Newest Version of Systemd Includes Experimental Feature for A/B-Style Updating (theregister.com) 182
"Let's popularize image-based OSes," writes Lennart Poettering, "with modernized security properties built around immutability, SecureBoot, TPM2, adaptability, auto-updating, factory reset, uniformity — built from traditional distribution packages, but deployed via images."
Or, as the Register puts it, the Systemd Linux init system "continues to grow and develop, as does Linux itself." They delve into the rationale for the new systemd-sysupdate and kernel-install features, noting "The former is still described as an experimental feature, so relax — for now." No, this does not mean that systemd is becoming a package manager. Like it or not, though, the nature of operating systems is changing. Modern ones are large, complex, and need regular updates, and as The Register has examined in depth recently, this means that the design of Linux distributions is changing radically....
ChromeOS doesn't have a package manager; neither do Fedora's Silverblue and Kinoite versions. You get a tested, known-good image of the OS. Updates are distributed as a complete image, like they are today with Android or iOS. ChromeOS has two root partitions: one live and one spare. The currently running OS updates the spare partition, then you reboot into that one. If everything works, it updates the now-idle second root partition. If it doesn't all work perfectly, then you still have the previous version available to use, and you can just reboot into that again. When a fixed image becomes available, the OS automatically tries again on the spare instance.
The idea is that you always have a known-good OS partition available, which sounds like a benefit to us. Presumably the users are happy too: Chromebook sales may be down, and they only have a fixed lifespan, but there are still well over a hundred million of them out there.
So, no, systemd is not going to become a package manager, because ordinary distros won't have a package manager at all, except maybe Flatpak, or Snap or something similar. The new functionality, including managing installed kernels, is to facilitate A/B type dual-live-system partitions.
For some insight into this vision, Lennart Poettering, lead architect of systemd, has described this in a blog post titled "Bringing Everything Together."
Other updates include "changes to systemd-networkd, such as systemd-resolved starting earlier in the boot sequence, and more cautious allocation of default routes," the article points out, adding that new releases of systemd "ppear roughly twice a year, so the chances are that this will appear in the fall releases of Ubuntu and Fedora...
"If you still prefer to avoid systemd, don't despair. There are still a selection of distros that eschew it altogether, including Devuan GNU+Linux, Alpine Linux, and Void Linux.
Or, as the Register puts it, the Systemd Linux init system "continues to grow and develop, as does Linux itself." They delve into the rationale for the new systemd-sysupdate and kernel-install features, noting "The former is still described as an experimental feature, so relax — for now." No, this does not mean that systemd is becoming a package manager. Like it or not, though, the nature of operating systems is changing. Modern ones are large, complex, and need regular updates, and as The Register has examined in depth recently, this means that the design of Linux distributions is changing radically....
ChromeOS doesn't have a package manager; neither do Fedora's Silverblue and Kinoite versions. You get a tested, known-good image of the OS. Updates are distributed as a complete image, like they are today with Android or iOS. ChromeOS has two root partitions: one live and one spare. The currently running OS updates the spare partition, then you reboot into that one. If everything works, it updates the now-idle second root partition. If it doesn't all work perfectly, then you still have the previous version available to use, and you can just reboot into that again. When a fixed image becomes available, the OS automatically tries again on the spare instance.
The idea is that you always have a known-good OS partition available, which sounds like a benefit to us. Presumably the users are happy too: Chromebook sales may be down, and they only have a fixed lifespan, but there are still well over a hundred million of them out there.
So, no, systemd is not going to become a package manager, because ordinary distros won't have a package manager at all, except maybe Flatpak, or Snap or something similar. The new functionality, including managing installed kernels, is to facilitate A/B type dual-live-system partitions.
For some insight into this vision, Lennart Poettering, lead architect of systemd, has described this in a blog post titled "Bringing Everything Together."
Other updates include "changes to systemd-networkd, such as systemd-resolved starting earlier in the boot sequence, and more cautious allocation of default routes," the article points out, adding that new releases of systemd "ppear roughly twice a year, so the chances are that this will appear in the fall releases of Ubuntu and Fedora...
"If you still prefer to avoid systemd, don't despair. There are still a selection of distros that eschew it altogether, including Devuan GNU+Linux, Alpine Linux, and Void Linux.
