Linus: “I no longer feel like I can trust ‘init’ to do the sane thing”
lkml.org
lkml.org
EDIT: Here are links to comments discussing the actual content of his email, in case anyone else came here for that and left wanting.
https://news.ycombinator.com/item?id=14734868
https://news.ycombinator.com/item?id=14733973 -> https://plus.google.com/+TheodoreTso/posts/EJrEuxjR65J
EDIT: I read all the comments on this post, and I am disgusted to report that (as of this EDIT, nitpickers) the 2 above-linked comments are the sum total of actual replies to the content Linus was trying to discuss. I hope LKML had more self-control than HN.
EDIT: 14733558, it was really excellent of you to try and defuse things, too.
And yes, a large part of this may be that I no longer feel like I can trust "init" to do the sane thing. You all presumably know why.
It's not hard to understand what he meant and for some of us it's articles like this on HN which give an opportunity to talk about systemd without being off topic.
IMO the take away from the article is that systemd has become enough of a moving part that it would be unwise to rely on it for default anything. Maybe I read it wrong but there's my take on the technical argument against using init to set default limits on processes. There are already defaults in the kernel so use them and allow an override for admins that need one.
Complaining that people aren't having the conversation you want to have is just weird. Start the conversation you want instead of complaining that others are failing to have it. No one is obliged to hold the conversation you want.
So far I haven't found a page like this by the SystemD advocates. I haven't really found much of an argument by them, to be quite honest. It would be nice to hear of one, of course.
I would argue it is off topic because it's barely relevant to the submitted link. Just post "Tell HN: systemd sucks" and you can talk systemd all day long.
You could even include some actually real and concrete concerns, like trying to bloat the kernel with some linux-on-teh-desktopz IPC which will probably be replaced by the next big thing by 2025, bundling some weird syslog replacement, bundling a half-assed DNS cache (it was originally vulnerable to spoofed answers right off the bat, besides the recent remote code execution oopsie), polluting dmesg with systemd log messages, asking the kernel to workaround systemd crashing when the kernel is booted with debug output, Poettering's dismissal of the UEFI bricking fiasco, the recent failure to drop privileges on usernames starting with 0, ...
You guys behave as if some overblown pretext were needed to flame systemd :p
Also with respect to your comment: "I am disgusted to report that ... the 2 above-linked comments are the sum total of actual replies to the content Linus was trying to discuss." Linus did not post this here and in no way was attempting to discuss this with the HN crowd. cnst posted this with the intention of discussing Linus's apparent distrust of init.
"For me a lot of "good taste" is about adding complexity when it's needed, and avoiding it when it's not needed. And if you do have complexity, making sure you have the tools so you can debug things when they break. And for me one of the things that I don't like about systemd is that it has added a lot of complexity, and when it breaks, trying to debug it can be almost impossible."
Also:
"Heck, I don't even I want to file bug reports, just so I can get abusive messages from Lennart. At least when Linus flames me, it's because I did something wrong which hurts users and which I d*mned well should have known better, given my years of experience in the community. Lennart just flames you because you're wrong, and he's right. By definition."
And:
"The high bit, in my opinion, is "not being able to admit you are wrong". If someone (such as Lennart) is always right, then it's impossible to have a technical discussion in a post-mortem to avoid similar problems in the future. In a company, if there are personnel issues like that, you can escalate to the person's manager, or use other mechanisms. In the open source world, all you can do is route around the damage. Whether you call the inability for someone to admit that he or she has contributed to the problem a "lie" or just that they were "mistaken" is really beside the point as far as I'm concerned."
The problem is with more recent people, that seem to be more "american" their approach. Always looking to climb the ranks and find new domains of development to undertake rather than stick with one area and become deeply proficient.
This kept failing. It turned out to be systemd having a hardcoded timeout of something like 30 seconds on any process launched by udev, that couldn't be changed without recompiling it from source, or having it fork off some other process group that lacked this limitation.
From what I can tell, systemd has some of the worst sort of developer myopia, where whole swaths of legitimate use cases are actively destroyed.
> ...where whole swaths of legitimate use cases are actively destroyed.
Your use case is fine. Your implementation doesn't fit with udev's design; that's all.
systemd kills the entire udev process tree (all children) including the ripping script after 30 seconds.
Those who need the complexity or some features should adopt the technical debt, there is zero rationale to enable it by default and impose it on everyone.
For instance if you have too many network interfaces and naming is a problem then switch on predictable network names which btw is anything but predictable for human beings and let the 97% who don't carry on.
12 digit random names are simply not predictable or better for human beings than eth0, wlan0? If there is a problem predictable network names is not the solution.
Similarly if you need binary logging and an audit trail then turn on binary logging but don't impose it on everyone else.
Its time to move beyond silly accusations of 'hate' etc every time there is a discussion about systemd because at the moment it just serves to deflect criticism and avoid accountability for questionable decisions.
Indeed. Even if I have multiple ethernet interfaces, I'd prefer to map a MAC address to eth[0-9] myself. But most machines only have one ethernet interface and one WiFi interface anyway.
These things pop up all over the place. E.g.: why would I want to have NetworkManager on a server, which never switch networks and need to set up a VPN connection? Why do I need firewalld on a server? I rarely need a zone-based firewall on my laptop (probably like most users, I just block all incoming traffic), let alone a server.
Layer upon layer is piled to solve hypothetical situations that do not arise for > 90% of the population.
(I do like systemd as a service manager, but systemd and the surrounding ecosystem has made it terribly difficult to understand and debug UNIX systems.)
Because systemd wants to be The One Ring of Linux, and tries to be all things to all systems and all people.
> why would I want to have NetworkManager on a server, which never switch networks and need to set up a VPN connection?
It's funnier because NetworkManager must wait for the filesystems to be mounted, but then, networked filesystems (that are, those ones, important there) need a functioning network that NetworkManager postpones to after it's up.
In the bug you reported, however, I noticed that he pushed a commit to work around the underlying bug, so he's not completely impractical.
https://github.com/systemd/systemd/issues/6237
systemd runs services as uid 0 because it can't parse a valid user name (of a non-uid 0 user). Of course it's not systemd's fault. It's not even a problem at all.
dies
> OpenRC is default
> we have to ignore the (unixy) distinction between OpenRC and sysvinit
Nice try :)
Found this article: https://www.pcworld.idg.com.au/article/129776/after_controve...
I wonder, how many people reading that article would predict the rise of stuff like GitHub and GitLab and the success of Git to the point of crushing CVS and making SVN decline in popularity?
I didn't dream of it in terms of GitHub.
In my experience most people are using git with a svn workflow. They use github (gitlab, or something internal to the company) as the central server, and work only from the central server. Make github speak svn, and alias all the git commands to svn and they couldn't tell the difference.
Yeap, but with a lot of local branches, and that makes a huge difference. Branching with SVN seriously sucked (although I've heard it's gotten better).
CVS was already on the way out even prior to git's inception, because of SVN. SVN's unofficial goal is essentially to be a better CVS, and it was a great success on that front. However, IMO its success made its core dev team complacent.
