From my experience that mostly only applies to the application layer. Especially updates on distributed databases are often not as easy as "stop and rollback" a

From my experience that mostly only applies to the application layer. Especially updates on distributed databases are often not as easy as "stop and rollback" as you have to achieve a stable quorum or other similar dynamic states like that which you can't easily achieve from restoring a snapshot. There your best shot often is to learn how to deal with all the failure scenarios manually and do as many tests on a staging system until you feel comfortable.