Let's ditch systemd (Score:5, Insightful)
It does too much, does it badly, and in a non-unixy way. It was a mistake to let it get this far. It'd be a bigger mistake to stick with it.
Re:Let's ditch systemd (Score:5, Informative)
Red Hat got behind their employee's vision. It's had many problems based on fundamental architectural goals of Leonart Pottering:
* The unwelcome and unnecessary binary log formats of journalctl
* The intricate interweaving of the init tool with previously distinct parts of the system, such as logging and file system mounting and networks.
* The unnecessary rewriting of the file system hierarchy to make systemd's components the most critical part of configuration, combined with the inconsistent agglomeration of alternative and interwoven layouts.
* The ever-expanding role, such as attempting to terminate processes of users when they log out, with no record of the termination. This was eventually disabled by default, the concept that the init tools should shut down individual user's processes without trace is indicative of Leonart Pottering's disregard for the ordinary user.
* Leonart's stated goal of creating "immutable Linux", one in which all configuration and system management is in systemd itself and configuration files in "/etc/" are completely locked down. This is in violation of the FSH, and isn't compatible with any operating system but Linux. It's the self-absorbed model of someone who is unwilling to even acknowledge cross-platform development work or the free software world and its collaboration with UNIX environments such as AIX, OpenBSD, or any of the other BSD environments.
Re:Let's ditch systemd (Score:5, Informative)
>Leonart Pottering's disregard for the ordinary user.
As further evidence, note that you can't push ctrl-c on your own system to tell systemd to stop waiting for a stuck Start or Stop job. You know that message you get when your NFS goes down and you attempt to reboot, and systemd decides that it should wait until the heath death of the universe before giving up on a mount/unmount? Yeah, that shit. In the bug report on it (https://github.com/systemd/systemd/issues/838) Poettering dismissed the frustration of users by saying "well systemd might be on a server where the person sitting at the serial terminal shouldn't actually have permission to hit ctrl-c. If you don't like it, tough. We're not adding a configuration option, not even one you have to manually enable if you're using a desktop and want this feature."
Re: (Score:2)
Don't let systemd control NFS or CIFS or any other network based mounting. Use autofs or other automounting gools for network based filesystems. It won't entirely eliminate such problems, but it should help reduce the likelihood of such hangs.
Re: (Score:3)
Re: (Score:3)
I hate to defend the guy,
Then don't. Problem solved.
but what's actually wrong with his solution?
You can't configure it to meet your preferences, only his preferences.
That was in fact a truly dumb question.
Also, note that you can press C-A-Del 7 times within 2s which will expedite the shutdown...
Sweet, bleeding Jesus.
Re: (Score:3)
the systemctl enable debug-shell trick only works if you have access to a shell, which you don't during startup/shutdown at that point in time. If you enable it all the time, as he suggests ("runs continuously from earliest boot on") you're actually introducing a real security flaw to get around his mostly imagined security flaw. The systemd debug-shell gives everyone on tty9 _passwordless root_ on your system! Why does he prefer that solution to an option to allow users to hit ctrl-c?
Re: (Score:3)
Indeed. But lots of people cannot admit having been wrong until all hell breaks loose.
Re: (Score:2, Insightful)
So? Ditch it if you don't want it. Leave the rest of us be. Especially those of us who don't give a fuck about the unixy way (otherwise known as "getting in the way of functionality way").
It was a mistake to let it get this far. It'd be a bigger mistake to stick with it.
And yet everything works, in many cases better. Doesn't fit your use case? Well that just your own stupid fault for picking the wrong distro. Let others be.
Re: (Score:2)
The problem is that everything was fine, until systemd came along.
Right now? I can't have some functionality in KDE without systemd. This is bad, and unneeded.
Systemd introduced a host of "new" (translation: complications) stuff, which only makes things more complicated. On top of that, systemd was forced on many distributions, exactly because some packages require systemd (or, more specifically, a part of it called elogind) for no apparent good reason.
Do I want to be rid of systemd? Yes, absolutely. It's c
Re: Let's ditch systemd (Score:5, Insightful)
What you see as a weakness is really a strength. Next you will want to do away with KDE and XFC because gnome is all you need. Then you will be happy to just go with one web browser.
