- cross-posted to:
- [email protected]
- [email protected]
- [email protected]
- cross-posted to:
- [email protected]
- [email protected]
- [email protected]
You must log in or register to comment.
Void Linux doesn’t use systemd.
I’ve been using systemd on most of my systems since it was released; I was an early jumper to upstart as well.
The thing I don’t like about systemd is how pervasive in the OS it is. It violates the “do one thing, do it well” Unix philosophy, and when systemd went from an init system to starting to take everything over, I started liking it less.
My issues with systemd is that it isn’t an unmitigated success, for me. journald is horrible: it’s slow and doesn’t seem to catch everything (the latter is extremely rare, but that it happens occasionally makes me nervous). There are several gotchas in running user services, such as getting in-session services working correctly (so that user services can access the user session kernel keyring).
Recently I’ve been using dinit on a system, and I’m pretty happy with it. I may switch all of my systems over to it; I’m running Arch everywhere, and while migrating Arch to Artix was scary the first time, in the end it went fairly smoothly.
Fundamentally, systemd is a monolithic OS system. It make Linux into more of a Windows or MacOS, where a bunch of different systems are consolidated under a single piece of software. While it violates the Unix philosophy, it has been successful because monolithic systems tend to be easier to use: users really only have to learn two command-line tools, vs a dozen. Is it categorically better, just because the user interface is easier for new Linux users?
Systemd is modular not monolithic. Distros choose which parts of system d to implement and it just happens to be most of it since its really good at what it does.
You cannot even decouple SystemD from Glibc, never mind separating the various components from each other. It is a bunch of processes but it is designed as a monolith.
It is not modular. This is a lie Poettering keeps pushing to defend building a huge edifice of interdependent systems.
Look at the effort required to factor out logind. It can’t just be used in it’s own; it has a hard dependency on systemd and needs code changes to decouple.
I will repeat that journald is really bad at what it does, and further assert that you can not run systemd without journald, or vice versa. That you can not run systemd without getting timed job control. Even if you chose not to use it, it’s in there. And you can not get time job control without the init part. In most unix systems, init and cron are utterly decoupled and can be individually swapped with other systems.
Systemd is not modular if you can’t swap parts out for other software. Systemd’s modularity is a bald-faced lie.
The one exceptions are homed and resolvd, which are relatively new and were addedlong after systemd came under fire for being monolithic. And, ironically, they’re the components most distributions don’t use by default.
Systemd boot, Systemd network and elogind I’m pretty sure can be decoupled. There’s 69 modules so I’m not entirely sure about all of them just the ones ive encountered.
I know logind can’t easily be, because I ran Artix for a while and they were using a decoupled version of it, and there was a big discussion about swapping it for something else because it was so hard to maintain.
What do you mean by modular though? I assume there’s serious coupling amongst systemd modules that make “modularity” just theoretical
Its like 70 different files and not all are required. You can swap out parts of systemd like run a different init system.
Sure but I wouldn’t say something is modular just because some things are modules. LIke yeah you can swap out networkd, but how about journald?
I would. There are a lot of modules only a few have hard systemd dependencies
I hated it and still do because for a period of years every weird, difficult to find issue on a bunch of servers was caused by systemd. It may be fine now, but I switched to Devuan and have had incredible stability. Poettering’s response to security issues was also terrible and honestly the dude seems like a real piece of shit.
Systemd on a server is disaster.
Only admins with 25+ years of experience may downvote this
Excuse me but wtf? How many machines do you manage?
Right now five servers. Three with Alpine Linux and two FreeBSD servers.
Ok I see. Without any intention to sound offensive, 5 servers is not enough to really see the pro cons of either init system. People handling 50 times those numbers encounter issues where it starts to matter, and those people tend to claim that, while it ain’t perfect, it is a lot better than any alternative
You see, I scaled down a little in recent years because that is not my primary job any more. But, I am working as an admin from 1998. so my word should have some weight, right?
BASED,
Try dinit too. It’s probably the easiest to use alternative to systemd
Though I see Systemd as an improvement, I still do not like it.
