I agree with all of this except for your bottom edit.

Nutanix and others are helping a lot in this area. Also really like Talos and hope they keep growing.

stego-tech 6 months ago | parent | next [-]

That’s fair! Nutanix impressed me as well when I was doing a hypervisor deep dive in 2022/2023, but I had concerns about their (lack of) profitability in the long run. VMware Tanzu wasn’t bad either, but was more of an arm-pull than AWS was for K8s. Talos is on my “to review” list, especially with their community license that let’s you manage small deployments like a proper Enterprise might (great evangelism idea, there), but moving everything to kube-virt was a nonstarter in the org at the time.

K8s’ ecosystem is improving by the day, but I’m still leaning towards a managed K8s cluster from a cloud provider for most production workloads, as it really is just a few lines of YAML to bootstrap new clusters with automated backups and secrets management nowadays - if you don’t mind the eye-watering bill that comes every month for said convenience.

▲

esseph 6 months ago | parent [-]

If you work for any of the CISA-Identified 16 critical infrastructure sectors, one of their recommendations is for organizations to be able to expect to operate for more than 24h without an Internet connection.

Kinda hard to control real-world things with no Internet connection that rely on an internet connection

Note: Nutanix made some interesting k8s-related acquisitions in the last few years. If interested, you should take a look at some of the things they are working on.

	▲	stego-tech 6 months ago \| parent [-]
		If I were still in that role, I’d absolutely be keeping my Nutanix rep warm for a possible migration. Alas, I’m in another org building them a Win11 imaging pipeline for the time being, and Nutanix doesn’t want to play nice with my personal N100 NUCs for me to try their Community Edition.

▲

nabeards 6 months ago | parent | prev [-]

Exactly the same as you said. Nobody rents GPUs as cheap as I can get them for LLM work in-cluster.