Hacker News Reader

Comment by grey-area | Hacker News Reader

Not great news if you use container Linux:

https://www.redhat.com/en/blog/faq-red-hat-acquire-coreos

Container Linux and its investment in container-optimized Linux and automated “over the air” software updates are complementary to Red Hat Enterprise Linux, Red Hat Enterprise Linux Atomic Host and Red Hat’s integrated container runtime and platform management capabilities. Red Hat Enterprise Linux’s content, the foundation of our application ecosystem will remain our only Linux offering. Whereas, some of the delivery mechanisms pioneered by Container Linux will be reviewed by a joint integration team and reconciled with Atomic.

View on HN

Yeah, not great. That sounds like an EOL announcement.

EDIT: Thanks for everything CoreOS team, and congrats on the sale! This immutable distribution was a breath of fresh air in a crowded "enterprise / kitchen sink" space.

Man have I really enjoyed CoreOS/Container Linux, etcd and generally all the work the CoreOS guys have done, and boy have I hated Red Hat's business practices. I moved away from Red Hat back when they split the OS into Enterprise and Fedora flavors.

I was a big fan of Red Hat back when they were leading the way in the early days of linux with things like application packaging, and a general vision towards simplicity. This is the same thing that attracted me to CoreOS' elegant engineering. I don't object to companies focusing more on revenue, as Red Hat's Enterprise move clearly was. But I cannot abide the erosion of vision in pursuit of profits. In particular when elegance and simplicity are replaced with complexity, and user hostile design to increase switching costs.

I realize there is probably a natural selection process at work here, and user hostile design is clearly a more profitable strategy, but I can't keep from hoping that one of these times the well engineered, simplicity focused, user centric design will find the right business model and survive... vision intact.

Interesting, I totally disagree.

The split of Fedora and RHEL was a great move for users IMHO. Serving the needs of desktop users and enterprise (servers) with the same distro just isn't doable. They have wildly different and incompatible needs. I've been a happy fedora user for years.

Red Hat has also done tons of things that directly fly in the face of a company seeking profits. Like continuing to release your full source code, even when punks like Oracle just rip it off and sell it for less. They even directly support the CentOS project, even tho it cannabalizes people who might otherwise buy RHEL.

Red Hat is an amazing company that has poured countless millions of dollars in both money and time to make all of our lives better. Even the Gnome project may not be a thing without Red Hat's contributions, and everybody on linux benefits from that work. I wish there were more companies out there like Red Hat.

Disclaimer: I have zero affiliation with Red Hat. I'm just an appreciative observer and beneficiary.

Companies that dont want to pay would simply use another free Linux, or maybe BSD. Centos keeps Red Hat relevant to those users and ensures they have an easy path to commercial Red Hat. It also makes sure accademics and students have an option to get familiar with it and provides a platform for developing skilled Red Hat aware talent. I'd be surprised if it costs them much more than 1% of sales, and as a free taster Im sure brings in much more.

Then there’s projects like OpenShift, we deployed OpenShift Origin on a cluster of CentOS Atomic VM, just over a year ago - at this point we are looking at buying dedicated hardware and OCP licenses from Red Hat now that we’ve used it in production for a while and need/want support when things go awry.

“Free taster” is certainly an apt anaology, I know there’s plenty of people who will never buy the supported products but there are likely a lot of other organizations like us that decide to buy in AFTER trying the upstream projects first.

Recent example: they open sourced Ansible Tower. They must've lost quite a few sales, but even after only a few months community contributions started flowing in.[1]

Red Hat really believes in the power of open source. It must be a great place to work at.

[1]: https://github.com/ansible/awx I wish more companies had that kind of courage.

Sounds like they are now going to support it as Open Source at least, even if RH is not interested in it:

https://groups.google.com/forum/#!topic/coreos-user/GR4YlF2c...

Super disappointed in this part of it. Is anyone else aware/familiar with a decent immutable lightweight Linux distro like Container Linux?

