[Date Prev][Date Next] [Thread Prev][Thread Next] [Date Index] [Thread Index]

Re: piece of mind

hi matthias,

On Tue, Oct 21, 2014 at 05:39:49PM +0200, Matthias Klumpp wrote:
> Did you play around with systemd already?

not as much as would be ideal, but I have been running it on one
machine, adapted a few things that I run for starting, and trieid the
monitoring/restart. but really, my main concerns are fueled by pages
like this [0].

> Well, you just now described systemd. The systemd project develops
> many small tools which do one thing ant interact together via defined
> interfaces.
> The init daemon is a bit more powerful that sysvinit, but it still
> only does what it is supposed to do: Starting, stopping and monitoring
> services. The other tools under the systemd umbrella do different

ok, the way you put this sounds great, but let me contrast that with
the originall paragraph:

On Tue, Oct 21, 2014 at 04:35:20PM +0200, Matthias Urlichs wrote:
> [...]
> first place. Having ten processes responsible for bits&pieces of what
> systemd-as-PID1 does instead of one isn't a benefit -- not if all you
> gain by that is nine additional processes.

I don't quite see how these two paragraph line up. to me these seem to
be contradicting a bit...

The other point is what you expect from the interfaces between
components. In the paragraph towards the top you just say "defined", but
I am sure your quality requirements for an interface of that type are
higher than that. I would argue that the most important aspect is
compatibility across large range of versions of the interface, and a
looseness that allows either end of the interface to be (partially)
reimplemented in a different way. so for me "POSIX" is a well-defined,
stable interface (not without problems, mind you). "DBUS" means next to

There are a few symptoms that make me think that while an
interface may actually be defined in some way, it is so specific that
there really can only be a single implementation. In other words, what
should really be an optional part now becomes a mandatory part of the
implementation of the interface. For example, one of the core complaints
about systemd is the fact that it is linux-only. I understand that
cgroups is a nice facility, but if an init system can't start a process
without it, then it simply is far too specific (for my taste). Equally,
if my window manager has such specific and absolute requirements that
there is only a single init system that can start it, then something is
terribly amiss. it's just another process after all!

but note that I have nothing against anyone using systemd, or it being
supported in debian! what kinda rubs me the wrong way is intreoduction
of something new and shiny with the expectation that it is now everyone
else's job to make sure that the existing still works. The GR just says
that your package (with the exceptions) must not be specific to an init 
system. That seems to be a no-brainer to me, of course a normal package
must work independently of how or who started it! 

regards  robert

[0] http://www.freedesktop.org/wiki/Software/systemd/InterfaceStabilityPromise/

Robert Lemmen                               http://www.semistable.com 

Attachment: signature.asc
Description: Digital signature

Reply to: