Their infrastructure is meant to all be HA from memory, and unless you're doing stupid things like caddy did[1] small windows of downtime shouldn't cause much issue for ongoing operations anyway - you will likely be starting to attempt renewing certs weeks before they expire.

1: https://github.com/mholt/caddy/issues/1680