Which is more or less what CoreOS's `fleet` already does on top of systemd. They used a different protocol for remote communications (etcd, ie. HTTP+JSON) vs. local, which I guess can be a sensible choice as requirements are quite different.
Yes, and they had to create an ad-hoc layer over systemd using a particular K-V store. Contrast to the more generic and seamless option of RPC interfaces, which are more versatile in making policy decisions.
Which is more or less what CoreOS's `fleet` already does on top of systemd. They used a different protocol for remote communications (etcd, ie. HTTP+JSON) vs. local, which I guess can be a sensible choice as requirements are quite different.