Instead of having open standards it sounds like you want one way for all and piss off if you don't like it.
Why are you not just a Windows or Mac person if you want someone else to tell you what software stack you are about to run? I'll stick with have choices, thanks.
Re: (Score:2)
Next you will want to do away with KDE and XFC because gnome is all you need
Sadly that seems to be the direction for RHEL. We use RHEL 8.x at work and trying to use KDE on it is next to impossible, xfce some people can get working, but some have issues. fvwm and others, impossible. GNOME3, all works fine.
I am OK with GNOME3, would rather use something else, but personal customizes are being slowly removed from RHEL.
Poettering REALLY needs to be a Windows dev (Score:5, Insightful)
> Why are you not just a Windows or Mac person if you want someone else to tell you what software stack you are about to run? I'll stick with have choices, thanks.
That's exactly it. Poettering is all about building a Windows style system. Okay, great. He should apply at Microsoft. They hire Windows devs. Many of his ideas would fit perfectly with Windows. They are antithetical to *Linux*.
Re: (Score:3)
Why is it that nobody seems to be able to build anything better though? Surely if systemd is so terrible and everyone can see the obvious flaws in it, it shouldn't be difficult to make something that works better, right? In fact, people keep insisting that the old init systems are already better, so there should be no reason to use systemd...
And if systemd is just Windows, why is Microsoft heavily supporting Linux, to the point where it's the most common OS on their Azure cloud service?
If you want systemd t
Re: Poettering REALLY needs to be a Windows dev (Score:4, Insightful)
Why is it that nobody seems to be able to build anything better though?
Not having systemd is better in more ways than having systemd, so the whole idea that nobody can build anything better is bullshit. When I say "nothing is better than systemd" I mean it literally, not figuratively.
Re: (Score:2)
IMC it's led to lost log messages and non-working RAID functionality. Bluetooth audio was working for me before, but I only ever used it with a headset.
Any other init system is better (Score:2)
Pretty much any other init system, Upstart or whatever, is more reliable, more understandable for troubleshooting, lighter weight - pretty much better in every way. In ADDITION, there a lot that follow the Linux / UNIX way of doing things. Everything is a file, do one thing and do it well, make things discoverable / readable, etc.
Perhaps you're asking for a more Windowsy init system?
In that case, see the Windows 10 init. It's even slightly more Windows like than systemd, though only barely. Systemd would b
Re: Let's ditch systemd (Score:4, Insightful)
People who complain about systemd need to just stop using it
Done.
If you write software you quickly realize how bad Linux is with its myriad of init systems.
I seriously doubt you know understand those init systems.
Re: (Score:3, Insightful)
I seriously doubt you know understand those init systems.
I seriously doubt most of the anti-systemd people do. One of the things those other init systems all share in common is the realisation that despite Slashdot's armchair expertise, sysvinit was childish trash as far as init systems go, and the only reason it worked at all is the myriad of other helper utilities that had to be bolted on top of it.
Re: (Score:2)
sysvinit was childish trash as far as init systems go, and the only reason it worked at all is the myriad of other helper utilities that had to be bolted on top of it.
sysvinit did its job just fine. If you wanted more functionality you could add it through things like daemontools which followed the Unix philosophy or not, which is also the Unix philosophy. systemd runs counter to the whole idea of the system by being one big turd that's difficult to replace.
pclinuxos is another distro without systemd (Score:2)
Since they bothered to mention linux distros without systemd, I thought I'd throw in pclinuxos as well. I'm using it now to post this.
Re: pclinuxos is another distro without systemd (Score:5, Informative)
http://www.slackware.com/ [slashdot.org]>Slackware is the oldest Linux distro still maintained, alive and well, that also does not rely on systemd.
Re: (Score:2)
I belive mxlinux does not use it either.
A/B/AB/O (Score:2)
Re: (Score:2)
It's great for appliances like routers, and also pretty useful for non-technical people as there is a clean/automated recovery path. The downside is that upgrading a single package can be harder (or at least unsupported).
Re: (Score:2)
Re: (Score:2)
The benefit is that the complete system can be tested. *CAN* being the operative word; past experience doesn't suggest it is a reliable expectation.
