[HEADS-UP] systemd for F14 - the next steps

[Simo Sorce]

On Fri, 23 Jul 2010 02:43:45 +0200
Lennart Poettering <mzerqung at 0pointer.de> wrote:

> On Thu, 22.07.10 20:40, Lennart Poettering (mzerqung at 0pointer.de)
> wrote:
> 
> > 
> > On Thu, 22.07.10 08:05, Simo Sorce (ssorce at redhat.com) wrote:
> > 
> > > > to make real; give reality to (a hope, fear, plan, etc.).
> > > >  
> > > > but its seems quite an abstract term to associate reality with
> > > > an abstract computer object.
> > > 
> > > Dave, I am not a native speaker, but I have the exact (or may be
> > > even worse) problem. For as much as I try the syntax there is so
> > > obscure I cannot "realize" what it means *at all*, just by
> > > looking at it.
> > > 
> > > 
> > > Lennart, "realize" really is a bad bad bad choice, please consider
> > > changing it while there is still time.
> > 
> > Kay and I have discussed this now. We agreed to fold systemd-install
> > into systemctl entirely, and replace --realize by --now. Also, we'll
> > drop some of the options --realize had, and always imply that the
> > init system configuration shall be reloaded after all changes took
> > place. This basically means that this
> > is what will be done in %post in the general case:
> 
> Kay and I discussed that even further now and decided to scrap --now
> entirely now. There's little reason left to keep this as flag around.
> I think it is more descriptive to do this on %preun:
> 
>     systemctl stop postfix.service
>     systemctl disable postfix.service
> 
> then it would be to do this:
> 
>     systemctl disable --now postfix.service.
> 
> Especially for "enable" this is even more the case, since depending
> on the case you might or might not want to restart, or reload the
> service on package upgrade. For example, restarting D-Bus or the
> gettys on package uprgade would be a really bad idea, but restarting
> ntpd might be a good idea. Given that packagers would have to
> specifiy anyway whether they want to reload, or restart or nothing at
> all a service on upgrade we think it would make more sense if they
> simply delcare their choice explicitly and do either one fo this in
> %post:
> 
>     systemctl enable foobar.service
>     systemctl try-restart foobar.service     ### restart if running
> 
> or:
> 
>     systemctl enable foobar.service
>     systemctl reload foobar.service          ### reload if running
> 
> or just:
> 
>     systemctl enable foobar.service
> 
> or, for debian folks which want to start services after package
> installation:
> 
>     systemctl enable foobar.service
>     systemctl restart foobar.service         ### restart if runnning,
> start if not running
> 
> I think this scheme is really simply now, as the operations issued are
> first class commands, and no switches necessary. Also, the verbs here
> are 1:1 from the LSB specs, and hence should offer no surprises to
> anybody.
> 
> Everybody happy?

I am personally quite happy, this looks much much better.

Juyst one very minor nitpick, why 
 systemctl enable foo.service
instead of
 systemctl enable service foo ?

And of course
 systemctl enable socket bar
etc...

Simo.

-- 
Simo Sorce * Red Hat, Inc * New York