Goes without saying that Heroku is a great example of how it should be done -
Note: The applications in question have had zero interactions (deploys, restarts, scaling, etc.) which might have caused their dynos to be unallocated/re-provisioned within Heroku's own service.