For example, one complaint I hear often of SVN is that it's slow. Go use CVS for a week -- and I mean not just doing a checkout and make a few commits, but merge, diff, switch branch, add / remove / rename sub-trees, etc.; bonus points for involving >3 people trying to do the same simultaneously -- then try and tell me SVN is slow.
Nonetheless, SVN is not fast, but it's fast enough when there was no viable competition. Then git came along, and even the staunchest opponents must conceded that it's lightning quick. And people will tolerate a lot of BS from a tool if it does the job and does it quickly.
SVN's performance has improved significantly since the 1.4.x days, around the time when git first made its appearance. No doubt some improvements were on the roadmap regardless, but surely having competition lit a fire in getting them out the door.
Git is extremely well engineered, it's just that its UI is not conducive to learning it.
1) git log uses a pager by default and is lightning fast compared to svn log which needs a server connection and is not paged
2) the "current commit id" is the same across all directories in a repo, while in svn every directory can be at a different revision
3) side benefit of #2, a "git status" in a subdirectory also shows the status of files in upper-level paths of the repo
4) backup-ability: every git checkout is a full clone, so if the server is down, or you're on the road without wifi, you can still work even with history
5) partial commits (git add/commit -p) is a godsend, I don't get why svn still hasn't implemented this
6) the tooling around hosting is way better - websvn is a laugh compared to gitweb and even more so compared to gitlab/bitbucket/github
7) support for branches
Depends on the business area. I am still using SVN in 100% of our customers.
And they don't seem keen to move into Git anytime soon.
Based solely on my personal experience (which is mainly Ubuntu desktop and server) Systemd is a hot broken mess. To me it feels like good goals with poor implementation.
I actually miss how simple init used to be... I understand the motivations for improving it, but if I had to choose ease of grokability or better boot times, I'll take grok every time.
It's a sad day that the biggest Linux company is doing Embrace, Extend, Extinguish, but I guess I shouldn't have expected more from RH.
Git didn't have this problem because:
0. Version control systems don't have to interact with as many other software packages anyway (editor/shell integration is nice, but you can live without them and just run git directly from the command line);
1. There were compatibility layers that let Git talk to popular VCSes;
2. The fact that you use Git personally does not affect anyone else, obviating the bulk of the need for coordinated upgrades (there were stories, from before GitHub was a thing, of 'guerrilla Git users' in companies that officially used CVS).
These three things ensured that a relatively painless incremental upgrade path existed; in the case of init systems, it will be much harder to establish.
[0] He talks about it in https://www.youtube.com/watch?v=1Mg5_gxNXTo ; can't find the timestamp right now, but it's in the latter half of the video IIRC.
* http://blog.darknedgy.net/technology/2015/09/05/0/
* http://skarnet.org/software/s6-linux-init/
Most that came before were good at forking, but not so good at merging.
""And yes, a large part of this may be that I no longer feel like I can trust "init" to do the sane thing. You all presumably know why.""
Systemd walks into a bar. It shoots the bar owner and proclaims itself to be the new owner. It then turns the bar into a farm, brewery and distillery, opens a casino, a freight rail line and an investment bank. Oh, and there's an init system thrown in there somewhere too.
What do you mean by ./just/run/some.sh?
Doesn't the fact that I have to google an alternative workflow for something that behaves consistently across pretty much all other software make its UI worse, at least in that regard? I get that maybe it's a justified trade-off, but that doesn't make the UI any better.
> What do you mean by ./just/run/some.sh?
I suppose just dumping an exec command into an executable file and call it a service.
Why must I have it if I don't need or want it? What is this, Windows?
and the scary part: we are not even fully there yet and it already reeks.
Init scripts are static. They don't change, they aren't ad-hoc, they aren't fragile. All modern Sysv init script systems have static scripts which 'include' a basic key=value pairs of environment variables from a config file. Nobody modifies scripts, or at least, they should not be doing it, and don't need to.
Many init scripts on my system have timestamps from the year 2002, and this system was just installed. They don't change because they don't need to change.
OTOH, Declarative unit files are configuration management taken away from the configuration management system, adding complexity.
Then you have your init scripts for random other services added by $USER, consider yourself lucky you haven't come across these.
What systemd "solved" was already solved.
When you know this you understand that the motivation of the people behind systemd was not technical.
One of the top contributor of systemd is banned from the linux kernel because of bad coding. The other top contributor gave us the fiasco of pulseaudio. Both work at the same major company. Draw your own conclusions.
Really? so you really think:
systemctl <command> <service>
[optionally, some initsystem] <service> <command>
The systemd UI is a hot steaming pile of garbage that has made my day-to-day life on the CLI harder, not easier.
systemctl status <service>
Different services have different failure modes with nuances that can never be generically captured by whatever means you chose to launch the service, in any kind of meaningful way. When a Database service stops serving data, my first though isn't "Lets see what the init system thinks", my first thought is "Lets see what I can find in various Database service logfiles" which will, 99.9% of the time, be something specific to the database service.
Inits' job in my view is "start this, and get out of the way". "Start this", and do logging, and do DNS/DHCP/manage-daemons/manage-restarts-on-fail/manage-the-fs/here-is-a-kitchen-sink/do-you-want-fries-with-that/oh-yes-and-here-is-NTP isn't what I am personally looking for in an init system, and - frankly - is a ridiculously stupid and idiotic approach from a server architecture perspective.
This isn't about the merits or lack thereof of binary logging, or the merits of whatever else systemd brought to the party. This is about needing tweezers, and being given a 500 function swiss army knife, where the other 499 functions get in the way of it being used as a very good tweezer.
Systemd is trying to be my default OS via the init system backdoor. I'll choose my own OS, thanks.
Then I'll check those logs if required.
Regarding DNS, DHCP et al, yes you have a point. Managing restarts on fail is a logical extension of init however.
* http://blog.darknedgy.net/technology/2015/09/05/0/
* http://jdebp.eu./FGA/run-scripts-and-service-units-side-by-s...
It's so cleanly engineered, just like ZFS.. makes me sad that solaris died a death.
(Their package management was crap though tbh)
* https://github.com/ServiceManager/ServiceManager/ (https://news.ycombinator.com/item?id=10212770)
I am certain that right now the choice of POSIX OS's at large companies such as mine is Debian or RHEL/CentOS. Smaller companies get away with ubuntu.
But nobody is making new infrastructure with IllumOS/Solaris. The most people will deviate is probably Free/OpenBSD.
While dmesg did that by default.
IMO this idea with copying is in itself even less sane than init using rlimits on PID 1.
He sees his domain beginning and ending with the kernel, and that's it.
Sadly certain people below him, that he trust deeply, are not so content (some even support Poettering and crew). And i fear the day he hand off the keys to the kernel to them.
If even the creator have concerns, man, that's not a joke!
So while it isn't perfect, it's IMO a step in the right direction, if only by lowering the barrier to entry to actually using and managing the thing.
If you are worried about stability then you shouldn't rely on a single machine anyway. Personally I manage systemd screw-ups (though I haven't had one in production yet - all of them were on my personal machines while monkeying around with Archlinux) just like any other hardware failure - by making sure my app/service stays available even if I yank the power cord from the machine.
