That all sounds very counterintuitive. The "fallback" amounts to a statically configured list of DNS servers, separate from the other lists of statically configured DNS servers, with the property that if no DNS servers were explicitly configured by any method the systemd developers assume DNS servers will be configured by then the system will use the DNS servers that systemd developers thought were a good idea to use. Not only is it complicated and brittle, it also is not clear why a project that intends "to create a unified architecture for all system daemons" is rewriting the rules of how DNS servers are configured. For that matter, why is this logic (which is of questionable value) in systemd instead of NetworkManager (which already deals with DNS server configuration and is user-facing and better able to inform users that the fallbacks are being used)?
