How I learned to not worry and love systemd (sort of)

I wrote this little piece originally on June 25, 2013 but did not push the publish button because I think it is incendiary, in more of one sense. But…

I’ve decided to continue adding to and editing this little piece: we’ll see how it ends up.

Systemd is truly a remarkable piece of technology (add sarcasm). In the beginning I was skeptic for two reasons:

a.) I tend to orthodoxy when it comes to the core. I started using old skool init more than 28 years ago and it was with *real UNIX®*; at some point I even deemed SysV init style startup scripts the work of the devil. Ha! Like I miss BSD 4.2 System 7 init scripts! Everything mashed up in one single 5000 line shell script, *shudder*. Before I go tangential into a rant that is truly for another day, let me say this: All those weenies that say “FreeBSD start up scripts are the simplest and coolest thing since sliced bread” are a bunch of Christmas geese[1].

b.) I’ve my share of battle scars with “advanced” init systems. Ever had to write a startup script for HP-UX? AIX? What about that horrid abomination Sun Microsystems (R.I.P.) invented for Solaris (their SysVR4/BSD FrankenUNIX)? Startup scripts written in a horrible dialect of XML? Honestly, I suspect SMF was written in the middle of a Colt 45 and crack brogrammer frat binge.

Then when Canonical pulled upstart out of the hat (or should I say, arse?), I certainly put attention. I was already a user of Ubuntu, though not a convert. I had the chance to exchange a couple of mails with Shuttleworth when the planning and design of the PPA system was in swing and he came across as a narcissistic know-it-all. In fact, I was vindicated some years later when the man had to do exactly what I told him: People need the option to have several PPAs under the same account, particularly if they are a project that hosts in Launchpad. There and then I decided I was not invested enough in Ubuntu nor would I be ever. I just kept using it because it was the best GNOME experience at the time. And I stuck until they moved to unity. After some time back with Debian (my first real comfortable desktop GNU/Linux in the 90’s after all the false starts with Slackware and its predecessors), I moved to Fedora but found it was inherently unstable in my hardware, due to the serial patching to all software, something that also breaks many things in Debian. So I ended up using Arch, clean not-patched source code, and there I met systemd close and personal.

I was happy with the simplicity of the original scripts Arch used to boot with SysV init although I was very aware of the ugliness of the back-end. When the developers announced that they were planning to replace it with systemd, I was not enthusiastic even for a moment; why change something that works even if it is half-broken? Luckily, they took their time to implement it. In the end it took the longest part of a year to replace the old init with the new. I was wary it would have a negative impact on me and the distro (in that order; I’m selfish, so what). I decided to plunge in and install the test packages to have a go while I still could fall back into the old init system.

At first I was totally astounded. It wasn’t as anything I’d seen yet, so naturally I was very unhappy. I didn’t understand the whole damned thing and even the man pages seemed to be written in classical Cantonese. But digging through the supplied service files while reading the manual pages and the articles published by the authors made my mind click and after some months I began to like and appreciate the simplicity and power of expression of the systemd configuration system.

But not everything is wine and roses, I was perplexed when the systemd project subsumed udev and the efforts of late that make the project the only source of basic parts of user-space plumbing. I wonder if this can create a single-point of failure considering the questionable quality of core parts of systemd’s source code (the parts written by the two main developers!). I can see why a few very shouty people make such ruckus, but to be honest, those making the most noise are the most ignorant, thus the most damaging. They drown in their noise the voices with valid concerns expressed by knowledgeable and reasonable people who don’t need to resort to violence to express their opinions and ideas. In the end if systemd becomes unmanageable and quality plummets, there is always the option of a fork. I stay open to change, but I am keeping an eye on Void Linux.

Why Void Linux? Well, because they are the first to have the cojones to go their own way and adopt a different init: runit. Now, runit has been around for 11 years, and there are other tools as old or older that try to address the init and the PID 1 problems, that is both the problem of being the first process that starts everything else and that of process supervision; problems that are to some extent orthogonal. Dan Bernstein’s daemontools or monit address the second problem; tools like OpenRC, runit and s6 address both. You can read about many more in this research paper.

These are interesting times in Linux land to live in. I am enjoying the show from the peanut gallery and to all those who shout madly and throw tantrums I have this to say: Go forth and create your own Linux distribution; why didn’t you do it 4 years ago when Poettering showed up with this thing you hate? The owner of Distrowatch will be grateful and perhaps the 5 fools that end up using it. But believe me, those 5 fools will be all a bunch of self-entitled ungrateful bastards that will make you consider suicide as an honorable exit. If you can’t create your own distribution that works as you see fit because of your technical inaneness, why don’t you try PC-BSD? Sorry, it doesn’t run in computers bought 2 years ago, but  who cares? ZFS is way cool!!

[1] If you lost the reference, it means they are full of shit.

Advertisements