If you're looking for immutability, I can recommend NixOS. However what is drawing me to container Linux is mainly its update process

The draw of CoreOS Linux for me is its curated pairing of the Linux kernel, Docker, and etcd. There was a commercial entity reviewing upstream changelogs and making sure that they were pairing components appropriately. I wonder if an non-commercial district can really get this right.

Take a look at Linuxkit by Docker: https://github.com/linuxkit/linuxkit . I didn't develop it, but I use it indirectly via Docker for Mac, which I use a lot. The latest version supports Kubernetes. This means Docker commercially maintains a Linux/Docker/Kubernetes stack that needs to run reliably on what I think is a very large install base. They have open-sourced the system they use for this, it's called Linuxkit and it's a very cool, underrated project.

+1 too :-)

I'm using NixOS happily since 2014 after getting messed up in pacman dependencies on Arch and haven't looked back. But their too radical shift from LSB makes for a less attractive enterprise story. Hopefully GuixSD[1] can plug that hole due to the GNU brand as it has been progressing nicely.

[1]: https://www.gnu.org/software/guix/

GuixSD is just as disconnected from the Linux Standard Base as NixOS.

We've also built https://github.com/vpsfreecz/vpsadminos recently which is based on not-os / NixOS

RancherOS: http://rancher.com/rancher-os/

Rancher is awesome. As was CoreOS, and I am really pissed RH bought them out. RH manage to break everything they touch.

Linuxkit: https://github.com/linuxkit/linuxkit

Fedora / CentOS Atomic, which is Redhat's "clone" of container linux.

Atomic (or at least OStree) predates ContainerLinux by quite a bit.

Also Atomic is more than just an OS; it's a system of reliable upgrades / versions switching. So it's got something in common with NixOS.

Sort of, but not quite. I'm familar :)

I used to be a gnome foundation member (I trained Andrea Veri on the sysadmin team who trained Patrick Uiterwijk, the current lead sysadmin for GNOME) when Colin Watson wrote OSTree. He was talking about it and some of the ideas from it at the KDE/GNOME desktop summit in Berlin (2011ish) in some of the comments after https://desktopsummit.org/program/sessions/improving-how-we-...

At the following Boston (GNOME) developer summit one of the big conversational points sans Ubuntu pulling away from GNOME to build Unity was the idea of GNOME OS because Linux was hard then. He built rpm-ostree after that, and it was after CoreOS as a distro was a thing.

rpm-ostree's first commit was Dec 21, 2013: https://github.com/projectatomic/rpm-ostree/commit/958dfa435...

https://coreos.com/blog/coreos-fourth-birthday CoreOS as a company was created on July 13th, 2013.

TL;DNR: It's complicated, but Atomic as a distro most definitely does not predate ContainerLinux as rpm-ostree didn't exist when CoreOS the company was founded.

OStree's mailing list starts in April 2012. I'm sure Colin did a bunch of work before then.

Oh OSTree was definitely around then, but it wasn’t building a competitor to ContainerLinux until rpm-ostree, which made it insanely powerful in that it could build images from existing (and pervasive) redhat package repos. I’m definitely not arguing that much whatsoever. Only that it wasn’t building a full distro in a way to put it on par with ContainerLinix until after ContainerLinux was a thing and CoreOS was a company.

Thanks for the info and I agree. Also thank you for your work for Gnome. Awesome!

Forgive me for poking around a little more but I like reading computer history. That rpm-ostree first commit was an import from something called pkgsys-ostree. The first commit in pkgsys-ostree was only a little earlier; Aug 2013.

Heh, good on you for digging deeper. I think ultimately, the two were developed pretty much independently (Atomic and ContainerLinux), one from ChromeOS, and the other from Fedora / RHEL. They both prove the idea of a small container only os with atomic updates is a solid design to build a container orchestration system (k8s/openshift) ontop of. I'll still stand by the gist of my original comment though in that CoreOS was a thing before Atomic and they had a working P.O.C. before ostree was used to build an entire distro (such as atomic). Either way, we both learned something new so net win.