Re: (Score:2)
IBM's AIX has had something similar to this A/B, using split mirror OS environments. At least since 2017, (when I first had to work with it), but probably many years before that.
TrueNAS Core, based on FreeBSD, uses ZFS for boot environments similar to So
systemd-linux (Score:5, Insightful)
There needs to be a name for whatever disorder that Lennart Poettering has. A highly specific combination of hoarding, megalomaniacal attempts to replace anything he didn't create, and a passive aggressive attempt to shit on anything POSIX. Why the hell should my INIT SYSTEM be shadowing my packager manager and doing system upgrade tasks? And before you go "lol it's not an init system, bro," you need to address how it was very specifically adopted to be an init system.
At this point it's undeniable it's a hydra, but instead of cutting off a head to spawn two more, Poettering creates two more features for every bug report he gets that complains about systemd's tendril-like structure. The only disagreement seems to be whether you're better off with or without a hydra in your Linux system.
Re:systemd-linux (Score:4, Insightful)
Well it might be more accurate to point the blame at RedHat who is allowing and facilitating Lenny Poetter with this shit. Which is very typical of how they have run their parts of the Linux ecosystem in general since forever. Everything is non-standard, bizarre, bloated, slow, and often broken.
And please God no flatpak or snap. They are the worst packaging systems ever. I just had to reinstall a system that was corrupted with flatpak garbage and running really poorly. snap is even worse.
If you do want a AIO package system then AppImage is the only sane choice. It's designed from the ground up to work as a regular user, no system-wide corruption BS, no root needed for any part of it, and its "install" is nothing more than a single executable file.
Beside that, the old package managers work fine. I don't know why panties are all in a twist. We've had the ability to manage different library versions in regular package managers since forever.
Re:systemd-linux (Score:5, Informative)
Did you read the article? This isn't meant to compete with package managers. If you use a distro based on a traditional package manager, you don't need this feature and your distro probably won't even enable it. It's meant to support distributions like Silverblue that treat the whole OS as a single unit, rather than a collection of individual packages. There's interest in that kind of update model, because it makes sense for some (not all) use-cases, so systemd is aiming to make it easier to build distros that way. But package-based distros aren't going away any time soon (if ever); nobody's forcing you to switch.
Re: (Score:2, Insightful)
Re: (Score:2)
Did you read the article? This isn't meant to compete with package managers. If you use a distro based on a traditional package manager, you don't need this feature and your distro probably won't even enable it. It's meant to support distributions like Silverblue that treat the whole OS as a single unit, rather than a collection of individual packages. There's interest in that kind of update model, because it makes sense for some (not all) use-cases, so systemd is aiming to make it easier to build distros that way. But package-based distros aren't going away any time soon (if ever); nobody's forcing you to switch.
Yes, the same "systemd won't take over this", "systemd-whateverd is optional, not all distros will use it", "stuff X you're concerned with won't happen"... The past showed that most of the stuff we were concerned with and yet were dismissed condescendingly the same way you're doing now has sadly come to pass.
Making traditional distros go away to replace them with redhat's uniformed one true linux is the end game, it's the plan and it has probably been the plan since redhat started supporting systemd.
redhat
Re: (Score:2)
Really? I must have missed that. I run Debian, and Debian switched to systemd as init because it's useful, but Debian still runs a conventional syslog daemon alongside systemd's journal, and it works
Re: systemd-linux (Score:2)
No, there needs to be an INSTITUTION and MEDICATION for whatever Puttering has.
Re: (Score:3)
There needs to be a name for whatever disorder that Lennart Poettering has.
Providing software his customers want? (note, you're not a customer, you're a silly end user of his customers).
There's a reason systemd has widespread adoption, and there's a reason it does that despite the *apparent* universal hate as voted by Slashdot comment section.
Re: (Score:3, Insightful)
You need a link between the packaging systems and init because you need a standard way of installing software. Have you ever written Linux software and tried to support the different systems? It is a nightmare. Should I use apt? Yum? Emerge? Who cares? It is 2022
That's why you leave it to distribution packagers, of course. Why would overstep your field of competency, especially if you're aware of where it ends?
