[systemd-devel] Remote api for systemd via separate daemon

[Peter Paule]

While reading the news about dkr swarm and seeing what's going on at
CoreOS' fleet, a mail to the ML from 2013 came into my mind.

There was someone asking for some network API for systemd, if I remember
correctly. I think the answer was no, because you don't want systemd to be
exposed to the "outer" world. I'm not keen to expose systemd to the
network work either.

But looking at fleet there seems to be some need to start/stop
services/machines from remote. I asked myself if it does make sense to
expose a subset of systemd's functionality to the network by introducing
a small daemon - e.g. systemd-remoted - to the systemd ecosystem.

I can image, that the daemon could expose:

* Starting/Stopping of services
* Starting/Stopping of VMs/containers
* Pulling machines using machinectl pull-*
* Maybe even something about the load of node where the daemon is
  running.

The daemon could use some kind of rpc like REST+JSON or even the new kid
on the block "google's grpc" (https://github.com/grpc/grpc, HTTP2, with
bindings for c, c++, ruby, python, go, ...) to expose the API.

I would see this daemon more as ground layer for some other cluster
daemon which makes use of systemd-remoteds api.

Does this make sense to you or is this totally non-sense and should
better be solved elsewhere?

Cheers
/pp