little point if you're running a home lab, might be cool for bigger firms

>>106789867
this software is cool cause it has something called kubelet which sounds like niglet haha lol

>>106789867
Google product, can you really trust it?

the idea sounds great, in pratice it adds unneeded complexity and makes it a fucking mess to troubleshoot things.
your worker node crashed and want to know why? fuck you

>>106789867
Skill issue: The Thread

>>106789867
If you're running your services in containers and you have multiple old computers to use, plus you already know how to use K8s and GitOps deployment tools like Argo CD, it is actually great.

While there are Ansible playbooks out there to set up a cluster, Talos Linux makes things much easier, and it is an immutable distribution. Install it on bare metal, use a TrueNAS server for storage (coupled with democratic-csi on the K8s side) and you can enjoy using Git to deploy anything you want.

Let the scheduler take care of assigning Pods to the right nodes, then just buy some refurbished ThinkCentre Tiny's.

>>106789867
The point is that I get paid for writing yaml files

I have no clue what im doing. At work i mostly just update a tag in a helm chart (whatever that is) for Argo and then I just find the right box to click in Argo to see the logs. I dont get 99% of the stuff I see in Argo. I typically just hyper focus on solving whatever problem there is at the time.
I wish I understood this stuff more, but im struggling to find resources that actually make sense to me. Doesn't help that at work we use a thousand different processes for these kinds of things. Nothing is in a simple form.

>>106789867
The point is getting paid.

>>106789867
The point is that it forces you to grapple with the complexity of deployment upfront. Availability zones, distribution, load balancing, failover, virtual IPs, software defined networking, port whitelisting, networked storage, high availability, blue/green strategies, delegation of permissions, administrator/developer least privilege, metrics, autoscaling, metric aggregation, log aggregation, alerting, its all in there, and you have to think about it. It makes it so that you can describe a desired state of your entire world in plain text (tragically that text is either JSON or YAML, but maybe that can be fixed in the future).

The big problem is, as >>106789954 points out, if you don't need any or even some of those things, then its a huge drag.
However, if you don't need any of those things, what you're doing doesn't matter. Either downtime means lost revenue, or, you know, people just straight up fucking dying, or it doesn't. If it doesn't, you can probably get away with a raspberry pi, a nuc, or a mac mini in a closet, or something like a digital ocean droplet if you want to be "in the cloud", and you'll do just fine.

If what you're doing actually matters, you either use kubernetes, or a half-baked, home-grown, worse version of kubernetes, or you fail. Simple as.

>>106792149
If you want to get better, get yourself some mini PCs or raspberry pis or whatever, and run through https://github.com/kelseyhightower/kubernetes-the-hard-way . If you want extra credit, use a more powerful machine and QEMU or libvirt to make virtual machines to run it in instead.
Then, install, by hand, no argo, each of the tools you use at work. Once you get that to work, start over and use argo to do it.
There's no 101 textbook for this stuff yet, its too new for that, the only way you can learn it is to do it.

>>106792938
pure fucking brainrot
if I want to "get better" at home I will study actual Computer Science

I'm not touching all that useless shit unless you pay me to

>>106792815
>or a half-baked, home-grown, worse version of kubernetes,
Literally my life right now. Long term the product is supposed to move off Windows Server and some bespoke clustering bullshit, but realistically, it isn't happening given it's been given to pajeets to implement.

Pray for me.

What CSI do you guys use? I've been using SMB3 mostly because it's the easiest to provision and actually secure without krb5 autism.

>>106789867
kubernetes suck at work to orchestrate and use as SWE (and I imagine as a SRE) in very large enterprise environments because it's infested with subhuman room temperature IQ brownshit skin pajeets.

>>106793081
Ideally you set the cluster up once and then interact with them as little as possible. So far, k8s is the least shit way to do anything at my company, second only to windows server and that's only because of the very generous configuration advantage of being domain joined properly and not have a totally buttfucked pam stack + dogshit Linux spyware.

>>106792973
>I'm not touching all that useless shit unless you pay me to
Lucky for you, there are tons of companies that will pay you to do it.

8 years of devops experience, many projects behind me
I am yet to be in a project that actually needed k8s, instead of an auto managed ecs fargate

>>106793477
>actually needed
Isnt that the biggest issue with this? People only use it because they hear about other people using it.
Nobody actually find reasonable solutions, just hop on another bandwagon.

>>106793531
I'll be real with you, no one needs k8s
I understand why for some reason, google, running 10 billion clusters and a billion network settings needs it, but yeah, no fucking commercial company needs anything but fargate lol

>>106789914
kek

>>106793541
Tell that to my work. What makes it even worse is that our office doesnt actually control the aws side of things and have to coordinate with another office on the other side of the country for this crap.
All the while, we have to ensure it interfaces with aws stuff correctly. Heck, we have to manually update and commit the tags for Argo in a separate git repo everytime we make a commit to our own projects and we only get the tag after the pipeline built the docker image and uploaded it which takes anywhere from 10-20 minutes.
Literally every 6 months they change up our processes to be even dumber than the last change.