Poettering seems to take the advice of "release early and often" to heart and pushes things into widespread usage long before they're quite fully baked. PulseAudio was awful for years, and then it was good and we all stopped complaining about it. systemd was kinda crap for months after it became widespread...a ton of breakage, including in ways that were occasionally dangerous. We're lucky, I guess, that systemd didn't take as long as PulseAudio to stop sucking all the time (systemd still sucks some of the time, but what it replaced sucked a lot, too).
So, yeah, you can't break userspace if you want Linus on your team, and systemd has broken userspace now and then.
Don't get me wrong. For years, ALSA was a pain. But in the last 10 years, ALSA has always "just worked" for me. I think the only time I had to fiddle with ALSA in the last decade was to get the mic on a headset to work.
Also had issues with bluetooth, as you seem to have. I will permit the idea that this is possibly unrelated (bluez is garbage, yes), but bluetooth audio was working for me previously, and only broke when pulse entered the picture.
FWIW from an API perspective I still think OSS seems better than ALSA... Linux audio seems to be a repeated case of people confusing interface with implementation, and the interface only ever gets more complex.
Of course you can still trivially disable the daemon and you lose software mixing & resampling, but things that work within these limits will run ok. But there really shouldn't ever be need to disable it, it just works.
The sndio api is pretty damn nice & simple too. https://man.openbsd.org/sio_open.3
I use Slackware, which tends to be more conservative about these things; the fact that PA is now the default in Slackware is a good indicator of PA having overcome its growing pains.
Additionally, I somewhat doubt BlueZ would've made PA such a hard dependency if they weren't reasonably sure that it's stable enough for primetime.
Anyway, I haven't had to google about a pulseaudio problem in at least a few years. I've been using Linux as my primary desktop OS since 1995. I've hated Linux audio for most of that time, and I've broken it in every possible way. But, today, my Fedora system has sound that Just Works. I don't know how it Just Works (though I know it's pulseaudio), because I haven't had to learn...because it Just Works. It works for Steam games, it works for browser audio, it works for system notifications, it works for movies, it even works for audio and music software. It's pretty weird.
Anyway, from what I can tell, the bugs, misfeatures, poorly thought out implementation, etc. all got worked out somewhere along the way. It took a long time. But, it did get fixed.
Have you actually used it in the past couple of years?
Exactly. Does it support redirecting audio streams to another machine on the same network? That's a vital feature in our hackspace, where everyone can send audio to the room speakers via PA, and it's also used extensively in my living room (where the speakers are connected to the home server).
I've read a pulseaudio guru post explaining that when you know what you are doing you can work out a pulseaudio config that reduce pulseaudio latency issues, real time resampling, drop outs and high cpu usage. Just don't use the default config.
This is a common issue. It happens on Windows and Mac, as well. Am I to understand you don't experience it with some other way of handling sound (e.g. not using PulseAudio)? Are all of them being driven by the same program?
I DJ, and have used multiple independent interfaces in the past; there were frequently pops and warbles as the various timing crystals in the various interfaces fell out of sync and were forced back into sync by the software. It happened on both Windows and Linux (though it was slightly less pronounced on Linux, and I never figured out why). JACK might be the best way to deal with that problem on Linux, though I never tried it (I just switched to a dedicated DJ controller with multiple outs and headphone output).
But, it's a common enough problem that most DJ software has something about the problem in their FAQ.
Just two more random examples:
RPi 3, audio would hang randomly when opening a new process using audio from another using audio. The solution would be to either play an audio sample from the command line to unclog the audio system, or uninstall PulseAudio.
Ubuntu 16.04. Rewinding the video while using mplayer could cause the audio to play at a faster rate. Again uninstalling PulseAudio would solve the problem.
I often feared PA, and something in ALSA (I almost know zero about linux audio stack) made it a zero effort / 99% working thing that matched what I want for audio, that is a no sweat thing that just push sound. I can live without multi room networked audio, but having local audio crash randomly is too much a PITA. Feels like having lisp macros on top of VBA.
That said I wish we could find a better solution. Something between ALSA and PA. Also PA demands quality driver, IIRC Lennard told he cannot get blamed for that, it's a bit like GPUs in a way; and I wish we'd have open hardware audio chip that were simple and not lying.
That said a default fallback would be so good. I can't stop thinking that LP writes things for his own world and leave it there when he's "satisfied" with it; even if it means shit for the rest. Which means the issue is that he shouldn't be in charge of socially impactful components.
Unless you have examples of solid patches fixing these kind of problems that he rejected, this is a significantly unfair statement.
At least that’s what’s been apparent to me from the things that I’ve read by him and about him.
And I'm not sure it's really ego.. I'd rate him at about 0.07geohots.
Works every single time. last time I did it was around March. (I still don't know how it creeps back into my computers.)
Last time pulse was creating problems for me I decided that since I was now writing an application that links to the sound system, let's stop linking to low level functions and use pulse abstractions that some people claim to be better.
Turns out that the pulse "abstractions" are basically verbatim copies of alsa. Just with some added badly documented options. Some with very descriptive names, other with cryptic ones, but whatever, no option seems to do exactly what its name implies. Besides, pulse abstracts away most of the flexibility of the sound hardware, so you can't use it, and well, every time I tried running something with it, I got an assertion failure on pulse code... Always with some completely non-descriptive description.
The only bug that's affected me in the last couple of years was that the default volume on USB headsets was always set way too low. Pavucontrol helped sort it out
Something to do with opening the card in different profiles I think.
- a while ago pa got audio routing wrong on a new/obscure laptop, wrong audio routing as in - speakers would not turn off if you plug in headphones. That eventually got fixed with dnf upgrade and no effort on my part;
- on ultrabook, pa sucks noticeable amount of battery by waking up 100 times a second (while no audio is being played). Something akin to what this poor chap is describing:
https://www.reddit.com/r/Fedora/comments/6d7776/pulseaudio_b...
However I've been using the MacOS sound system (not just the mixer) for many years through many devices (mics, external dacs, HDMI transports, DP etc) and it has always worked without any problems.
https://www.apple.com/media/us/osx/2012/docs/OSX_for_UNIX_Us...
We are using it for develoment but it's considerably more work to setup the develoment environment than installing a Linux distribution with everything we need available by default or a command away. Also in El Capitan the root user is disabled by default.
http://www.infoworld.com/article/2988096/mac-os-x/sorry-unix...
Linux != Unix. Darwin is much closer to BSD.
> Also in El Capitan the root user is disabled by default.
Lies. I never touched a thing and (on 10.13 ß):
$ sudo su -l
password:
Galileo:~ root# ps -u root | wc -l
115
- SEP prevents you (or any random script/software/installer you run) from meddling with system files and that is a good thing too, no different from sensible SELinux or AppArmor policies.
How are you managing different package versions on Linux – PPAs to avoid conflicts with the system Perl, Apache, etc,
That said, macOS is just as Unix-like as it's ever been. As lloeki said, it's actually closer to BSD than Linux, which may explain some of your gripes.
As for the root user, it's been disabled by default since Mac OS X first came out. Being able to log in as root is considered a security risk, and there's no need for it. If you want root access, use sudo.
Also people at work are allowed to use desktop linux at work as long as we don't bother IT.
