[Nottingham] Someone doesn't like systemd

[Andy Smith]

Hello,

On Thu, Aug 21, 2014 at 12:13:42PM +0100, Martin wrote:
> There's a very succinct blast on the LinuxVoice forums:
> 
> http://forums.linuxvoice.com/viewtopic.php?f=4&t=74
> 
> 
> The killer for me that undoes all the goodness is the disaster waiting
> to happen of vast rapidly changing code all running as PID 1. Even
> stranger is for *why do that* when completely unnecessary?!

Note that the different components of systemd are all separate
binaries, although the one that runs as PID 1 is indeed a couple of
MB in size now, which is much bigger than sysv init.

> And if there is some fundamental need or missing feature that absolutely
> must have PID 1, then why not improve the kernel to fulfil that whatever
> assumed deficiency?

This sort of thing very much does not belong in the kernel. Although
if there is much more friction between the kernel community and the
systemd community then it is just possible that Linus "does a git"
and implements his own init system!

> Or is there just crass over-hasty blindly compromising rush?...

In my opinion, we really need this stuff, and Canonical shot
themselves in the foot by insisting upon a CLA. I think things could
look very different if they hadn't done that.

For those who don't think the CLA is/was a big deal, here's a video
from a DebConf where the topic of the Canonical CLA came up in the
context of Debian devs contributing to Upstart:

    https://www.youtube.com/watch?v=CUOAUQJ-y00&t=42m19s

Since Ubuntu gave up on Upstart after Debian ruled it out, I think
they made a huge mistake treating this issue in such a cavalier
manner and brushing it off as "lol Debian and licenses, deal with
it, others have." Yes, well, others weren't critical to your
strategy…

> The modular and secure Unix philosophy has worked very well for Linux
> systems thus far. Why abandon that now?...

systemd is quite modular you know. The problem is that it is its own
protocol, as in: they are not documenting a protocol for this stuff
and then writing the reference implementation of that protocol; they
are writing the reference implementation (systemd) and then
documenting what they wrote. So any competitors could only follow
behind.

However, it is pretty well documented. So replacing components of it
is possible (and has already happened for some bits of upstart).

> Something like systemd is needed. However, who else is there to be able
> to race beyond Red Hat?

Gentoo have been working on a competitor called OpenRC which is
quite interesting:

    http://wiki.gentoo.org/wiki/Comparison_of_init_systems

it runs on top of sysvinit and bolts on a lot of the features people
want from systemd, though not all of them. I'm not sure how
successful this is going to be because it is hard to make brain
space for something which is obscure and doesn't do everything that
the main competitor does.

Though there is still the open question about which init system
will be used on architectures that don't have Linux cgroups (e.g.
GNU Hurd, Debian kFreeBSD). Maybe OpenRC will see some use there
too.

Cheers,
Andy

-- 
http://bitfolk.com/ -- No-nonsense VPS hosting