As someone who has run nodes, no it is hard and expensive. Every time a geth node dies it has to resync and no persistent volume mounts and stateful sets are not solutions. They are problems. If you need to scale horizontally you get strange consistency issues with the API. All of this makes for a very unpleasant experience. It's built for TLC on a beefy box not a herd.
And that's the rub. The new node doesn't have the same state as the old one. So clients making requests assuming that latest is the same start having problems. If you haven't seen them you just haven't been running a production quality service.
