Skip to content

TODO — Set Up Rancher (Later)

Status: Deferred — cluster is running fine without this for now. Confirmed 2026-10-26: Rancher is a cluster UI, not a secrets store; until there are several nodes or operators, use kubectl, helm and k9s over WireGuard (decisions log, Configuration and secrets management). Network constraint (decided 2026-10-25): the Rancher UI must be reachable over WireGuard only, never on a public 80/443 as in the quickstart below, and the cluster’s Rancher agent must reach Rancher over the private network, since the K3s droplet accepts no inbound connections after the lockdown (decisions log, Droplet network lockdown).

Context: The router-africa-prod K3s cluster on the [redacted] Droplet ([redacted-ip]) was deployed without Rancher, since no Rancher server existed yet. This doc picks up that thread whenever you’re ready.


Rancher gives you a management UI on top of the K3s cluster already running: cluster-wide monitoring (CPU/memory across nodes and pods without installing anything extra), a visual view of deployments/pods/logs instead of kubectl commands, and — most relevantly for Router.Africa’s growth — a much easier path to horizontal scaling later (Rancher’s UI over Kubernetes’ native autoscaling, rather than hand-editing YAML). None of this is required for Bifrost to run; it’s an operational convenience layer.


Step 1 — Decide where Rancher itself runs

Section titled “Step 1 — Decide where Rancher itself runs”

Rancher needs its own compute — it shouldn’t run on the same Droplet as Bifrost/K3s/Postgres, since that box is already sized tightly for just that workload. Options:

  • A new, small dedicated Droplet for Rancher (2GB+ RAM recommended for Rancher itself)
  • An existing box elsewhere if you have spare capacity
  • A managed Kubernetes service (e.g. DOKS) if you’d rather not manage the underlying OS for Rancher’s own host

Whichever you choose, treat this as a separate piece of infrastructure from the Router.Africa gateway cluster — don’t consolidate them onto one node.


On the dedicated Rancher server, the fastest path is the official Docker quickstart (fine for getting started; move to a proper HA Helm install before relying on it for production):

Terminal window
docker run -d --restart=unless-stopped \
-p 80:80 -p 443:443 \
--privileged \
rancher/rancher:latest

Once it’s up, get the bootstrap password:

Terminal window
docker logs <container-id> 2>&1 | grep "Bootstrap Password:"

Log into the Rancher UI at https://<rancher-server-ip>, complete the initial setup (set a real admin password, confirm the server URL).

For a production-grade install instead of the quickstart, use Rancher’s Helm chart on its own small Kubernetes cluster (K3s works fine for this too) — check Rancher’s current docs for the Helm install steps at that point, since chart versions and values shift over time.


Step 3 — Import the existing Router.Africa cluster

Section titled “Step 3 — Import the existing Router.Africa cluster”

Once Rancher is running:

  1. In the Rancher UI: Cluster Management → Import Existing → Generic
  2. Name it (e.g. router-africa-prod)
  3. Rancher generates a kubectl apply command with a registration URL + one-time token — copy it exactly from the UI (don’t reuse an old one, tokens are single-use/time-limited)
  4. Run it on the [redacted] Droplet, where kubectl is already configured:
    Terminal window
    curl --insecure -sfL https://<your-rancher-server>/v3/import/<token>.yaml | kubectl apply -f -
  5. Back in the Rancher UI, wait for the cluster to show Active

From that point on, you can manage the cluster either via kubectl/helm directly on the Droplet (as we’ve been doing) or through the Rancher UI — both control the same cluster.


Step 4 — What to revisit once Rancher is in place

Section titled “Step 4 — What to revisit once Rancher is in place”
  • Cluster monitoring — enable it under Cluster → Monitoring in the Rancher UI for CPU/memory visibility without extra setup
  • Scaling — once additional nodes are added to the cluster (more Droplets or a managed node pool), Rancher’s UI is the easier place to configure autoscaling than hand-editing Helm values directly
  • Access control — if others on the team need cluster access, Rancher’s RBAC/user management is a much friendlier layer than sharing kubectl config files directly

The Router.Africa Bifrost deployment on [redacted] runs fully via kubectl/helm without Rancher — this is a “nice to have later” layer, not a dependency. Pick this up whenever there’s a natural moment (e.g. when a second node/Droplet gets added, or when someone else on the team needs cluster visibility without shell access).