Debian To Switch To Systemd Or Upstart

[ArneBab]

Actually, I didn't find out how to setup a bridged network - the former gentoo way seemed not to be supported, so I just hacked all the commands I need to set up the very network infrastructure into a shell script and ordered systemd to launch that file in order to set up the network... Very easy.

That?s good news.

[b]Basically all distributions have the same needs

?Basically? is the key word here, because the details can well differ.

I don't know how hard the dependencies on the kernel version are, but typically, I am using a recent version anyway - whatever Gentoo declares stable.

I don?t and luckily Gentoo does not force me to.

Here systemd adds restrictions which aren?t in Gentoo. It even requires me to add selected kernel features which I would not need otherwise - and I would not be able to opt out, because leaving them out would not just cost me some pretty graphics or cause some flickering. It would stop my system from booting at all. That?s a real hard blocker. And this moves in lockstep with features they provide which some other program might depend upon - even though the other program does not require that kernel feature by itself, so not updating systemd would stop me from updating bigger and bigger parts of my system.

You already got into that situation yourself with Gnome dependencies on systemd. Now imagine what happens if systemd requires something you really do not want.

And as I quoted Lennart in my blog, they actually state publicly that their intention is to ?gently push everybody? - in this case to use the kernel versions and configurations they want.

Can you trust a central, single source of control?

Then, nobody stops people from porting systemd to other kernels, keeping the APIs intact.

Since the systemd developers claim themselves that what they do requires Linux, it is likely very hard to untangle systemd from Linux. Also they would be coding against a moving target, because the systemd developers do not care about other kernels - and coding against a moving target is a recipe for disaster.

Wouldn't it be nice if one could reuse the knowledge about init when switching distributions?*There is still package management and default desktops distributions can argue about

Wouldn?t it be nice if one could reuse the knowledge about the package manager and the desktop when switching ? uhm? the pretty logo?

[oleid]

?Basically? is the key word here, because the details can well differ.

I've never seen any system which has different needs, from Linux on mobile phone (Jolla seems to use systemd there) to super computer clusters (I'm currently setting up a cluster for our department with 40 Infiniband connected nodes ? 16 cores per node - not counting HT - and actually yesterday I actually HAD to switch the init system on the nodes from OpenRC to systemd to make rootfs on iSCSI+OCSF2 work - if you like, I can elaborate that. Maybe I just missed something).

Here systemd adds restrictions which aren?t in Gentoo. It even requires me to add selected kernel features which I would not need otherwise - and I would not be able to opt out, because leaving them out would not just cost me some pretty graphics or cause some flickering. It would stop my system from booting at all. That?s a real hard blocker.

Just checked the ebuild. The minimal kernel version for the latest systemd ebuild is 3.0. For the oldest ebuild it's 2.6.39. That seems not very greedy for me. Even debian stable is at 3.2 nowadays.

Considering the needed kernel config options: Is there any reason not to use them apart from possibly longer build times? A kernel containing the needed options seems to work even on a mobile phone (--> Jolla)

And this moves in lockstep with features they provide which some other program might depend upon - even though the other program does not require that kernel feature by itself, so not updating systemd would stop me from updating bigger and bigger parts of my system.

You may be right about that - though I don't know if that ever occured so far.

You already got into that situation yourself with Gnome dependencies on systemd. Now imagine what happens if systemd requires something you really do not want.

Whatever that might be

I'm currently not sure if that systemd dependency of Gnome is not really gentoo-made. It's supposed to work with ConsoleKit, too - although people say that ConsoleKit is quite broken.

And as I quoted Lennart in my blog, they actually state publicly that their intention is to ?gently push everybody? - in this case to use the kernel versions and configurations they want.

Can you trust a central, single source of control?

Well, I guess if they ever started pushing to hard, there is always the option of forking an older version without the unwanted requirements.

Since the systemd developers claim themselves that what they do requires Linux, it is likely very hard to untangle systemd from Linux. Also they would be coding against a moving target, because the systemd developers do not care about other kernels - and coding against a moving target is a recipe for disaster.

What I meant by porting is creating something with the very same interfaces which can be used by desktop environments and those who need it. According to what I read the interfaces of systemd always stayed the same, even if the internals changed over the time.

Wouldn?t it be nice if one could reuse the knowledge about the package manager and the desktop when switching ? uhm? the pretty logo?

