I used to think so as well, but not any more. Any non-trivial service needs persistence, so it must checkpoint its state periodically. Also, the service is dist

I used to think so as well, but not any more. Any non-trivial service needs persistence, so it must checkpoint its state periodically. Also, the service is distributed to prevent a server from being a single point of failure. In this age of lightweight containers and microservices, restarting the program on an upgrade seems to me a perfectly reasonable alternative.