The Chimera Linux FAQ captures my thoughts quite well:
https://chimera-linux.org/docs/faq#what-is-the-projects-take-on-systemd
I use it because I’m frankly too dumb to use something else, but if that wasnt the case, i dont think id be speaking fondly of it.
I’m a ram usage fetishist, I absolutely disagree with the “unused ram is wasted ram” phrase that has caught on with people.
I see some of these distros running a graphical environment with only 90mb ram usage and i cream myself. All of them run something other than systemd, usually avoid GNU stuff, and…require you basically to be a developer to use them.
I already run a half broken, hacked together system due to my stubborness, I can’t imagine how fucked I’d be if I tried one of these cool kid minimalist distros.
So you just like having ram doing nothing? Unused ram is wasted ram. Distros cache a lot in ram because they can. I mean why hav RAM is you just want to stare at it and say ohh look at all the free RAM.
Even a system that uses 90mb of ram on a cold boot will accumulate gigs of stuff in cache if you’re using it. (assuming it has the memory for it) That isn’t what people have a problem with though.
Maybe this is an incorrect use of language on my part, but I feel like I’m not the only person who means “memory actively being used by a process” when referring to memory usage. I understand the whole linux ate my ram thing. That just isn’t what I or what I assume a lot of people mean when talking about this.
When I boot up my system, pull up my terminal, run htop, and see 800-1200mb being used just by processes (not in buffer, not in cache), that doesn’t raise any flags or anything, but I also know that some people have gotten their systems so streamlined they use 10x less than that. That’s all memory that could be used by other things. That could be the difference between a low memory system running a web browser or not. Could be the difference maker in a game someone wants to play on their system. There are endless possibilities.
Could be the difference maker in a game someone wants to play on their system.
One reality of the world is: the developers choose what hardware/OS configurations they target. If the makers of your game don’t target your RAM efficient system, you’re outta luck. Developers make their choices for their own reasons, but even with the ever-growing FOSS communities, the majority of developers work for a paycheck, that paycheck comes from profitable businesses and those businesses have very strong influence on what the developers work toward. The businesses only exist because they are profitable… FOSS may not be bound by those realities, but it lives in a world dominated by them.
Unused memory IS wasted memory and my Linux machines, AFAICR, always have buffered everything possible since twenty years ago, it’s not a systemd thing. It also speeds up things, why the long face?
90mb ram
If you’re in a system where 256mb of RAM is the limit, sure - go for the RAM efficient OS options, they’re out there.
Can you even buy less than 2GB of RAM in a desktop system anymore? Even the Raspberry Pi 5 starts at 2GB (and, yes, the older models have less, but I did say desktop system, implying: reasonable desktop performance.) Maybe if you feel the need to use a RasPi 3 as a desktop for something then you should dig around for one of your more efficient OS configurations, but I’ll note… back when RasPi 3 was the new model, Raspbian came default without systemd, but offered a systemd option. The systemd option booted from power off to the desktop (such as it was) in about 1/3 the time.
I’d say the main bad part of systemd is how it’s used and now expected everywhere.
If you search for some Linux guides or install something complicated or whatnot, they always expect you to have systemd. Otherwise, you’re on your own figuring how things work on your system.
This shouldn’t really happen. Otherwise, yes, it’s great, it integrates neatly, and is least pain to use.
There is no authority delegating responsibilities of writing tutorials for Linux. It is the responsibility of nobody and everybody. If you can’t find one for your problem, write it yourself when you have figured it out.
Sure, but I can’t single-handedly create an entire knowledge base on doing everything with X, so it’s a real and big limitation.
If you think so and that this is important, maybe you could be the one that makes it happen. Start a project and gather like-minded people. That is how Linux, FOSS and community driven efforts operate. It’s useless to complain that nobody else makes the effort if you have the capabilities but can’t be arsed making an attempt yourself.
In my opnion, systemd is like core-utils at this point.
It’s so integrated into most things and the default so many places, that most guides assume you have it.
I suggest Gentoo.
