systemd: Yet Another Init Replacement

Bralkein · 2010-04-30 22:28:23

Just wondering what the Arch community makes of this. Not sure if it's exactly KISS, but some interesting ideas there nonetheless. Plus he has code, which is nice.

Last edited by Bralkein (2010-04-30 22:28:43)

keenerd · 2010-04-30 22:56:35

Claims it can be used as a drop in replacement for sysv. Since our init scripts are built on top of sysv, we might be able to use it without too much trouble. If this weekend is slow, I'll probably give it a whirl.

Diaz · 2010-05-01 06:11:56

I read the while think in the afternoon, and the concept and ideals behind it make sense. I will be here to watch what will come out of it.
At least some good ideas will prevail i believe even if the project fails.

For now, i am all for it, develop it, i'll wait to really see the diference ^^

falconindy · 2010-05-01 18:23:43

Definitely some interesting ideas, but not the most KISS solution. I threw a PKGBUILD up on the AUR for this, though I suspect that a few things may be out of place.

http://aur.archlinux.org/packages.php?ID=36902

This reminded me of exherbo's up and coming init system 'genesis', which also has some interesting ideas, but may eventually be dubbed the 'Duke Nukem Forever' of init systems (despite the author's best efforts).

timofonic · 2010-05-02 03:35:47

This project is getting a lot attention, lots of people says it's gonna replace the current way and even replace Upstart in all ways (and lacks the copyright asignment as mentioned the Systemd author).

It's suffering the HYPE syndrome, meaning that got attention due to the sensationalism of certain self-proclaimed "geek" (but lacking scientific rigor and a strict objectivity) websites.

He's the guy behind projects like...

- Avahi
- RealtimeKit
- PulseAudio

Love or hate him, he's quite relevant in the Linux ecosystem.

Fedora and OpenSuse guys seems to adopt "systemd" if the experiment succeeds.

There are guys working on "systemd" from Novell, Red Hat, Intel and Nokia. The project is gaining mommentum.

He's also a Gnomer. I dislike both KDE and Gnome equally and avoid them all time, so no problem.

EDIT:

After looking at PKGBUILD...

makedepends=('git' 'libgee' 'vala')

Seriously...

Libgee and Vala as make dependencies for a system wide thing? Do they try to taint everything with Gnome stuff?

(edited to remove "hate", it was a joke but it can be undertood wrongly as offensive to Gnome or KDE users)

Last edited by timofonic (2010-05-02 17:00:27)

Dieter@be · 2010-05-02 09:52:25

timofonic wrote:

A bad point: he's a Gnomer (and I hate both Gnome and KDE)

And he once told me "if you would use a proper distro..." when I was talking to him about Arch he sees things a bit too complex imho.
i mean, i also like the ideas behind systemd, but his disgust of shell scripts is silly, and systemd is already more then 30k sloc! it contains too many features imho. it tries to be compatible with everything.

PJ · 2010-05-02 11:06:55

The concept is intersting but I am not a fan of the implementation, it seems to be too complicated because it tries to solve to many problems. Maybe it will replace sysvinit but I have my doubts, I mean there have been other implementations before and non of those have actually been able to replace sysvinit. I don't think this will do it either but I guess time will tell. Besides, it doesn't fit to well into the UNIX Philosophy of having one program that does one thing well, but I guess that is just my opinion...

timofonic · 2010-05-02 15:40:54

I talked with the people from #systemd at freenode, they were nice and helpful.

They said Vala and such are dependencies for UI tools and then being optional, but due to a script bug they are still needed (that's funny to me, but well). Anyway, they don't see the sense at separating them from the main project. I do.

I hope the project matures and these decisions get into consideration, other than that I'm not going to taint even my test systems. It needs lots more eyes and hands on it, let's see if Linux Foundation gets involved too.

timofonic · 2010-05-02 16:46:51

He replied me the following...

