Re: systemd

[Tom H <tomh0665@gmail.com>]

On Tue, Jul 31, 2012 at 7:20 AM, Roger Leigh <rleigh@codelibre.net> wrote:
> On Sun, Jul 29, 2012 at 02:05:15PM -0400, Tom H wrote:
>> On Tue, Jul 24, 2012 at 10:48 AM, Camaleón <noelamac@gmail.com> wrote:


>> I'd *guess* [...] there are hardly more problems with
>> systemd as there are problems with sysvinit on squeeze.
>
> Without any objective evidence, who knows?  systemd is hardly the
> only contender as a sysvinit replacement.  There's upstart, openrc
> and others.  Which of those is best suited to being the default?

It's *pure*speculation* on my part that had wheezy moved to systemd a
year ago, systemd would be in an even better shape that it is in now
and would be a good default init system for wheezy's upcoming release.
Nothing more.

I know that systemd's not the only candidate. In fact, purely from a
familiarity perspective, I prefer upstart to systemd because it's
closer to sysvinit. From a technical perspective, systemd seems to be
more comprehensive and better thought out than upstart. AFAIU, OpenRC
requires some development to qualify as a sysvinit replacement. You
(Roger) made what seemed like a call for OpenRC testers in a earlier
post to this thread to which I meant to reply.

Technical considerations aside, my pie-in-the-sky wish would've been
for Solaris' init system to have an open source license and to be used
by all Linux distributions - or at least those that I use...


>> Had Debian decided a year ago to transition to systemd (not taking
>> into account the difficulty of surmounting the opposition of the
>> "conservatives" who don't want to transition to anything, of the
>> kfreebsd and hurd ports, of the anti-RH crowd, of the anti-Poettering
>> crowd, etc), would it have been in a good enough state for a wheezy
>> release? Yes IMHO (especially since there'd be testing/unstable
>> testers and reporters to add to the Fedora users), certainly no for
>> many others... Just because it was first released as a distribution
>> default 14/15 months ago doesn't mean that it's hopelessly buggy.
>
> I don't think that anyone at all claims that it's "hopelessly buggy".
> Not having had the amount of testing that the alternatives have, it's
> certainly not ready to be the default.  Maybe for wheezy+1, but for
> wheezy that would be a bit premature, and work is still needed for it
> to properly integrate (I only got a patch for update-rc.d and
> invoke-rc.d two days back, so if a package wants to update or run its
> init scripts or service then it won't work properly when using
> systemd).  That's not a problem for just trying it out, but it's
> certainly not ready for production use, let alone being the default.

When I called it "hopelessly buggy" I was - perhaps wrongly -
extrapolating from the claims that it's unstable, experimental, etc.

(I'll refer you to my first para's "pure speculation" explanation in
this post; and I'll defer to your judgement regarding wheezy v/s
wheezy+1.

If I were to install systemd on Debian, I'd expect to use systemctl
but it's nice to know that there are plans to make update-rc.d and
invoke-rc.d "speak systemd."


> systemd is technically superior on some levels, compared with
> sysvinit.  But it's also a lot less flexible, due to swallowing up
> many smaller tools and daemons into a single monolithic package.
> That has consequences which might not be so great down the line.
> See:
>
> http://lists.debian.org/debian-devel/2012/04/msg00751.html
>
> for my take on this.
>
> This isn't about being "conservative" (though I would hope that in
> general our users would appreciate us being conservative--it's one
> factor which results in Debian being a solid and reliable system).
> It's about taking a longer-term view of what is in *our* best
> interest, rather than jumping at the latest cool and shiny thing,
> and tying ourselves to the interests of RedHat, which may not be
> in our interest at all--it could significantly hamper our ability
> to tailor our own distribution for our own needs.

My reference to conservatism wasn't in respect of Debian as a whole. I
was alluding to those, for example, who'd like to have an insserv-less
file-rc driven init system. (Debian's makes too much of an issue of it
being solid and reliable but everyone's entitled to have a specific
PR/advertising/marketing angle...)

I agree that Debian - and other distributions - should look after
their interests and not necessarily follow Red Hat's lead. But it
looks increasingly difficult to my observer, non-developer eyes. As
you note, systemd's swallowed up other projects.

udev's the largest such project to my knowledge. I'd expect that, for
the time being, the udev maintainers of distributions not defaulting
to systemd can pull udev only from the systemd repository. Can anyone
guarantee that this'll still be possible going forward? Won't the
systemd developers want to tie udev more closely into systemd proper
(rightly or wrongly)?