If LP doesn't see the obvious benefit of making systemd optional, I
can't see that he'd go to the trouble of creating a plug-and-play setup
to allow alternatives to systemd to get in on the act. (Would a dictator
suggest the tryout of other political systems, to see how they stacked
up against autocracy?)
I think that the thing that I found tantalising was the idea of sniffing
what systemd tried to do, and then deciding whether or not to do it, and
what responses to send back to systemd. I had a mental image of a
convicted gangster directing the mob from his or her prison cell -
he/she can give out the instructions, but is entirely dependent on what
feedback is relayed from outside the jail. ("The Taking of Pelham 123" -
they're trying to get the money to the subway station before any more
hostages get shot, and aren't going to make it. In a flash of
inspiration, Garber, the train dispatcher, realises that the crooks have
no way of knowing what's happening on the surface, so tells them the
money's arrived before it actually does).
One possible real benefit of a glue/abstraction/jail layer did occur to
me. Eventually, it would itself become an init system, with a
distillation of (hopefully) the best bits of all the init systems which
had used it.
I totally agree re. PID1 and the need for simplicity. Maybe I should
just get out more (eventually, anyway ;) )
On Wed, 2020-05-20 at 18:45 -0400, Steve Litt wrote:
> On Tue, 19 May 2020 10:29:02 +0100
> Peter Duffy <peter@???> wrote:
>
> > Apologies for following up on my own post - just an afterthought.
> >
> > When I originally encountered systemd, the word was that it was so
> > pervasive that it couldn't be removed (obviously, now we know
> > different ;) )
> >
> > Given the alleged non-optionality of systemd, I started to wonder
> > about some kind of an init system wrapper (or even jail) - an
> > abstraction layer which would sit between the init subsystem and the
> > main system, and sanitise and homogenise interactions between the
> > two; init systems, including systemd, could be plugged and unplugged
> > into the top surface as desired; the abstraction layer would manage
> > commands and responses (including lying to the init subsystem if the
> > latter tried to do something dangerous or antisocial).
>
> Oh please don't suggest this: Poettering might do it.
>
> The job of an init system (not systemd, that's a software analogy of
> those old electronic circuits encased in epoxy so nobody could reverse
> engineer them) is to:
>
> 1) Run as PID1, listening for certain
>
> 2) PID1 forks off early boot stuff to mount, unencrypt, construct
> devices, set up the network, and the like.
>
> 3) PID1 forks off a daemon handler. The best daemon handlers are, in my
> opinion, djb style process supervisors like daemontools, runit and
> s6.
>
> 4) Upon receipt of a certain signal, PID1 forks or execs the shutdown
> script.
>
> It really is just that simple. There's no need to add anything to
> accommodate badly behaved init system authors.
>
> SteveT
>
> Steve Litt
> May 2020 featured book: Troubleshooting Techniques
> of the Successful Technologist
> http://www.troubleshooters.com/techniques
> _______________________________________________
> Dng mailing list
> Dng@???
> https://mailinglists.dyne.org/cgi-bin/mailman/listinfo/dng
::
Re: [DNG] without-systemd.org not w…