Re: (Score:3, Informative)
These developers have hated the OSS distribution model for decades because it takes away their
Liven up the holiday weekend (Score:3, Informative)
Nothing like a systemd story to waken up the Slashdot dead. This ought to be fun reading over the long weekend.
Great for systemd users (Score:2)
So you say (Score:5, Funny)
"So, no, systemd is not going to become a package manager, ..."
I'll bet you $20 that it is going to include a package manager before too long. Poettering seems like the exact sort of developer that led JWZ to come up with Zawinski's law.
Re: (Score:3)
My personal theory is that a given programming language will expand until all forms of brackets "([{)]}' have meaning in every context.
Re: (Score:2)
I'll bet you $20 that it is going to include a package manager before too long.
But does it matter? Between docker, snap, apt, and manually installing software there's one thing that is constant across all package managers: You can completely ignore their existence.
Re: (Score:2)
systemd-flatpakd...
I'll bet another 20€ with you that if that A/B style updating manages to get pushed onto big distros, systemd or flatpak will start to depend on the other.
This is the issue... (Score:3)
Modern ones are large, complex, and need regular updates
And this is the problem, if people stuck with simple Operating systems you would not need something like systemd. But simplicity was tossed out with the Bath Water years ago.
I think there are a few systems trying to be simple, OpenBSD for sure and Slackware is doing all it can to try and be simple, but seems to be fighting a loosing battle with freedesktop.org :(
..
Re: (Score:3)
if people stuck with simple Operating systems you would not need something like systemd.
The world stopped being simple. Your computer is not a fixed device operating in isolation anymore. We live in a world where your computer can go to sleep and wake with different hardware, on a different network, with different connectivity, rendering a different GUI display resolution, different audio output device, and accessing different files all at the same time.
And we expect a modern OS to not only cope but to automagically setup everything in a functioning and working way. Shit I remember the "simple
Finally, news for nerds (Score:5, Insightful)
"Let's popularize image-based OSes," writes Lennart Poettering, "with modernized security properties built around immutability, SecureBoot, TPM2, adaptability, auto-updating, factory reset, uniformity — built from traditional distribution packages, but deployed via images."
I don't understand a single word, so it must be news for nerds.
Re: (Score:3)
Shorter Poettering: "To make the cargo planes come back, we need to include a kitchen sink. Maybe even two."
Re: (Score:2)
I don't know but immutability is naturally orthogonal to auto-updating, adaptability, and factory reset.
Re:Finally, news for nerds (Score:5, Funny)
Re: (Score:2)
but of course he didn't think of the exceptions. The next thing he does, is send his hordes of trolls out to every major Linux distribution to scream: "you don't need your package manager, you've got A/B images now, thanks to systemd!" Of course Debian, Arch, Ubuntu, Fedora, and Red Hat can't resist and immediately implement this, killing off their package manager in favor for flatpak/snap/appimage... and then start wondering why the power users are leaving in droves for Devuan, Pclinuxos, Void Linux, Slack
BSD (Score:2, Informative)
Could also avoid systemd by using a version of BSD (Net, Free, Open).
Re: (Score:2)
Agreed, and OpenBSD is doing what this article is suggesting in a much simpler manner via pledge(2) and unveil(2).
No complex flatpack, snap or whatever many other type things exist on Linux
Also FreeBSD has jails which is much superior than what Linux has, but more complex than pledge/unveil.
Comparing to Android and IOS (Score:5, Insightful)
Well, if those are considered examples of Images based systems, then I can't think of any better proof of why *not* to do this. I'm not about to RTFA, but IOS updates are only successful because Apple rabidly controls the hardware. Google does not control Android hardware, and as such Android updates are among the most nororiously farked in the industry. Slow up dates? Lack of updates? That's android for you. If either of those scenarios are likely outcomes of non-package based update systems, then, whoo-boy, keep that nasty shit to yourselves,
Re: Comparing to Android and IOS (Score:3)
He is not entirely right about iOS updates. They are distributed as differential images within a major release. When an iOS device contacts the software update server, it requests the image for the diff of whatever version it has and latest.
