No, I ran three-node etcd clusters for years that Kops set up. Kops deployed an etcd-operator to babysit them and take care of backups and the like. It was set-and-forget. We had controlled disruptions all the time as part of Kubernetes control plane updates, no issues.

And you know... etcd supports five-node clusters, precisely to help support people who are paranoid about extended single node failure.