Great documentation, systemd optional.
Lol, this is borderline evil advice
But yeah, it works!
Gentoo is an easy OS, and works like a charm.
Alright you broke the edge of evil into a very different realm
I have struggled with Fedora for couple of years (graphics drivers after major updates), then Ubuntu got me down a couple of times (snaps and other malice).
Zero issues with Gentoo after the initial setup. You build it, update it, and IT WORKS. Also you can easily remove parts of software you’re building with USE flags. -telemetry, -x11, and you never care about it anymore.
I must assume Gentoo does require manual intervention here and there, and updates must take eternity for packages that are not precompiled, ain’t it so?
I have an 8 core CPU, but I have to admit I don’t use any DE.
Updates can take several hours if I don’t upgrade for a while, but PC is usable during them (you can set number of build threads).
Manual intervention is what I’ve said needed way more in Fedora, which left me without any video after updates, or Ubuntu which broke integrations or replaced my software.
Gentoo just… is.
There are sometimes updates that would require intervention if you do something special, nothing too difficult though, and you get a link to Wiki with working solutions.
I need to donate more money to that project.
I’m over here still using OpenRC. Mostly because I want to. Some servers I run have systemd on them. systemd is generally nice. OpenRC has finally gained the ability to run user services, which is also very nice.
dinit also has the ability to run user services, FWIW.
I totally agree.
I hate to admit I didn’t want anything to do with systemd because it took me forever to get somewhat familiar with some other mainstream init systems.
Then, I didn’t care for a while until I developed software that had to keep running using some sort of init system. The obvious choice was whatever the default I had (systemd) and I fell in love with the convenience of systemd (templates, timers, …). I started shipping sample systemd with the things I provide & yes, you are on your own if you use something else.
I’ve never used any other init system since I’m relatively new to Linux (8 years of use). So, systemd is all I know. I don’t mind it, but I have this one major issue with it. That “stop job for UID 1000…” Or whatever it says. It’s hands down the most annoying thing I have ever experienced in Linux. Making me wait for 3 minutes sometimes is just insane. I know I can go in and make it wait for 5 seconds
/etc/systemd/system.confor whatever, but why? Also, another one usually pops up.Other than that, I really like how I can make timers. I like how I can make scripts run on boot, logout or login. And I like how I can make an app a background service that can auto start if they ever crashed. Maybe all of this can be done with other init systems? I wouldn’t know, but I like these in systemd
I decided to finally lean into using systemd more while i’ve been using NixOS, since the OS already relies heavily on it anyway. Created targets for my window managers, starting all my programs with services instead of autostart scripts, etc. And it worked fine for the most part, except for some reason, in qtile the systray widget refuses to load the nm-applet when it’s started through systemd. Waybar does not have this problem. I can’t help notice that systemd is not just a little slower, which isn’t the biggest deal in the world, but it also tends to hang more often when shutting down, which is a bit annoying and reminds me of windows lol. Before NixOS i used Void, and while i never really cared too much about what init system i’m running, i can’t help but really appreciate runit for being so simple and fast. I’m thinking of moving back to Void but using the Nix package manager on top. I recently found a solution to the nix driver problem when using it on other distros, so now i should be able to combine the best of both worlds.
It’s refreshing to read to someone that actually says “I was so wrong”
I was wrong also with systemd, I hated it mainly because I already knew init.d, where files are, where configs where etc. Some years later hate is gone, I’m not a power user, but I just now know how to handle my things with systemd and all is good.
You do realize that it’s not a binary between systemd and SysV, right? There are modern replacements for SysV other than systemd, like dinit, OpenRC, s6 and they all strive to address the shortcomings of SysV
And they are all playtoys, unused in production by anyone serious for a reason.
I see most often that it’s the people who live in init.d - interact with it multiple times a day - who are most vocal about systemd hate. I’m going to call “old dogs don’t like new tricks” on that one.
I do get into that layer of system maintenance, but it’s maybe 1-2% of my time, mostly a set-it and forget-it kind of relationship. There was a time when the old ways were easier due to more documentation and guides on the internet, which I lean on heavily because I interact with this stuff so rarely. Those days passed, for me, 8-10 years back.