"experimental feature" (Score:2)
Hail Bob (Score:2)
Duplicate whole partitions sounds wasteful. (Score:2)
I get wanting a backup when you update, but why not just package up the OS as a SFX archive? Then it's a file like any other. And you can have 2 copies for a bit... Then delete the backup when you don't need it anymore.
I'm not a fan of spreading little crumbs all over the system for one single app, either. Group up your crap so I can look at it or whatever easily.
Lets not (Score:3)
It's an inferior duplicate. I can already snapshot my system, then update as I please. If I don't like the result, I can roll it back.
Redhat wants this because their dependency management has been too broken to successfully use the package manager to upgrade between major versions, unlike Debian where that is standard procedure.
Something like it could be useful for embedded devices, but systemD is a little heavy-weight for that.
Re: (Score:2)
Redhat wants this because their dependency management has been too broken to successfully use the package manager to upgrade between major versions, unlike Debian where that is standard procedure.
Exactly. Redhate is de-Unixifying Linux in the process of making up for their deficiencies. Instead of admitting RPM is garbage and moving on, they're making everyone else suffer for their incompetence.
Launch codes (Score:5, Interesting)
I don't agree with pottering's musings and have zero interest in seeing Linux turned into Android.
A/B is garbage both wasteful and unnecessarily rigid. A logged, versioned filesystem is a superior solution for booting any number of arbitrary system versions and variants where only deltas need to be stored.
If you want to protect system partition from users the easy solution is to use something like Opal and write lock the range which comprises the system partition prior to switching to user mode.
As for evil maids it is a fantasy to believe computers or their data can withstand physical tampering. FDE is mostly worthless against real world threats. Old school relatively trivial solutions like Class 0 offer a better solution against physical access allowing not a single byte to be read or written from the storage device without a key. No Verity + TPM + secure boot nonsense required.
Re: (Score:2)
As for evil maids it is a fantasy to believe computers or their data can withstand physical tampering. FDE is mostly worthless against real world threats.
FDE is basically for when you get your laptop stolen while it was off or somebody stealing your server for the hardware and hence turning it off. A second use for FDE is that it secures disks being thrown away without wiping them. But that is basically it. For a running machine, it is pretty meaningless.
Re: (Score:3)
It may not fit your desktop use case, but there's millions of devices shipped every year that this would simplify implementation and probably fix numerous security holes..
No, I wouldn't use it on my personal desktop either, but I'd use it at work in a hot minute if it works correctly.
Re: (Score:2)
I don't agree with pottering's musings and have zero interest in seeing Linux turned into Android.
Then don't. The great part about Linux is that it caters for all use cases. This has nothing to do with you? Fine, move on. You won't be affected by it. There are other people and other use cases who are calling for such a thing. Not the least of which is in cloud provisioning.
Re: (Score:2)
This has nothing to do with you? Fine, move on. You won't be affected by it.
Yes, the same way I wouldn't be affected by binary journald, by broken and feature-missing resolved, by timesyncd and all the other "optional" systemd component you promised us wouldn't affect us, and still did because distros relented and used them.
The extreme vast majority of us have no choice but to rely on distros. The only choice we have is to choose which distro to use, to choose whose choices we'll submit to.
Just stop repeating that lie that we have full choice, we don't, we have very limited choice.
Re: (Score:3)
Debian is
Ah, yes, the cancer grows (Score:2)
Fortunately, I cut it out when there still was time, but looks like a lot of the lesser competent Linux users are now getting really screwed over...
Systemd (Score:4)
Okay, I get the original concept. The boot process was archaic.
But then, they included a DNS resolver, X11 auto configuration (which broke many desktop assumptions), user session management, syslog replacement, and the kitchen sink.
https://en.wikipedia.org/wiki/... [wikipedia.org]
I was okay with all of those.
But including a package manager? Oh, come on, that is a bridge too far.
Re: (Score:3)
You realize systemd isn't a gigantic monolithic kitchen-sink binary, right? The service manager runs in PID 1, but things like logind, journald, networkd, resolved are all separate programs running in separate processes. And the auxiliary stuff is optional; if you don't want to use systemd's DNS resolver, don't install it. Plain old dhcli
Re: (Score:2)
Re: (Score:2)
You realize systemd isn't a gigantic monolithic kitchen-sink binary, right?