(This with a serious company with even older more seriois roots.)
I think desktop linux is still on the rise.
https://blogs.gnome.org/uraeus/2017/06/20/fedora-workstation...
There really is a changing of the guard in Linux userspace, and the change is for the worse.
Now if only the same could be said for bluez...
https://github.com/systemd/systemd/issues/6237
Its going to end up exactly like glibc under Drepper, where distributions are forced to have a mile of patches because obviously broken shit isn't patched because the maintainers are too arrogant to admit fault.
> systemd is not the one coming up with the restrictions on user names, and while some distributions are less restrictive, many do enforce the same restrictions as we do. In order to make systemd unit files portable between systems we'll hence enforce something that resembles more the universally accepted set, rather than accept the most liberal set possible.
These are pretty clear "refuse to fix". Though I do agree that this is a poor example to illustrate the point.
As the owner for a critical service like systemd (I don't use it, I prefer openRC on gentoo), he should know better.
I don't like your name. Therefore you get root.
How on earth did this init system become so prevalent?
* http://jdebp.eu./FGA/debian-systemd-packaging-hoo-hah.html
The Arch process was rather different:
* https://news.ycombinator.com/item?id=11834348
The Debian people did look at who develops the various systems, but they mainly just measured the developer count and what else was on each bandwagon. (See section 3.3 in Russ Allbery's evaluation, for example.) They didn't make any deeper evaluation extending to things such as the way that bugs got responded to in practice or what design steps people were taking to limit bugs.
Neither process was extensive enough to have spotted a User=0pointer gives superuser rights bug.
I didn't see Russ' evaluation, but I get the point.
systemd is a task manager (which happens to manage init along with everything else it can and often cannot). Referring to systemd as an init system is just muddying well-tread waters by people who are lazily misunderstanding the technical issues. The problem with usernames is that there is no standard so schemes are OS specific. It's not really a systemd problem at the core. The systemd problem (because it almost always is badly implemented) is that it uses it's own username validation scheme that defaults to permissive root when the systemd scheme is not matched for configured services. Running on top of heterogeneous environments, you get inconsistent behavior (even when you comply with the OS). This would be like apache spawning handler threads as root if it doesn't like the username it's configured to run as, regardless if the username is valid under OSX and invalid under Fedora. Too bad, username schemes are now dictated by systemd, is Poettering's position.
Yes there is: https://github.com/systemd/systemd/issues/6237#issuecomment-...
In regard to usernames, I can definitely forgive Poettering's initial failure in design to confusion. His assertion that he is following "the standard" fails on both a practical and logical level.
The problem isn't the format of the usernames, it's that they detected an error, then ignored that the error happened while doing the opposite of what the user wanted. Failing the unit with an error would have been fine. Alternatively, proceeding with the indicated setuid(2) while warning about an invalid username would is also a sane response.
You might be able to modify your descendants' environment, but the rules under which that can happen are maintained and enforced by the kernel. It seems to me that if user space programs can do undesirable things, that's the fault of the kernel itself. To be sure, setting those boundaries and rules is one of the kernel's main jobs.
It is not a secret that kernel devs have a love-hate relationship with systemd, and have for years. That's why Linus said, "You all presumably know why." He was speaking with context that everyone on the LKML already knows.
First result when I google it (but it's been a thing for a long time and on many fronts with many interesting bugs and arguments): http://www.phoronix.com/scan.php?page=news_item&px=MTY1MzA
I don't really subscribe to the "systemd is terrible and is the worst thing that ever happened to Linux" theory, and I'm not trying to be one of those folks that takes every opportunity to make a thread about how much systemd sucks. I like systemd. I mean it when I say that. But, in this case, it seems obvious it is about systemd. I have a hard time coming up with another plausible theory.
Edit: Also, now that I'm thinking about Linus and Poettering butting heads...well, that's hilarious. A couple of brilliant assholes arguing is funny to me. I dunno how much it's actually happened (and how much happened through proxies), but still. Funny.
What did you mean by this? It seems like a complete nonsense statement in the context of user space breakage of the form the kernel is responsible for.
The main clash between the kernel maintainers and the systemd maintainers was over systemd reading the kernel argv in an error prone way to trigger debug information. Other than that the interactions seem completely minimal. Linus Torvalds uses systemd at work and at home.
Does it? systemd is the init, the service manager, the interface to D-bus (and kernel devs had a bit of a tussle with the systemd devs about kdbus), the logging interface of most services (possibly including kernel logs), etc. systemd interacts with both the kernel directly (as the system init, as one interface to cgroups and namespaces, etc.) and with users (as everything else it does). I like it, but I understand the Borg accusations.
"The main clash between the kernel maintainers and the systemd maintainers was over systemd reading the kernel argv in an error prone way to trigger debug information."
Searching the LKML for systemd seems to indicate otherwise. It seems like there have been a few scuffles over the years. I don't subscribe to the LKML or read it religiously and haven't for many years, so I'm mostly guessing based on context, but the context seems clear to me.
Again, I ask, if not systemd, then which init is Linus talking about?
The issue isn't him not knowing what systemd is, or not knowing that that is the init system Linus is talking about. He is aware of both.
His issue is with your claim that systemd has broken userspace. You haven't seemed to provide any evidence of this being the case.
But, I'm happy to take the "broke userspace" assertion back if you don't like the term. Let's just say systemd has had a lot of bugs, sometimes stupid ones, and sometimes stubbornly held on to despite kernel devs or other devs asking for changes. I would hope we're not arguing about whether systemd has had an exciting number of bugs and compatibility issues (I think we can all agree on that, even if we like systemd).
I inferred a bunch from my recollection of how the LKML has talked about systemd in the past. I assumed Linus was talking about systemd and that when he says he doesn't trust it, it was about systemd being sloppy and stubborn about compatibility and fixing bugs. I could be wrong about that.
systemd doesn't really have the power to completely break userspace in the manner Linus refers to, though it certainly has the power to break a system.
(One of my favorites was when a systemd update shut down dhclient but did not restart it, so once dhcp leases expired, servers lost connectivity. It was fun seeing several thousand machines go offline all at once. "Fun.")
If definitions are inconsistent in the minds of people participating in a conversation, it can muddy an issue and make people needlessly argue because they aren't communicating effectively about what they mean. It also makes it more difficult for a reader to understand the debate.
Specifically, in this discussion, when I said "breaks userspace," I and others were referring to the term of art as developed over the past 25 years, and I expected (this being HN) that others would also understand it as such and discuss accordingly. Instead, this thread has turned into a morass of misunderstanding and confusion, because still others believe the term means something else, and are arguing a tangential point as a result.
Systemd and the whole controversy around it is anything but funny. It's upsetting and it already divided the community.
systemd has some great stuff in it. I have my nits to pick with it but, on the whole, it's better than what we had before. And, don't we all just want things to be a little better every day?
I think it's all gonna work out fine in the end (and if not, we'll all be dead someday, anyway).
A behavior that is even default Alsa these days if it detect that you use sound hardware without mixing.