I’m in Guix Linux land right now and I miss
journald. I’m supposed to wade through all the log files in/var/logmyself??I still have no idea how to find the right record to read but at least I can run a journalctl --follow till my crash happens
If that approach is enough then
tail -f /var/log/*could work too with multiple files, it’ll “follow” all the files and display only new lines.
journalctl is the one part of systemd I really do not like. For whatever reason, it’s insanely slow, taking multiple seconds before it gets around to display anything. It also has all the wrong defaults, displaying error messages from a year ago first, while scrolling to the bottom again also takes forever and consumes 100% CPU while doing so.
There are flags to filter and display only the relevant parts, but not only are none of them intuitive, doing a mistake there just gives you “-- No entries --”, not an error. So you can never quite tell if you typed it wrong or if were are no messages.
Maybe it all makes more sense when studying the man page in depths and learned all the quirks, but /var/log/ kind of just worked and was fast, without any extra learning.
I’m supposed to wade through all the log files in /var/log myself??
You configured your logging. You could have made them all one file.
Uh, sorry. I don’t follow. Is there a way to tell all programs to write to one file in Guix?
Yea, symlink all the files to the same master log file. … /s
I totally agree. I used to hate systemd for breaking the traditional Unix philosophy, but the reality is that a tight init and service-tracking integration tool really was required. I work with and appreciate systemd every day now. It certainly didn’t make things simplier and easier to debug, but it goes a long way towards making a Linux system predictable and consistent.
Poettering can go fuck himself though - and for PulseAudio too. I suspect half of the hate systemd attracted over the years was really because of this idiot.
predictable and consistent.
Or none of those.
Oh. My NIC didn’t ‘start’ because systemd and network manager are fighting again? Neet.
Just use systemd-sudo to replace network manager with systemd-networking
I don’t know why they are downvoting you, it’s true. I’m dealing with this kind of problem currently… sometimes the boot lasts forever to the point that I have to use AltGr+SysRq commands to force kill everything… other times it simply boots as normal. It’s not consistent at all.
At least before with the old init it was relatively simple to dig into the scripts and make changes to them… I feel now with systemd it’s a lot more opaque and harder to deal with. I wouldn’t even know how to approach the problem,
systemd-analyze blamedoes not help, since the times I actually get to boot look normal. But I do believe it must have to do with the mountpoints because often they are what takes the longest. Any advice on what should I do would be welcome.Also, I have a separate Bazzite install in my living room TV, and while that one does not get locked, sometimes NetworkManager simply is not running after boot… I got fed up to the point that I wrote a workaround by creating a rc.local script to have it run, so I can have it available reliably when the system starts (that fixed it… though some cifs mountpoints often do not get mounted… so I’m considering adding the mount command to the same rc.local script too…).
Any advice on what should I do would be welcome.
You can play around with the mount option
nofail, if that’s set, systemd will not wait for the mount point to be ready and continue booting normally. Can be useful with HDDs that take a while to spin up and aren’t needed for the boot process (e.g. backup drives, etc.).Another thing to look out for: SDCards or USB flash drives that might randomly fail to “spin up” and hang, unplugging those helps.
Thanks! I’ll try with
nofailand see if the lockups stop!Another thing to look out for: SDCards or USB flash drives that might randomly fail to “spin up” and hang, unplugging those helps.
Honestly, that could be it now that you mention it… I have had for a while an external hard drive plugged in that I’ve used for some backups.
systemd is easy to work with, other init systems introduce kinks, I rather break philosophy than deal with that shit
Is it really breaking it? As far as I’m aware, it’s more like gnu. It has components and you can select what you use (here meaning distros and packagers).
People mistake this for a monolith because it’s all named systemd-thing. Integration, like you said, was and is needed. But what if all those separate utilities and services are actually disconnected and speak some protocol different to pipe? Does it make it less unixy?
