changing the configuration and then, in the event of a severe problem, restoring from the snapshot.
However, during the interval the new configuration is being tested, various services might run that change the state in ways that should not be lost in the event or a restore. The number of different services running makes it difficult to journal each of them for restoration after the snapshot restore.
What is the best practice to handle this treacherous interval?