The one "issue" that PA "fixed" was that of temporary audio devices (bluetooth, USB). But then i find the whole notion of USB headphones (a pair of headphones soldered to a minimal USB sound card) a massive abomination. and i fail to see why paired Bluetooth audio devices are exposed directly into the /dev tree rather than behind the Bluetooth dongle device.
https://fedoraproject.org/wiki/Changes/KillUserProcesses_by_...
Their reasoning is pretty good, IMHO. It provides more control at low cost, and it's a one-line configuration change if you want the old behavior. But, everything I need it for (pretty much just screen and tmux) will already be aware and will handle setting up the new systemd unit automatically.
Not only that, but there was an existing mechanism for this. On a disconnect, SIGHUP would be sent to all processes. A background process can explicitly ignore SIGHUP in order to remain running on disconnect. I would be fine if systemd were sending SIGHUP to processes, because that would fit with the existing method of opting out. Sending SIGKILL, and having to use systemd's API to opt out of being killed is ridiculous.
This adds more complexity for no added benefit. My standard .bashrc now needs to have a line checking whether KillUserProcesses is enabled, so that I can bug the admin to disable that idiocy if it is.
The problem with SIGHUP is that a background process can't explicitly ignore SIGHUP, it can only implicitly ignore it. If you send SIGHUP to a process and it doesn't exit you have no idea if it didn't exit because it wants to hang around or if it didn't exit because the process is hung.
Part of the motivation for fixing that is to eliminate things like gnome-keyring-daemon hanging around after the user logs out.
Since gnome is the one that requires behavior and integration with systemd beyond what can be done with signals, a reasonable workaround would be to have the extra logic in systemd apply only to gnome. As it is, they are changing to a default that is entirely unreasonable outside of a desktop environment, and requiring others to work with the new system.
As for the Gnome project though, that's just an example. I'm talking about cases where the application should terminate upon receipt of SIGHUP but doesn't. That default isn't changing, the method to avoid the default action is. Back when SIGHUP was created all software that wanted to ignore it needed to make a handler to avoid the new default behavior. If you want to be able to kill misbehaving processes that were supposed to terminate on SIGHUP then the design needs to change. There's no way to fix this without making some breaking change.
As to requiring others to work with it, your distribution is the one that's making you do that. systemd upstream has had killuserprocesses for a while now. It's just a configure option when building systemd. It's not like your distro is having someone else build packages for them, setting appropriate config options is entirely on them. Some distros are building systemd with that on by default, and some aren't.
I'm fine with having KillUserProcesses exist as a concept. There are some odd machines, like public terminals, where you would want to forbid any long-lived processes. I'm not okay with it being the default option. Yes, the distributions can change their default, but systemd's default in an endorsement that shouldn't be there.
The only things that will ever create a new scope are things like tmux, screen, and maybe some variant of nohup.
Here's the kinds of problems that misbehaving processes actually cause.
https://bugs.freedesktop.org/show_bug.cgi?id=94508#c10
And yes, KillUserProcesses reliably fixes crap like "hp-systray" from hanging your whole session in a way that isn't obvious to the user. If there's a deadlock when trying to log out, somethings gotta give otherwise your session will just persist forever.
False. People ran such softwares via the nohup command. No program modifications were needed. They still aren't.
Can't. Fucking. Wait. Seriously, that day cannot come soon enough. systemd might have some useful ideas, but their implementation is crap, and the overall intrusion on how to get stuff done is significant.
Define brilliant. One changed the computing landscape by starting (and running for decades) the biggest opensource copyleft project in the history of mankind; the other is a bog-standard corporate engineer who simply leverages his parent company status to piss off most people and get away with it. Placing them on the same level is an insult to Linus, who might occasionally be an asshole but is still the most effective leader the opensource community has ever seen.
EDIT/NOTE: I don't have opinions on either Linus or Lennart. I use both of their software daily.
He will gladly chastise himself for errors etc, and is quite humble in public.
Poettering on the other hand have a history of hubris and assholeness.
Consider for example that he not only heckled a presenter at a conference. But when the presenter bowed out early he climbed the stage, beer bottle in hand, to grab the microphone.
It would not surprise me of Poettering have never met hardship during his school years (apparently the signing scheme used in journald was based on his brother's doctoral thesis no less) until he hit the net with his projects, and simply do not know how the handle negative feedback.
Never mind that they have faced off (sort of) on camera at least once, with Poettering basically missing the point again and again that not just Torvalds but also other long term kernel devs on the stage was raising.
runc has been broken many, many, many times by systemd.
I agree with your dislike of systemd, I disagree with the term you used.
By your argument, whenever any program used by anything else has a bug, that's breaking userspace -- which defeats the "userspace" qualifier if every breakage is breaking userspace.
userspace was working, systemd got an update, userspace now not working.
And the systemd opinion on breaking userspace[1] is that compatibility doesn't need to be maintained:
>> The kernel's policy is "don't break userspace" - isn't the init(1) equivalent "don't break the rest of userspace"?
> No it isn't. We need to break things where progress in the base OS is more worth than compatibility.
[0] https://lists.freedesktop.org/archives/systemd-devel/2011-Se...
[1] https://lists.freedesktop.org/archives/systemd-devel/2014-Ja...
$ nohup ./long_running_process &
$ logout
You could interpret this as "breaking user-space", as it's contrary to long-running practice and user-expectations.
Or you could say that systemd is superior as it's reliably cleaning up unwanted lingering processes which may have bothered some before. And this is worth the hassle of explicitly running intended background processes via systemd-run(1), or adding code to e.g. tmux/screen doing the equivalent via the dbus-api.
* https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=825394#221
The whole purpose of nohup is to get your process to ignore your session ending. Why on earth systemd would up and decide that you didn't do that on purpose is beyond me.
When people say they hate systemd, it's precisely this kind of thing that we're talking about.
Systemd does give us extremely quick boot times, though! Too bad I generally don't reboot my Linux systems more than once a month.
On the other hand, error handling is pretty awful, especially during shutdown: Every now and then systemd would decide to give that one unresponsive sshd process a generous couple of minutes to shut down. Or will wait for a random filesystem to unmount itself for another 90 seconds. Sometimes outgoing systemd will lock up completely, until you do the magic ritual of pressing ctrl+alt+del 7 times within 2 seconds all while spinning around on your toes trice counter-clockwise and barking - to force "immediate" restart. Better yet - on special occasions this ritual devolves into farce: _after_ hammering ctrl+alt+del at machine gun rate it will print "rebooting now" and... lock up again. Until you hammer ctrl+alt+del again - only to see another empty "rebooting now" promise. SysRq would - of course - have no problem at all remounting filesystems ro and rebooting from this sad limbo.
Systemd is not about building refinements on unix, it is about turning Linux into something like OSX or Android. a OS that use unix semantics only for bootstrapping convenience.
Effectively they are turning the Linux kernel into a convenient source of drivers, but beyond that could not care one bit about unix.
I really wish i could find the interview again where it mentions that Poettering in the past have advocated throwing away the chapters on unix programming from one of the seminal works on programming unix and Linux.
