Kubernetes Control Planes

Home Platform Materialisation Kubernetes Control Planes

Platform Engineering · Capability

Kubernetes Control Planes.

Running RKE2 at scale — multi-region, multi-cluster, air-gapped. We design, deploy, and harden control planes for teams serving AI workloads in production.

Scope

What we do

  • Install and harden Rancher RKE2 control planes (bare-metal, cloud, air-gapped).
  • Design multi-region federation with GitOps (Fleet, ArgoCD).
  • Integrate SeaweedFS-backed CSI persistent storage.
  • Wire Envoy Gateway for ingress, mTLS, and policy-driven routing.

Practical

Exercises we run

Small, repeatable drills we use on engagements and teach in workshops. Each has a lab setup, step-by-step outline, and measurable output.

Air-gapped RKE2 in 90 minutesBootstrap a three-node cluster with no external egress, a minimal CNI, and a sealed bundle of workload images.
Multi-region federation on budgetTwo-region RKE2 federation using Fleet or ArgoCD for GitOps deployment of workloads to both regions.
CNI swap to SeaweedFSReplace the default CSI driver with SeaweedFS-backed storage; benchmark against Longhorn.

References

Four Kubernetes distributions we choose between

RKE2 is our default, but not every team's. Here's how we decide between it, K3s, OpenShift, and vanilla kubeadm on real engagements.

Project Best for Trade-offs When we reach for it
Rancher RKE2 Production clusters with CIS-hardened defaults, FIPS-140 mode, and a sane upgrade story via system-upgrade-controller. Works identically on bare-metal, cloud VMs, and air-gapped bundles. Stateful operator assumptions — tolerant of Rancher UI but doesn't require it. Slightly slower cold start than K3s; the hardened defaults are deliberate overhead. Default choice for AI/ML platforms, multi-region federations, and any workload where audit-grade security posture matters out of the box.
K3s Edge, single-node, IoT, and resource-constrained hosts. Single binary, SQLite default datastore, rapid cold start (<1 min). Datastore tradeoffs: SQLite default is fine for 1–3 nodes; anything bigger wants embedded etcd or external SQL. Fewer hardened defaults than RKE2 — you add them. Dev clusters, edge inference pods, labs, workshop environments, and any place where the operational overhead of RKE2 isn't worth the weight.
Red Hat OpenShift Enterprises with an existing Red Hat contract and regulated workloads (pharma, finance, gov). Opinionated Source-to-Image pipelines, integrated IdM, and certified compliance paperwork. Heaviest footprint of the four — CPU, RAM, and licence cost. Opinions around routes, SCCs, and Operators create friction for teams used to upstream Kubernetes patterns. Engagements where the customer already runs OpenShift and we're a consumer, not a re-platformer. Rare pick for greenfield work.
Vanilla kubeadm Full upstream control — you pick the CNI, CSI, ingress, security defaults. Matches the reference docs exactly, which is useful for teaching. All assembly, no batteries. Every hardening decision is yours to make and maintain — CIS benchmarks, certificate rotation, upgrade orchestration, the lot. Customers with a seasoned platform team who explicitly want a bespoke stack, or certification/audit scenarios where "upstream kubeadm" is the stipulated baseline.

We've shipped production clusters on all four. The honest summary: if you have a Red Hat shop, use OpenShift; if you're a single-digit-engineer platform team, use RKE2; if you're at the edge, use K3s; if you have a specific reason to be different, use kubeadm.

Further reading

More on Kubernetes.

Workshops we teach + field notes we're writing, all linked back to what you just read. See all workshops → See all field notes →

Workshop

Hands-on: Air-gapped RKE2 — 1-day workshop

Bootstrap a 3-node RKE2 cluster on a sealed network from a single pre-staged tarball. Bundle pipeline + bootstrap script + CIS-benchmark evidence path.

Scheduling soon →

Field note

Air-gapped RKE2 bootstrap in 90 minutes

Pre-staged bundle + bootstrap script that survives a procurement-grade transfer to sealed hosts. From cold iron to `kubectl get nodes` under 90 minutes.

Draft →

Engagement

Hands-on: Air-gapped RKE2 — 1-day workshop

Packaged engagement — we scope, build, and hand over with runbooks, against a specific SLA. Add to cart to request delivery; no price is billed up-front.

Neux Ltd

AI Infrastructure · Platform Engineering · London.
Since 2014.

Contact

LinkedIn

Legal

© 2014–2026 Neux Ltd
Registered in England & Wales.