In practice, it is. Using any part of it means using other big parts of it unless you have stubbed it away, which represents a lot of duplicated and unnecessary effort.
Re: (Score:2)
But including a package manager? Oh, come on, that is a bridge too far.
Don't worry, it's only a package manager if you don't RTFA.
I just installed Devuan... (Score:2)
... on my flock of RPI-4's. I am impressed.
No systemd, no dropped log messages, no binary gibberish, no failed reboots. And I haven't had my eyeballs accosted and shrink-wrapped by a single UUID all week.
In fact, it feels like Unix back in the days where men were men, and women were women, and small furry creatures from Alpha Centuri worked for MIT and Cygnus Solutions while Microsoft was busy cobbling together winsock.dll having missed the dot-boat.
But have at 'er, Pottering. Nobody ever got rich or
Systemd is a bonerkill (Score:2)
Maybe I'm too old to like systemd. I don't know. What I do know is that systemd is a bonerkill. But the alternative, BSD, is flaccid. I'm stuck between a pillow and a soft place.
Re: (Score:2)
Re: (Score:2)
The overall concept of doing more in userland is fine, and many of the ideas behind systemd are ok. Systemd itself is pretty buggy and opinionated which makes it obnoxious to deal with. Ann's it swept over the Linux world due to some rather poor politicking while it was incredibly buggy and not really ready leaving something of a sour taste.
Re: (Score:2)
Agreed.
elephants (Score:5, Informative)
Like it or not, though, the nature of operating systems is changing. Modern ones are large, complex, and need regular updates
How is everyone missing the elephant in the room?
This is a huge problem. It means that a) I can't hold my server stable for years, something that a LOT of big enterprises rely on - why do you think banks and insurances still have mainframes in the basement? and b) the OS is full of holes and security issues that need constant patching. If you don't see that as a huge problem, you need a look at the economic losses caused by security incidents.
We've been going the wrong way. Sadly. I used to build systems that then run for 10+ years. Several of my servers have overrun the uptime counter. Now my OS kernel wants an update every other week? No, fuck you. I want to be sure that you are behaving the exact same way today as you did last month or last year. In some of the contexts I work in (industrial cybersecurity, IEC 62443 etc.) there are safety certifications that become null and void if I update the system.
Let's go back to trying to build a system that is correct the first time. Maybe it can do without a few of the bells and whistles.
Re: (Score:2)
Except your musings aren't even remotely correct. The reality is you can hold your server stable for years, and there are even dedicated distributions precisely for your use case.
This change here has nothing to do with your use case and won't impact you. Take a breath, in through the nose, absorbing the goodness in the world, and exhale deeply releasing all the kneejerk thoughts that you had.
Sadly. I used to build systems that then run for 10+ years.
Yes that is sad. I can only imagine the security problems you created given the kernel level critical CVEs that have
Re: (Score:2)
He's right.
EEE (Score:5, Insightful)
Embrace, Extend, Extinguish.
Lennart Poetering seems to be learning from the Microsoft methodology.
One more step, one more proof... (Score:2)
This is next one more piece of evidence that redhat and their useful idiot poettering are hell bent on eliminating all diversity from the linux ecosystem to make a macOS style "one true systemd/linux" distribution, that they control.
Redhat's objective is clear: if there is only one distro that they control (and they will control it, thanks to their control over systemd, freedesktop, wayland, flatpak, gnome, etc.), they control it, it's easier to support, it's easier to sell to companies (companies love unif
I'm not interested in immutable Linux. (Score:2)
It is an unnecessary change that will confuse Linux raison d'etre for future generations of users. Linux is about freedom from control. Freedom from manipulation of what we do with our computers. If someone wants an immutable OS they should do it themselves individually, and/or immutable distributions should be very rare and never promoted as a Linux OS/solution.