You need to look at some of the history here. Over the course of the past six years, Zbigniew Jędrzejewski-Szmek has twice cycled around the same loop, once in 2011 and once in 2016. Xe goes to the developers of tmux, tells them that tmux needs to change to be systemd specific, and the developers of tmux ask why systemd cannot provide the same semantics for interactive login sessions that have been employed for the past 38 years, with HUP for session hangup and TERM/KILL at system shutdown; or why at the very least xe does not talk to the C library people. Zbigniew Jędrzejewski-Szmek goes away, and then comes back about 5 years later with the exact same thing, unchanged.
Unfortunately, buying into that Fedora wiki page is as unwise as buying into the rather erroneous Freedesktop explanation of how version 2 cgroups work. You will note that the wiki page dates from around the same time as Zbigniew Jędrzejewski-Szmek's last cycle around the loop, which hit the headlines last year.
* https://news.ycombinator.com/item?id=11797075
I couldn't find any mention of it in Zbigniew's blog.
[0] https://news.ycombinator.com/item?id=11785977
[1] https://news.ycombinator.com/item?id=14665518
Occasionally when I would do so, people would get confrontational about it, so I mostly stopped. But now that means I sometimes waste time trying to find someone's gender even though that's totally irrelevant, and/or try to work around needing a pronoun at all, and/or use "they" even though I don't like it. None of these solutions is great.
(Btw, I don't parse you as confrontational. Comments like yours wouldn't have bothered me.)
Not that I was a huge fan of ALSA either. It always felt a bit over-engineered for my taste. I can understand needing something like that if you want to make professional audio equipment with the Linux kernel but for my desktop use case OSS has always worked just fine for me and it didn't randomly break after an update.
As for systemd, it's bearable and it mostly does what it's supposed to do, but it still feels a bit unnecessary and its security track record is not stellar so far. I prefer the simplicity of the BSD init personally.
why do you think linus is complaining? because he can't get sound to work on his own box or because kernel invalid bugs are over the roof thanks to bogus init?
There is a vast gulf between two communities, people who run Ubuntu Beta for fun on their own PC, and people who run 10,000 RHEL boxes in Prod, and both communities tend to talk "at" each other and misunderstand where the other is coming from.
The scenario you describe is something that would make one of the latter community recoil in horror.
It's easy enough to understand their point of view: run through a Linux From Scratch install. You'll have to write your own init system. You'll be duct-taping a lot of things together, and your solution—while it might be something you're "proud of" in some sense—will definitely be lacking in engineering, no matter how much time you as one individual could ever throw at it.
After the LFS experience—which is essentially the distro creation experience—coming back to regular (pre-systemd) distros made one feel rather different: it was very easy to see how each distro was, at its core, just someone else's duct-taped-together init system with a pile of hacks accumulated on top. There wasn't any real Engineering; there was no shared knowledge-base of best practices for doing this or that. Each distro was a evolutionary Galapagos island.
From the LFS perspective, an init system that exists as its own standalone FOSS project, containing contributions from hundreds of interested parties and battle-tested engineering choices after coming-up-on-a-decade of use in virtually every scenario, is a wonderful thing. systemd is, finally, a reference implementation of an init system "at scale." If you want to build your own init system, you don't have to look at the init systems of 30 different distros to figure out what things each distro just did its own way for no reason, any more (which used to be "most things.") You can just start by looking at "what systemd does", and figure things out from there. Or you can use some systemd components, and some of your own.
I work on a distribution (openSUSE) and don't hold systemd in high regard. There are many people who work for distributions and don't feel that way. Of course, the general openSUSE community decided to go with systemd, but it's not fair to claim that everyone who contributes to distributions that use systemd must therefore like systemd.
The reason I don't like systemd is that in my experience, systemd forces you to integrate system management programs (such as container runtimes) with systemd. And when you are forced to integrate, systemd then causes you nothing but issues. cgroups are a perfect example of this. In addition, systemd doesn't obey SIGPWR and as a result LXC has to write special code for systemd -- which is just ludicrous.
I've talked to some other people from other distributions and they feel so strongly against this that it's possible you'll see a new init system in a few years that rectifies the problems with systemd.
> an init system that exists as its own standalone FOSS project, containing contributions from hundreds of interested parties and battle-tested engineering choices after coming-up-on-a-decade of use in virtually every scenario, is a wonderful thing
There are several things wrong with that description of systemd. To be clear, the pre-systemd init systems were also horrible. The original idea of systemd which was a declarative and parallel init system was a good idea. The problem is that systemd has far out-grown being just an init, and it has been shown historically that it is far from "battle-tested engineering" (with the additional note that the maintainers are strongly against people allocating CVEs for their project).
Several years ago Red Hat decided they also wanted something like initramfs-tools but which themselves had developed (a remark I recall that the developers made on the mailing list), so they built a new system called dracut that models very close to initramfs-tools, in which Debian's response seems to be: "Great, someone else want to support that complex tangle of dark magic, let's use theirs!", so instead of competition I suspect things will return to a monoculture again. A good thing is that during all this, distributions like openSUSE has also switched over to dracut which hopefully shows a trend that the age of large shell scripts in the boot process seems to be over.
It does take a lot of time, but one is protected from a lot of distribution nonsense.
I'm a distro-maker (Exherbo) and while we provide the option for systemd I wouldn't say anyone is exactly thrilled by it. Users use a variety of init systems including handrolled ones, and that choice is the users'. But we recognize it as the sensible current default and is too much effort to work against the grain here.
Now read https://news.ycombinator.com/item?id=11550802 and https://news.ycombinator.com/item?id=10357589 . (The mentioned van Smoorenburg rc script has since moved to https://github.com/dun/conman/blob/master/etc/conman.init.in .)
The TrueOS people have replaced Mewburn rc with OpenRC, retaining the FreeBSD init as far as I am aware. I suggested to them back in January 2017 that since OpenRC has s6 integration, they might do well to add s6 to that to gain full service management. I never received a reply. I haven't heard that Laurent Bercot was contacted, either.
* https://news.ycombinator.com/item?id=13458186
* http://www.mail-archive.com/supervision@list.skarnet.org/msg...
I personally use the nosh system and service managers on FreeBSD and TrueOS, of course. I released version 1.34 last week.
When I did that job, I would estimate maybe one day a year at most. But core functions just stopping working and no one knows why would have been a very big deal
As a sysadmin however, I hate it so much. Not because I have to learn a new system which is nice, but because it breaks so much things. I had to rewrite scripts that just worked for years, I had to do a bit of black magic to make old software behave on newer distribs and it's just so frustrating. I get it that most of these things were not done in the right way but FFS, don't break my production stuff.
I can trust someone who makes mistakes. I can't trust someone who refuses to accept that they regularly make mistakes. Nor can I trust their work.
Bugs happen, that's understandable, but simply the fact that they turn logs into a binary format shows their philosophy is different than the one I expect from a unix-like OS.
And almost all distros are following them sadly...
Sorry, this is a complete BS "solution"
You don't keep a broken piece of software running then put more redundancy to make it kinda stable.
Redundancies are for when things break, not for when they're broken from the start
Nobody takes off from a plane with one inoperative engine, both have to be turning, so in case one fails the other one can take over
http://www.commitstrip.com/en/2015/07/08/true-story-fixing-a...
