Alright, let's cut to the chase. Everyone loves a shiny managed service until you see the invoice for a fleet of bare-metal nodes. So you're left with the DIY route, which feels like a choice between two flavors of "you own the problem."
We've all heard the siren song of Rancher/RKE2. It promises a unified management plane, a nice UI, and air-gapped bliss. But then you peek behind the curtain. You're not just deploying Kubernetes; you're deploying Rancher's entire opinionated stack. Ever tried to untangle their bundled ingress or figure out which magical add-on is breaking your CNI? The "value" often feels like paying for a luxury car but being forced to use the dealer's mechanic, who only speaks in proprietary tools.
Then there's Kubespray. It's Ansible. It's ugly, it's verbose, and it feels like assembling IKEA furniture with a missing Allen key. But the parts are standard. You see every bolt. Calico, Cilium, MetalLB—you pick, you configure, you suffer the consequences directly. No middleman, no surprise "enterprise module" required for basic observability.
So the real question is: which pain is more tolerable? The pain of a vendor's "helpful" abstraction that locks you into their support contract for sanity? Or the pain of raw, unfiltered DIY where the only support is your team's collective will to live and a GitHub issues page?
I'm leaning towards the latter. At least when Kubespray breaks, you know it's your own fault and the fix won't require a sales call. But maybe I'm just scarred from one too many "contact your account manager to enable this feature" dialogues.
—DW
—DW