Well, I don't share your thoughts about Vala. It's an awesome language
and simply a lot nicer to use to wrap D-Bus interfaces than plain C. In
the end it just generates C code however, and hence is as fast and lean as
normal C code is.
Please understand that I cannot take into account that some people have
emotional issues with certain technologies. I try to be rational and
Vala appears to be a good choice to me.
Note that Vala is only needed for the client side tools (i.e. tiny
binaries calling into the server via D-Bus). The server part is native
C. There are two Vala tools: systemctl is a command line tool for
accessing the systemd server, systemadm the UI tool you saw in the
screenshots.
You are welcome to fork systemd and reeimplement the Vala parts in pure
C. A cleanly written pure-C systemctl reeimplementation I'd even be
happy to merge upstream. A reeimplementation of systemadm I am however
I am not interested in, programming UIs in pure C is just pure pain. I
am not interested in maintaining such a beast.

Profjim · 2010-05-02 17:25:37

You are welcome to fork systemd and reeimplement the Vala parts in pure
C. A cleanly written pure-C systemctl reeimplementation I'd even be
happy to merge upstream. A reeimplementation of systemadm I am however
I am not interested in, programming UIs in pure C is just pure pain. I
am not interested in maintaining such a beast.

That sounds reasonable to me.

PJ · 2010-05-05 21:36:17

I was just looking at the source code for systemd. Well, I was just curious how systemd had solved one thing mostly since I am playing around with my own implementation of an init system. Anyway, looking at the commit log it now seems like systemd has support for archlinux.

dieghen89 · 2010-05-05 23:02:56

I've recompiled the stock kernel (2.6.32.2-2 i686) with DEVTMPFS, I've uploaded it HERE (to download press the small click here)

But It doesn't boot yet....

falconindy · 2010-05-06 21:52:10

dieghen89 wrote:

I've recompiled the stock kernel (2.6.32.2-2 i686) with DEVTMPFS, I've uploaded it HERE (to download press the small click here)
But It doesn't boot yet....

Currently, it also requires CONFIG_CGROUP_DEBUG.

dieghen89 · 2010-05-06 22:06:10

Thank you falconindy! Now I'm compiling the stock kernel with that option enabled...

I have seen that there are a lot of options in CGROUP section...only DEBUG is needed or also the others?

dieghen89 · 2010-05-06 22:37:22

Something is changed....And here the stock kernel with CGROUP added....

falconindy · 2010-05-07 01:26:50

Please understand that this is turning out to be so ridiculously beta it's not even funny. Even after I've gotten my VM to boot on systemd, its not useable. Debug info is spewed nearly nonstop, and none of the services in the init script pack I have do things like...

trigger udev
remount root as rw
parse fstab to mount other devices
load extra modules
start daemons

Wanna reboot cleanly? Good luck. This may eventually be a "drop-in" replacement for sysvinit, but not yet. FAR from it. This is going to be many many man hours of work before its stable. If you want to help, we don't need screenshots. We need .service and .target files to fill in the gaps.

I've been in touch with Lennart and he's been helpful thus far. But, it's not really clear he's ready for even early adopters just yet. Regardless, I'll be hacking away...

For those dying to get a kernel26-systemd, I'll post it on the AUR shortly...

Last edited by falconindy (2010-05-07 01:40:38)

Bralkein · 2010-05-08 18:36:55

I guess he wasn't joking when he said it was version 0! Well done for trying it out though.

PirateJonno · 2010-05-21 11:42:56

For a fast and efficient boot-up two things are crucial:
* To start less.
* And to start more in parallel.

This isn't working for me:

$ echo hello | less | more
hello

flamelab · 2010-05-21 11:49:23

PirateJonno wrote:

For a fast and efficient boot-up two things are crucial:
* To start less.
* And to start more in parallel.
This isn't working for me:
$ echo hello | less | more
hello

omg i didn't see that happen

PirateJonno · 2010-05-21 12:04:58

flamelab wrote:

omg i didn't see that happen

oh wait, it actually does work if you echo more than a few lines .
pretty weird how he made it bold as well

tvale · 2010-05-21 20:26:33

PirateJonno wrote:

For a fast and efficient boot-up two things are crucial:
* To start less.
* And to start more in parallel.
This isn't working for me:
$ echo hello | less | more
hello

AHAHAHAHHA.