And while systemd may have started on the desktop (likely to "solve" the speed of repeated bootups when using a laptop "securely") increasingly its development is dictated by this cloud mentality (as is the development of various distros, as someone basically shouted to my virtual face over at LWN).
EDIT: I don't know why I'm being downvoted for asking a genuine question. I'm not a linux expert, so I'd love to hear your criticisms.
Aside from that, using it on my personal laptops has worked surprisingly well. Once I familiarized myself with the CLI tools, I haven't had any issues.
It's not on AWS. Your use case is just a hobby or one man shop or just one data point. It doesn't necessarily scale well for other use cases.
I think what other people are pointing out is that companies aren't lining up to use it or replacing it over a linux server or window server.
All you've got to do is tweak the PLIST.
Read the manages. They're out of date.
Looked at the PLIST tools (not my thing). Doesn't seem to be an option.
There used to be, back when PLISTs were actual files you could edit. That's gone away now though, best I can tell.
Got a Linux box up and running, that now accepts remote logs.
I can't say I blame you, though. Figuring out what all these tools did and what they were for required a fair amount of reading and experimenting on my end.
The problem is more about the project itself. The scope keeps creeping to consume more and more things. The developers have a "we know better, do it this way, your argument is irrelevant" type attitude to outsiders.
"Systemd is great! But it's an inconsiderate piece of software!"
Meaning it work nice, it does it's job it does it fast. But if you want to things your way, or want to run something not-systemd standard, you are in for some fun.
There are also technical reasons, but those tend to attract less of pure anger.
That anger comes from somewhere. We don't just go around picking random people to become angry with. Poettering has a lot of influence in core components of Linux (dbus, pulseaudio, systemd) and he's a bad developer. His designs are horrible, his implementations buggy and shoddy and don't fit into the overall architecture of the rest of the system. So, yeah, there's anger towards him and I think it's more than deserved.
> Also some don't like change, react aggressively
We like change just fine if it brings improvements. If not, we don't like those changes. Again, we don't go around hating on random changes.
>There are also technical reasons, but those tend to attract less of pure anger.
All of the anger is caused by technical reasons. Whether it's shoddy programming or shoddy reasoning about technical issues on the systemd developers part. Almost no-one personally knows Poettering or the systemd developers, yet we still think their software sucks.
There are all kinds of people and plenty in Linux community do pick up on people merely for not be deferential enough to authority. Or for asking question they find stupid (or don't know answer for). You can get flamed quite easily.
Not all, probably not you, but a lot of people there get quite aggressive quite easily.
> We like change just fine if it brings improvements.
I personally know people who get outrages over changes whenever change happen. Maybe not you, but you are not everybody.
> All of the anger is caused by technical reasons.
No, not all of it. A lot of it is cause by arrogance which is not technical reason. You don't need to know someone personally to get that feeling. A certain amount of the anger has faulty reasoning on itself.
Thing is, launchd's developers made a good effort at preserving backwards compatibility, but it's far from perfect. And most online docs (such as launchd.info that you're linking to) is subtly out-of-date. For example, I see no mention of "service-targets": new launchd learned about "bootstrap domains" (huh?), which are namespaces implemented at the Mach (macOS kernel) level. It's not a Unixy thing. Trick question: how do you define a user job that runs in the background even when the user's not logged in? How do you define a user job that runs if the user logged in via SSH (ie not GUI)?
It's late and I'm tired, I can try being more coherent tomorrow. There are some good sources of information hidden out there, but the first few Google responses are woefully obsolete. And fuck Apple's own docs.
[1] I'm not sure what the reasons are, but I suspect for a) merging iOS and macOS functionality and b) tightly integrating it with the kernel's XPC interprocess comms infrastructure (again, for iOS?)
I had to do some provisioning work with NFS and pf and the docs were maddening. I never actually got VirtualBox and NFS to play nicely but once I got a couple little quirks of pf it was pretty reliable.
Well, you can't. Only administrators get to define things that run when the user isn't logged in. That's an intentional decision.
> How do you define a user job that runs if the user logged in via SSH
Offhand, I don't think you can do that either, though I haven't researched this particular angle. And it would be surprising to me that SSH'ing into a machine would trigger daemons anyway.
You might want to find out about systemd, then, which does exactly this. (-:
A PAM hook indirectly causes a per-user systemd instance to be created at login and terminated at logout, which triggers all of the per-user daemons that the account has.
If one has a user that is used for lots of little SSH sessions, hundreds of thousands of times a day, the overhead of starting up and tearing down the per-user systemd for each one, not least in terms of the log noise, becomes a serious consideration, and one has to learn the systemd mechanisms for making the user a "lingering" one.
I've not mucked much with OSX admin, but the times I have, it's been fairly ugly. PLISTs particularly so.
/facepalm
Really? I read up on launchd a year or so ago to setup backups and I though the documentation was amazing...Which docs are you referring to?
For example, in the launchctl(1) manpage, can you elucidate what they mean for the "bootstrap" subcommand? (caveat: you're not allowed to use "bootstrap" in your definition. That's tautological).
What does "kickstart" mean? (again, tautological definition)
What's the significance of the various "domains"? What's user vs login vs gui domains? What's an "audit" session ID?
How do you reload a service's plist file?
Trick question: how do you kill your GUI by mistyping a common launchctl command? (I don't remember the command offhand, and am unwilling to experiment right now)
What does the LimitLoadToSessionType plist property do? What are the possible values?
etc...
My guess is, it's not as relevant because Mac OS isn't as widely used in servers as Linux.
Systemd is effectively Freedesktop/Fedora/Gnome dictating how every last Linux distro is supposed to behave. And doing so by invasive dependency chains.
Observe how LFS now have two parallel books. One continuing the one they have maintained from the start, and another dedicated to systemd. This is a good indicator how invasive systemd is in the basic construct of a Linux distro.
In theory, yes. In practice, systemd reinvents half of the rest of the system, poorly. Even to the point of shipping a half assed dns resolver and a dhcp client that doesn't even support lease renewal, and when people file bugs the invariable response is to blame the user.
BSD's choice to drop everything into a single rc file is not particularly elegant, but it can be coerced to do what you want. The sysvinit user interface, frankly, isn't bad:
/etc/init.d/<service> [start|stop|reload|restart]
/bin/sh -x /etc/init.d/<service> [start|stop|reload|restart]
I'll add that my distro of choice is Debian (or derivatives, which I generally mean here). One thing that I noticed when first switching to Debian, back in the mid 1990s, was that its init scripts were vastly easier to follow than Red Hat's. Debian has a fairly standard structure (/etc/init.d/skeleton is the bare-bones template), and most scripts follow that within some reason. There are a few exceptions -- scripts which initialise things (bootclean, mount, networking) rather than start services, tend not to look like service initialisation scripts, on the inside. On the outside, though, they do.
Further: once configured properly, those scripts don't change. Debian in particular does an extraordinarily good job of extracting out all necessary configuration to an /etc/ configfile (which it then won't touch unless you specifically tell it to).
