Debian Still Debating Systemd vs. Upstart Init System

[RahulSundaram]

So you are now accusing the TC to be biased?

Everyone is biased and this shouldn't be a surprise to people but as I noted before, several of the TC members are contributors to Upstart and that is a significant factor in addition to who they work for. My post was a public note for people who may not be aware of the dynamics at play here.

[RahulSundaram]

Thanks for reminding me of that page - it's a bit better than I remembered - but it's still the case that several important sections couldn't be reliably reimplemented, and very few of the systemd implementations are supported to work with other init systems.

Systemd developers can only state what interfaces they can commit to supporting and supporting alternative init systems isn't something anyone can reasonably except them to do. If you are developing other init systems, it is upto you to take those interface commitments into considerations. If Upstart doesn't want to copy code from systemd because of their CLA requirement that is a self inflicted problem.

[moonwalker]

I have never used systemd, nor upstart, and I cannot judge them on technical merits.

But whenever I read discussions about them, I've always found the explanations of systemd's advantages to be more compelling, those arguing for systemd to be more coherent in their explanations, and most arguments against systemd were.. non-technical.


SysVInit sucks, and OpenRC (which I currently use on gentoo) has it's flaws, too. From what I read, systemd neatly solves all my complaints about OpenRC, and then some. I'm looking forward to try it on my systems once gentoo integration has matured enough.

I've been using systemd on Debian in form of systemd-sysv package that completely replaces sysvinit package. There were some issues in earlier versions some year or two ago and a brief period where my system was for whatever reason booting actually a bit slower than before conversion, but right now on my Debian Sid desktop setup what takes the most time during boot is actually me entering 13-symbols-long LUKS password.

[leech]

I've been using systemd on Debian in form of systemd-sysv package that completely replaces sysvinit package. There were some issues in earlier versions some year or two ago and a brief period where my system was for whatever reason booting actually a bit slower than before conversion, but right now on my Debian Sid desktop setup what takes the most time during boot is actually me entering 13-symbols-long LUKS password.

Ha! I'll have to agree with that. My Arch system boots frighteningly fast, couldn't imagine it on an SSD, and same with Debian Sid with systemd enabled. From what I've seen in Sid, Debian has more or less already settled on systemd, if you're running the default desktop of Gnome.

[Ibidem]

Not really. Debian ports include KfreeBSD AND Hurd. I think supporting all of these forcing a lowest common denominator mindset which makes it unattractive for developers to integrate well with any kernel but free free to post an update when there is a Hurd port in progress too.

#1, the port to kfreebsd shows the way for a potential port.

#2, hurd-i386 (the only Hurd port) has several large divergences from standard packaging. Some of this is because of gratuitous incompatibility with POSIX, such as the refusal to define PATH_MAX.
(kFreeBSD is slightly closer to POSIX than Linux, thanks to the tty reset flag; but even if an init system were fully POSIX it might not compile on Hurd.)

#3, hurd is not an official port.

#4, Debian Hurd has long been and still is exceedingly buggy. Not just "a lot of things act unusually"; I'm talking about:
-iceweasel didn't have working https (as of iceweasel 9)
-dbus isn't working
-dhcp restarts the whole TCP/IP stack
-emacs doesn't build
-a critical security bug in the kernel was reported in Oct 1999. The fix didn't get uploaded till January 2006.
(Speaking of which, upstream lists the latest kernel version as 1.3, from 2002.)
I found this out from https://wiki.debian.org/Debian_GNU/Hurd, some of the linked bug reports, and https://wiki.debian.org/ArchiveQualification/hurd-i386.

And while I'm at it...
The impression I get is that some people are suggesting that Debian support systemd, launchd (what many systemd advocates point to for BSD systems; IIRC, it ran on standard FreeBSD but not as PID 1, work has just been revived, and no one has tried it with glibc), and openrc (usually considered the "best candidate" for hurd, though no one has tried building it there), plus initscripts for the transition.
This basically means "Everthing but upstart," and does not seem at all realistic.

Then there's the systemd on Linux/OpenRC everywhere else alternative. This will require the Hurd and kFreeBSD porters to provide testing for OpenRC, and everyone on Linux gets systemd.
(If this happens, I'd be tempted to install kFreeBSD--and a number of other Debian users have stated a preference for "not systemd".)
It still means that Debian needs to split systemd up; the maintainers have to maintain packaging for both configurations, and one gets little testing.

A systemd port has been alluded to, but if upstream says "You'd be better off writing a clone of systemd" it seems absurd to expect this.
A systemd clone is not going to be easy; porting upstart and using a split-up systemd atop that is likely to be easier.
(Russ actually suggested systemd on Linux + Upstart elsewhere).
It's definitely going to happen sooner, since I see no signs of anyone even starting on a complete systemd port or clone.

In summary: I don't see any way to simplify Debian's maintenance burden for all platforms while using systemd as pid 1 on Linux.

(Now, really, I'd like to see OpenRC everywhere, even if OpenRC is a second choice on Linux.)

[profoundWHALE]

+1 for systemd and openrc

[Ibidem]

@RahulSundaram:
You might try mentioning slightly more relevant details like two TC members being involved in upstart development.

Also, may I ask what your proposed _full_ solution is?
(I mean init systems and systemd packaging on all three kernels.)

[crymsonpheonix]

With two Canonical employees in the Debian Technical Committee and Ian (Ex-Canonical employee), there are currently atleast three votes in favor of upstart. Russ is favoring systemd but other members haven't really participated in the discussions much if at all. We will have to hear from them more to understand their perspectives as well.

Being that they recently brought Keith Packard into the committee (making it 5 instead of 4) I suspect that the team is evenly split between upstart and systemd and Keith was brought in to be a tie breaker. I wouldn't want to be in his shoes

[Guest]

The best idea for debian would be OpenRC or Upstart for portablilty reasons. I would be fine with both, just no systemd please

No, that would be the worst idea. I don't give a shit about compatibility. As far as I know freebsd doesn't provie Linux kernel, so screw them. Futhermore kfreebsd usage in Debian is nearly none. It's stupid to make decisions basing on some legacy crap. They should choose systemd.

[doom_Oo7]

In the end they will vote by counting the posts on phoronix anyway!
also, why not put GNOME in the kernel instead ? No need for an init system then ! It will be known as Knome and KDE fans will rejoice!