Diaz · 2010-05-23 02:29:45

OMFG awesome xD

ajk · 2010-07-15 11:51:51

Version 3 is out! Integration in Fedora is at full speed now.. Arch Linux initscripts are being created at http://github.com/falconindy/systemd-initscripts/

This is awesome

falconindy · 2010-07-15 12:52:25

Hey there. Just some comments about the initscripts...

They're far from useful, right now. They're based on the pre version 0 systemd-git and atm, systemd works better without the initscripts than with. In short: do not use them. At this point, the repo probably needs to be nuked and reinitialized. systemd has the ability to use Arch's /etc/rc.d daemon init files, but they need to have LSB headers added to them in order for system to be able to parse them into usable units. It may be wiser to start from scratch entirely and simply use .service files where possible -- i.e. where shell logic isn't necessary and its just a simple start/stop. I welcome anyone who wants to spend a few hours reading up on doc and contributing.

Last I checked, systemd had some boot time issues on Arch as well which I haven't been able to stomp. The biggest issue being that dbus's socket isn't connected to on boot. That said, I haven't taken a look at it since the dbus requirement was bumped up to 1.3.2 (which doesn't exist as a git tag yet).

Last edited by falconindy (2010-07-15 12:54:21)

ajk · 2010-07-26 08:10:18

Did anyone try systemd allready on their Arch Linux system? I just build v4 but the network still doesn't come up by default.. Anyone has a clue on how to fix this?

Arch Linux

#1 2010-04-30 22:28:23

systemd: Yet Another Init Replacement

#2 2010-04-30 22:56:35

Re: systemd: Yet Another Init Replacement

#3 2010-05-01 06:11:56

Re: systemd: Yet Another Init Replacement

#4 2010-05-01 18:23:43

Re: systemd: Yet Another Init Replacement

#5 2010-05-02 03:35:47

Re: systemd: Yet Another Init Replacement

#6 2010-05-02 09:52:25

Re: systemd: Yet Another Init Replacement

#7 2010-05-02 11:06:55

Re: systemd: Yet Another Init Replacement

#8 2010-05-02 15:40:54

Re: systemd: Yet Another Init Replacement

#9 2010-05-02 16:46:51

Re: systemd: Yet Another Init Replacement

#10 2010-05-02 17:25:37

Re: systemd: Yet Another Init Replacement

#11 2010-05-05 21:36:17

Re: systemd: Yet Another Init Replacement

#12 2010-05-05 23:02:56

Re: systemd: Yet Another Init Replacement

#13 2010-05-06 21:52:10

Re: systemd: Yet Another Init Replacement

#14 2010-05-06 22:06:10

Re: systemd: Yet Another Init Replacement

#15 2010-05-06 22:37:22

Re: systemd: Yet Another Init Replacement

#16 2010-05-07 01:26:50

Re: systemd: Yet Another Init Replacement

#17 2010-05-08 18:36:55

Re: systemd: Yet Another Init Replacement

#18 2010-05-21 11:42:56

Re: systemd: Yet Another Init Replacement

#19 2010-05-21 11:49:23

Re: systemd: Yet Another Init Replacement

#20 2010-05-21 12:04:58

Re: systemd: Yet Another Init Replacement

#21 2010-05-21 20:26:33

Re: systemd: Yet Another Init Replacement

#22 2010-05-23 02:29:45

Re: systemd: Yet Another Init Replacement

#23 2010-07-15 11:51:51

Re: systemd: Yet Another Init Replacement

#24 2010-07-15 12:52:25

Re: systemd: Yet Another Init Replacement

#25 2010-07-26 08:10:18

Re: systemd: Yet Another Init Replacement

Board footer