Don't forget the fixed location of the icons

[ArneBab]

Considering the needed kernel config options: Is there any reason not to use them apart from possibly longer build times? A kernel containing the needed options seems to work even on a mobile phone (--> Jolla)
…
You may be right about that - though I don't know if that ever occured so far.

I do. I had my GentooXO working nice and well for years, even though I could not update the kernel (I did not find a kernel configuration which worked with newer kernels).

Then a required udev-update made it impossible for me to update it for month, until I got a newer kernel to work. Now I face the problem again, because after the next (required-by-udev/systemd) kernel-update, WLAN ceased to work. My tries to get the newer kernel version to work till now only resulted in a non-booting system.

So you see, I’m not talking theoretically here. I AM using a fringe platform and I am forced to choose between not updating my system at all (with all the added security implications like browser-exploits, especially when on a hotel WLAN) and spending hours upon hours trying to get the newer kernel version to work - or ditching the Gentoo for the OLPC.

[oleid]

[...] So you see, I?m not talking theoretically here. I AM using a fringe platform and I am forced to choose between not updating my system at all (with all the added security implications like browser-exploits, especially when on a hotel WLAN) and spending hours upon hours trying to get the newer kernel version to work - or ditching the Gentoo for the OLPC.

What kernel are you using? And what OLPC do you have? A quick search yielded a x86 compatiple CPU for the older OLPC, ARM for the later ones. If yours is x86, maybe you could try a recent install image of Gentoo to quickly try a recent Linux version. If it's the ARM one, maybe check out archlinux. There seem to be images for the XO.

[Ibidem]

I don't know how hard the dependencies on the kernel version are, but typically, I am using a recent version anyway - whatever Gentoo declares stable.
Non-rolling distributions don't upgrade the init system during the life cycle of a release, so I don't see a problem here, too.

Debian supports upgrades. And a partial upgrade should not leave the system unbootable.
So Jessie should not require a kernel newer than 3.2 at the latest; and it's not going to freeze for a long time yet.
Now, there's also folks who have to run out-of-date kernels for various reasons:

http://forums.debian.net/viewtopic.php?f=5&t=103166

http://forums.debian.net/viewtopic.php?f=10&t=100348

And Debian actually does try to support that sort of use, so far as it can be done.

By the way, just because an architecture is commonly used doesn't mean that every feature of every system using that architecture is well supported. And a Geode is not really as mainstream as an i3 or i5.

@Awesomeness:
IIRC, the cdrecord maintainer added CDDL code to a GPL project (claiming that the two licenses are compatible despite what Sun and the FSF say), decided to insist on using Solaris "b,t,l" device naming on Linux (no /dev/* names!), and asserted that anyone who patched the software should rename the binaries so that the upstream reputation would not be hurt; in fact, he asserted that the GPL required that.
cdrkit was forked because he made the project impossible to legally distribute and also toxic.

ffmpeg and libav use the same SONAMEs and shared library names, so it's not possible to have both for packages to build against.

@chithanth: do it before telling me "linux is not about choice". If you don't want to make that claim, feel free to ignore the challenge.

[oleid]

Kernel 3.0 seems to be perfectly fine for the very latest systemd - according to the dependencies listed in the ebuild. So a partial upgrade to Jessie shouldn't pose a problem.

[Ibidem]

Kernel 3.0 seems to be perfectly fine for the very latest systemd - according to the dependencies listed in the ebuild. So a partial upgrade to Jessie shouldn't pose a problem.

Assuming that (1) systemd doesn't push that up by a couple versions within the next year or so (which would be a quite reasonable pace); (2) systemd can be relied on to not drop support for kernels that are ~3 years old; and (3) that Debian doesn't care about users who needed a kernel that was 3 releases older.

Systemd changes too fast for Debian, IMHO. 2 kernel releases is too little safety, and for Raspian that's actually 1 kernel release (they use 3.1.x).
And there are people asking about using Jessie with 2.6.32, which is _currently_ supported.

[doom_Oo7]

Does anyone knows if the Debian guys are starting to agree on something ?

[Vim_User]

Does anyone knows if the Debian guys are starting to agree on something ?

We will have to wait for the decision of the TC.

[Pajn]

We will have to wait for the decision of the TC.

Is it possible to see where they are heading and what arguments are brought up?
Although I don't like SystemD because of its monolithic model it would be very
interesting to see technical arguments by a group that tries to be neutral.