Re: (Score:2)
That's a use-case that Lennart wrote about wanting to support: you set up a system the normal way with packages, then snapshot it as an immutable image you can deploy to other machines. Later, you update the packages on the master system, then snapshot a new image to send to the other machines. Basically making your own custom immutable image-based sub-distribution out of a configuration you've built with a conventional mutable pac
Re: (Score:2)
Filesystem snapshots and COW subvolumes are another interesting way to do A/B images, but ZFS on Linux is controversial for license reasons and btrfs is controversial for technical reasons, so it makes sense to start with implementing A/B on partitions because that's the safe option that works everywhere. Partitions also mean you can use dm-verity to protect against changes below the filesystem level (e.g. corruption or malicious tampering). But I'm guessing systemd-sysupdate will probably gain btrfs supp
An alternative that can be widely used. (Score:2)
With as far as this unelected official is going we need a 2nd party to fend off bad ideas. Since these people now have what seems like total control of this important aspect of Linux it would seem there's an imperative to find and implement an alternative to systemd. Lenert does seem to be a politician trying to dictate the future of most Linux distributions. This is a devastating prospect.
Solaris Did It Decades Ago (Score:2)
Mainframes and Sun Microsystems Solaris had A/B upgrade systems for decades. It's not a new concept. Different disks / partitions / slices would be toggled as bootable. PC's had GoBack which kept a disk cache of all system changes and you could roll back in time including the OS as well as all your data. Windows copied this with System Restore in a somewhat limited fashion. Apple has Time Machine. Linux can revert back to a prior kernel release by editing the boot loader. But only for the kernel. Apple'
Re: (Score:2)
Not before you learn how to correctly capitalize the name.
Re: SystemD+EMACS (Score:2)
System Demon + Escape Meta Alt Control Shift?
This is a great idea! Two terrible ideas in one package. Now the Emacs users can enjoy the extra crashes systemd provides as a service *on top of* the constant gymnastics for their fingers.
Re: (Score:2)
Escape Meta Alt Control Shift
Actually "Editor MACroS" for those that don't know ... From 4.1 Where does the name “Emacs” come from? [gnu.org]:
Emacs originally was an acronym for Editor MACroS. RMS says he “picked the name Emacs because E was not in use as an abbreviation on ITS at the time.” The first Emacs was a set of macros written in 1976 at MIT by RMS for the editor TECO (Text Editor and COrrector, originally Tape Editor and COrrector) under ITS (the Incompatible Timesharing System) on a PDP-10.
Also noting:
Many people have said that TECO code looks a lot like line noise;
Re: (Score:2)
Systemd Linux
Great cthulhu, Systemd is its own distro now??!?!?
Re: (Score:2)
Defund the Poeter-lice!
What a trainwreck systemd has been for those serious administration of large fleets of servers.
If you want to use it to make your laptop seem to boot faster (while all the parallel running stuff hasn't really started yet), fine
Re: (Score:3)
What train wreck is this? Are you experienced in managing large scale deployments, server farms, etc? I am and systems was a vast improvement over init. In fact some modern uses of Linux ("cloud stuff", containers, etc) are facilitated by systemd in ways that old school init hacks never could. If systemd truly was the train wreck you say it is why aren't large Linux users saying anything? Amazon, Microsoft, etc. There have definitely been issues and bugs. But overall systemd has worked out rather well. I
Re: (Score:2)
Experienced admins are and have been saying systemd is bad, can find that with search engine "systemd bad design" "systemd broken" etc. You must be willingly ignoring the experience ones here who say it is (32 years Unix admin myself with over 2 decades Linux and BSD too)
It's not the Unix way to have a swiss army knife doing unrelated features, and making yet another place configuration can be when upgrading older OS.
Train wreck of troubleshooting with tossup if issues will even be in logging, loops and g
Re: (Score:2)
Indeed. Poettering has no understanding of professional computing at all.
Re: (Score:2)
He could fix PulseAudio. That's still borked.
Re: (Score:2)
Yeah he's the last asshole who should be allowed anywhere near any attempt to fix PulseAudio. It doesn't matter he created it, he's the last single person in the world who should be allowed to do anything with it.
Re: (Score:2)
It has mostly worked acceptably well with some minor quirks since Fedora 17, though it has been replaced with Pipewire. Pipewire integrates with JACK better.
Re: (Score:2)
Open *question
Re: (Score:2)
Category "Trump and Putin " can be brought down by political or military action.
Whether in a time of peace or war, there's no legitimate (alright, "legal") mechanism for putting any controls on Poettering. It's the Open Source world, so there isn't even an economic mechanism.