And poettering is an absolute good guy here. Pulseaudio wasn’t perfect, but did it improve things compared to what was there before? Sure it did. Even now, pulesaudio protocol is used within pipewire and it works just fine.
Perfect is the enemy of good. And while all these tools might not be perfect, they are the best in the Linux world.
poettering is an absolute good guy here
Agreed. But he’s also an abrasive know-it-all. A modicum of social skills and respect goes a long way towards making others accept your pet projects.
pulesaudio protocol is used within pipewire and it works just fine.
I wasn’t talking about the protocol, I was talking about the implementation: PulseAudio is a crashy, unstable POS. I can’t count the number of hours this turd made me waste, until PipeWire came along.
I always assumed that Poettering is an arse to people because of the hate he got for systemd. I imagine it’s hard to see the best in people when there’s a crowd of haters everywhere you go. Though I have no idea what he was like beforehand.
People are idiots.
Poettering got death threats for systemd.
Everybody who is hated and popular gets death threats. Hell, even the nicest actors get death threats.
They are easy to write and send, and there’s 0.01% of the population that is mentally unstable enough to actually do so. You and I don’t get death threats because we aren’t popular enough.
Nah bugger that. Famous actors are known by a vast majority of people. It is not normal for open source programmers to receive abuse to the level of death threats. That only happens when you get the attention of kiwi farms types.
Fucking hell, what is wrong with people? Looks at the US. Oh right.
Attitudes develop over time… his may indeed be a defense mechanism or simply a response to all that hate.
I feel that generally, when the issue is that the person is an arse, then the complaints are often not about the software. You might see people campaigning to boicot the software out of spite, but they won’t give you a technical reason, other than them not wanting the creator to get any credit for it.
When the complaints are about discrepancies in the way the software is designed (like it was with systemd), there’s no reason to expect the person to be an arse. Though him not being an arse does not make the criticism about his software invalid… in the same way as him being an arse would not have made the software technically worthless. Don’t fall for the ad-hominem.
Agreed. But he’s also an abrasive know-it-all. A modicum of social skills and respect goes a long way towards making others accept your pet projects.
This isn’t what I get when reading bug reports he interacts in. Yeah, sometimes he asks if something can’t be done another way – but he seems also very open to new ideas. I rather think that this opinion of him is very selective, there are cases where he comes off as smug, but I never got the impression this is the majority of cases.
I wasn’t talking about the protocol, I was talking about the implementation: PulseAudio is a crashy, unstable POS. I can’t count the number of hours this turd made me waste, until PipeWire came along.
PipeWire for audio couldn’t exist nowadays without PulseAudio though, in fact it was originally created as “PulseAudio for Video”; Pulse exposed a lot of bugs in the lower levels of the Linux audio stack. And I do agree that PipeWire is better than PulseAudio. But it’s important to see it in the context of the time it was created in, and Linux audio back then was certainly different. OSS was actually something a significant amount of people used…
Pulseaudio was introduced in 2004. How come it took almost 20y for it to be replaced if it was that bad?
Implementation, being what it is, improved the situation compared to alsa and other things before it. Again, while not perfect it made things better for everyone.
It’s funny that this is a thing attributed to poettering as bad since things before were way worse… why not throw Sticks and stones at those people?
I really don’t get it.
And all of these things are optional. The fact that distro people and companies select them is because they solve real world problems.
Pulseaudio was introduced in 2004. How come it took almost 20y for it to be replaced if it was that bad?
Did you learn nothing from X11 usage? May I remind you that X11 was invented by Xerox in the fucking 80s?!
Bad software attaches itself to OSs like a cancer.
things before were way worse… why not throw Sticks and stones at those people?
My earliest memories of Linux audio were in Slackware in the mid 90s, reading and re-reading the HOWTO that started off with a bunch of attitude about how real computer users don’t need audio, but we can do it anyway “so, if you must hear Biff bark…” and then a bunch of very unhelpful things to try following that never ever worked on any system I ever tried to use them on. Diverse systems that, of course, all played audio through Windows flawlessly.
