- signals (SIGTERM)

 - service-started, service-running, and service-complete
    - actually more like: stopped, starting, started, running, failed, and stopping.
    - ... but with `complete` added, to mirror `failed`.
    - service controls: reload?, restart, pause, unpause, force-stop ("blocked")
    - difference between "pause" and "block" is that "pause" stops the
      actual program instance but doesn't flag the service as
      non-started, while "block" pushes it all the way back to
      "stopped".

 - deal with logging
    - when syslogd is running, /dev/log is a unix datagram socket

 - service for `apk` operations - otherwise they clash

Service lifecycle (as freely imagined; really, a state machine similar
to this will be induced by assertion-driven
subconversations/facets...):

    digraph G {
        compound=true;

        subgraph cluster_runnable {
            label="Runnable";
            running [label="running\n(a.k.a. ready)"];
            starting -> started [label="deps satisfied\nstart svc"];
            started -> running [label="marked running\n-"];
            started -> failed [label="start failed\n-"];
            started -> complete [label="svc done\n-"];
            running -> failed [label="svc failed\n-"];
            running -> complete [label="svc done\n-"];
            failed -> started [label="restart\nstart svc"];
            complete -> started [label="restart\nstart svc"];
            running -> started [label="restart\nstart svc"];
        }
        stopped;
        paused [label="running\n(paused)"];
        blocked [label="stopped\n(blocked)"];

        stopped -> starting [label="required\nscan deps"];
        running -> stopped [label="unrequired\nstop svc"];

        running -> paused [lhead=cluster_runnable, label="pause req\nstop svc"];
        paused -> starting [label="unpause req\nscan deps"];

        running -> blocked [lhead=cluster_runnable, label="block req\nstop svc"];
        blocked -> stopped [label="unblock req\ncheck for requirement"];

        paused -> blocked [label="block req\n-"];
    }

Oh, how about having a set-of-states approach: one for the
supervision/dependency-resolution process, and one for the service
itself?