Common only in very high-level concept as the implementation details are vastly differing. NixOS is based on the Nix[1] purely functional package manager and configuration language while OSTree is termed to be "git for operating systems"[2].

[1]: https://nixos.org/nix/ [2]: https://ostree.readthedocs.io/

Atomic doesn't predate CoreOS.

CoreOS = 2013 Atomic host = 2014

OSTree (the foundation for it) was built as part of GNOME Continuous in 2011.

Yup, what I sort of alluded to in my comment above re: GNOME OS, however I was saying CoreOS the distro predates rpm-ostree. Colin was talking about some of these ideas in Berlin circa 2011 at the desktop summit (isn't that where we met btw? I forget).

Scale dude!

Gnome booth wet bar... too much good whiskey was consumed. I need to get back to that.

Absolutely.

But Atomic was architected as containers infrastructure and is/was a direct competitor to CoreOS and other lightweight container OSes.

It might appear that way to people that weren't involved in the history, but a lot of it was based on ideas learned from stateless linux, circa 2004.

People in the desktop team had been working on read-only root OS's for a long time, and it was only at a certain point that RH management bought off on forming a product/spin. But to say that work only started once Atomic was a thing would be inaccurate.

It's probably fair to say that ChromeOS/CoreOS lit a match under some upper management though.

There's Alpine Linux. https://www.alpinelinux.org/

Intel's Clear Linux is yet another option.

Alpine for sure.

Which delivery mechanisms? The ContainerLinux boot image updater comes from Chrome and ChromeOS (Omaha). They must be talking about the higher-level partitioning of a cluster into groups with separate policies, etc.

Congratulations to the CoreOS team, by the way.

tectonic, ignition, matchbox?

Also CoreUpdate is based on Chrome and ChromeOS but it's the only os that integrated it nicely. Running CoreOS is way simpler to manage than most other distro's because of that fact.

Atomic uses rpm-ostree, which looks similar to Omaha from a client perspective (update, roll back to previous version): http://www.projectatomic.io/docs/os-updates/

is there an alternative to locksmith? the "automatic" update (https://coreos.com/os/docs/latest/update-strategies.html) and/or https://github.com/coreos/container-linux-update-operator?

basically on coreos I don't need to run "rpm-ostree" to have an updated system. it is actually updated automatically.

And importantly, you can control updates across a whole cluster by limiting how many machines are allowed to update simultaneously.

I just can't get a decent "This is how to get started" from windows guide with atomic. It's all assuming I'm perfectly happy on linux. Not to knock it, but I just want to start some VMs on hyper-v and have a go.

This is good feedback. I'll bring it to the Fedora docs team.

Container Optimised OS (Which is just ChromiumOS stripped down) has been working pretty nicely for me on GCloud. Just reboot and you have the latest updates. https://cloud.google.com/container-optimized-os/docs/concept...

It's also open source

But yeh, Matchbox + Ignition for automatically bringing up configured nodes with iPXE was an extremely powerful combination. However, those ideas can be easily ported over or already exist through cloud-init.

Container Optimised OS also says they don't support use cases outside of GCloud, and it also relies on CoreOS' services itself.

So for bare metal use cases, I'm EOL?

well basically we use that on gcloud k8s in the future. but we still have a bare metal cluster. we only used ignition + vmware to provision it in the first place and it was a breeze.. etcd was configured on the fly in a HA setup, k8s was made ready to bootstrap (just needed to run kubeadm init --config...) and we had basically a working cluster. highly available, with as many nodes we wanted, just copy the worker.yaml and edit what's needed.

we actually play around with tectonic at the moment.

I was thinking of locksmith. It's nice, as you say, to be able to have no more than N nodes per pool restart at any time, so that rolling updates are automatic and in the background. You can't replicate it easily without a cluster-wide etcd and ignition. I forgot about matchbox. There was also good old fleet, but that one is dead.

>"The ContainerLinux boot image updater comes from Chrome and ChromeOS (Omaha)."

Could you or anyone else recommend any literature on ContainerLinux boot image and its use of ChromeOS as a foundation?

Reply on news.ycombinator.com