But he’s also an abrasive know-it-all. A modicum of social skills and respect goes a long way towards making others accept your pet projects.
You mean like Linus Torvalds?
And poettering is an absolute good guy here.
You obviously weren’t actually around when he was granted mini-king status and acted like a jackass to literally anyone who objected to pulse or systemd. As a result, redhat, canonical, and Debian had to eat criticism over pushing these before they were ready… because of “superstar” poettering.
Poettering is a disrespectful clown.
“It’s more like gnu”
You are correct. GNU has the bad habit of only working with itself as well. Systemd only works with Glibc so it fits in well.
The reality is that GNU is just a subset of the Red Hat Linux platform these days. Systemd is another part. GNOME is the other big chunk. They are all designed to work with each other and do not care if they work with anything else.
From what i understand the problem is that while systemd is technically split up into different components, you can’t really use it with the stuff you don’t want ripped out.
Linux audio has been a cluster^$%< of epic proportions since the mid 1990s. At least you can make single application systems work well these days, but Windows has really whipped the llamas ass on the audio front for 30+ years now, in terms of “it just works” user experience - without being hyper-draconian on the application ecosystem.
I had (and still have) way more issues with Audio on Windows then I ever had on Linux.
And I have seen it all, OSS, ALSA, aRts, EsounD, pulseaudio, pipewire and most likely some more that I have forgotten.
It definitely depends on what you are trying to get out of it.
I’ll grant: low lag audio performance in Windows is… dismal. Which is why everyone had conference call lag adjustment issues in 2020, “go ahead”, “no you go ahead”, “ok” - both start talking simultaneously again… It seems better these days, I’m sure that’s at least in part due to training of the conference participants, but maybe they have been working on getting the lag down without too many dropout / stutters.
We have a bespoke low lag audio system that was specifically implemented in Linux even though we put the GUI in Windows because of those lag / stutter issues, years back the audio was done on a dedicated DSP chip, but a Core i7 is more than up to the task on Linux these days.
The Linux audio pains I refer to were: A) audio just doesn’t work at all, and B) audio works, until you start to try to use two audio applications simultaneously - then they start to mess each other up. Both of those were better in Windows long before Linux came up to speed. But a lot of how Windows audio gets acceptable performance is big laggy buffers.
I’ll just go ahead and start the flame war.
I totally agree with the functionality of systemd. We need that. But the implementation… Why the fuck do we need to cram everything into pid 1? At least delegate the parsing into another process, god damn. And could we all just agree that ’systemd-{networkd,resolved,homed}’ don’t really have a reason to exist, and definitely not that coupled to a fucking init system. Systemd-timers are wonderful, but why are we running cron-but-better in pid 1?
We have an init-system where the developers are afraid of using things like processes and separation of privileges. I’m just tired of patching fleets of servers in panic every time Pöttering’s bad design decisions hit the fan with their CVEs and consequences.
The coupling in PID 1 is a bit much. I actually quite like
systemd-networkdfor some use cases, though. It lets me declaratively manage the network interfaces on my headless servers in a way that’s very similar to how I’m managing the services. Sure, it’s coupled tosystemd, but it’s mostly one-way coupling; if I want to use NetworkManager (which I do on my laptop), I can switch over, and nothing in the init system breaks.
What is the Unix philosophy?
I still doesn’t like it…
Because people here accuse Poettering of being an asshole: I’ve read some of his blogposts and seen some talks of his and him doing Q&A: He answered professionally, did his best to answer truthfully, did acknowledge when he didn’t know something. No rants, no opining on things he didn’t know about, no taking questions in bad faith.
As far as I can tell all the people declaring him some kind of asshole are full of shit.
He is not that bad, the issue is that, as all foss devs, he is not interested in solving problems he does not feel like are important.
The problem is, he disapproves when resources are allocated in his project to those problems and one main area he is not a fan of is support for legacy stuff.
It just happens that legacy stuff is the majority of the industry, as production environment of half the globe needs to run legacy software and a lot of it on legacy hardware
He answered professionally
Until you ask him about security and CVEs advisories…
Source?
