I am all for simplicity, but if that results in a service outage whenever you deploy new code, apply a security patch or reboot your VM, you need to rethink the architecture.
Why if a minute downtime on deployment doesn't impact revenue significantly ?
The frontend can even have some exponential cool down on requests when the backend is down and the users who are online in that one minute window will suffer just notice it being a bit slower.
He even mentioned his uptime in last years in the post.