My experience on MacOS (see elsewhere in this thread) is that stuff which ought be easy isn't. My experience with systemd, to date, hasn't evidenced this, as I am largely using the extant sysvinit scripts. But I've noticed numerous things Wot Used To Work ... do not. And the response from the sysvinit devs, particularly from the top, as well as various sycophants on numerous fora, has been exceedingly disturbing.
Case in point, my single most controversial Reddit comment:
https://www.reddit.com/r/linux/comments/2dvmdn/what_do_you_a...
(No comment on the general tendency of crowdsourced moderation systems to vote for popularity rather than truth.)
... is a description that is out of date by coming up to 2 decades. Mewburn rc came along at the start of the 21st century.
That's ... over a decade out of date.
Re-pitching my comment at that specific configuration: it wasn't elegant, but it could be coerced to do what I wanted.
With the sole exception of MirOS BSD as far as I am aware, the world has actually let go of that idea.
Cheers.
In recent memory, off the top of my head, it messed with NTP, DNS resolution, Syslog, DHCP, Nohup/Tmux, not to speak of taking over udev so it won't run without it (thanks for eudev).
Look upthread at the 'nohup and logout' "bug". That's not something you're going to encounter on MacOS, because it's not something you do on MacOS.
The problem with systemd was that there was potential to be better; the idea is good but the core developers reek of hubris.
Launchd is OSX, and osx has some horrors but that's ok because they're desktop systems and you don't work on fleets of thousands of them usually.
There was a time where Apple were threatening to upstream it to FreeBSD- but aside from that I think the consensus is that launchd is quite horrible too. It's just systemd could have been so much better and not tried to gobble up the entire Linux ecosystem.
I won't claim to speak for everyone here, but there's almost nothing I install third-party that I care about during the startup process (on my laptop). And even if I did, they're generally closed-source apps I can't access the source of anyway. 99% of my day-to-day usage of a laptop is with applications I'm manually starting and stopping.
I care VERY much about services automatically starting properly on a server. 99% of my day-to-day usage of a server is applications that automatically start and stop.
If you are running in a generic "This is my machine and I just did the normal install for a regular use case then everything works" it seems great. But the instant you try to do something a little out of the ordinary it becomes a absolute nightmare of opaque message passing interfaces and hours on Google.
A few examples: 1. In the old days you could network boot a machine with a kernel compiled with a couple of flags turned on and an NFS share. SystemD systems can be network booted, but you're going to spend a lot of time poring over obscure forums finding all of the tweaks you need to make, and it will be a lot slower and toss up error messages for things like UUIDs changing on different systems.
2. If you want to have the wireless card change MAC addresses before it connects for slightly improved anonymity at public hotspots, you had better be willing to upgrade to the bleeding edge. NetworkManager broke the existing solutions for doing this and implemented it's own model, but it required a version of wpa_supplicant that was never really released so you had to build it yourself from a snapshot and install it and tell your distro not to update it. The SystemD devs got tired of waiting for the wpa_supplicant devs to release the new version so they moved the functionality into their code. At the same time they pulled in the DNS resolver but broke it so it crashes on literally every other name lookup.
3. NetworkManager pulled in OpenVPN support as an option. It's awesome when it works, but when it dies it refuses to give you any sort of clue as to why the connection didn't work. You just get a generic "it failed" message like in Windows. Not "Authentication Failed", "Cannot connect to Server", or "Clocks unsynchronized", not even a cryptic error code. Every single error results in just a generic message and you have to either hope the logs have something useful or more often you have to fire up Wireshark and inspect the authentication exchange by hand to figure out what is up.
Lord help you if you find yourself with a machine that hangs at boot because some process didn't receive some message it was expecting and now you have to figure out who was supposed to send that message and why they didn't send it. This isn't the old days where you could trace shell scripts to figure out where the problem was, now you have to know the whole thing before you can fix anything.
And don't get me started about moving the syslog into some binary database thing that you can't grep without going through some export process.
Linus is clearly very smart, but user-facing tools is not what he's good at. Git is a perfect example of this.
Mercurial and Git were created ridiculously close to each other, and their main features/use model are quite similar, and yet mercurial is literally years ahead in terms of usability. The results to a given command are 99% of the time, what you would intuitively expect.
Now here's what I would love to see: Linus initiate a fork/rewrite/whatever of the init part of systemd, that keeps the concept of non-executing service definition files. They don't have to be compatible with systemd but that wouldn't necessarily be a bad thing either.
Edit: actually, aiming for compatibility probably means having to reimplement systemd weirdness, so probably is a bad thing as I think about it.
Git is a very powerful, flexible, reliable and efficient tool. I'd be over the moon if someone wrote an init system holding similar values.
If you haven't had those issues, you're one of the lucky ones.
Edit:
> Git is a very powerful, flexible, reliable and efficient tool.
I didn't say it isn't any of those things. I said it's harder to use than competitors such as mercurial.
> Furthermore, the proliferation of git wrappers says something. Mercurial has a lot of users too (Facebook), and guess what, they don’t write wrappers for it. They do write aliases and extensions, using hg’s established customisation mechanisms, but they don’t feel like the entire UI is so terrible that it has to be completely replaced by a different UI
Then let's consider this snippet, from the people that literally wrote the book on Git, in the Chapter "Git Internals" [2]:
> You may have skipped to this chapter from a previous chapter, or you may have gotten here after reading the rest of the book — in either case, this is where you’ll go over the inner workings and implementation of Git. I found that learning this information was fundamentally important to understanding how useful and powerful Git is, but others have argued to me that it can be confusing and unnecessarily complex for beginners.
If you don't know the internals, you don't really know git. I use git for client work on at least a weekly basis - not necessarily daily. I know enough to use it, and enough to know when I need to check the manual or search for something. I don't know the internals, and it's wrong that we 'need' need to.
[1]: https://lobste.rs/s/klqjwc/two_commits_wrecked_user_experien...
The point was that there are a lot of git "wrappers" created to solve the problem of Git's usability. The same doesn't happen with Mercurial, and it isn't just because of a lack of use, as demonstrated by X, Y, Z prominent/large software companies/projects using it.
I still remember the first time I used "git cherry-pick," I just guessed what it needed from my (then pretty limited) git experience, and it just worked seamlessly. The few little command-line quirks in git (like "checkout" being used to create new branches) are barely worth mentioning.
With all due respect, if you have zero experience with any other CLI solution, are you sure you can evaluate the qualities of git's interface? Nobody said git isn't capable.
We're saying the alternatives are easier to use. You freely admit you haven't used any of the alternatives - you haven't even used SVN from a command line, so you have literally nothing to compare to when using git commands.
My only point being, that while git is not perfect compared to others, even if you don't have experience with an alternative CLI, your experience with just that one CLI can be valuable.
Yeah, on one hand, relative ease of use is important, especially if you're trying to write a tool to replace others. And I'm not saying that is not important. Just that it's important to consider both, not only one.
Compare it to when trying to make a website easier to use. Ok, fine, compare it to other websites too, but your first-timers that have no experience with anything else are important too.
The best survey of Git's quirks that I've seen is, unfortunately, http://stevelosh.com/blog/2013/04/git-koans/
A good book on the subject might make things slightly better. However, it doesn't seem like anyone is interested in writing one :-(
Also "If you flag something, please don't also comment that you did."