Tested Deployments

Note: The primary deployment has migrated from Civo k3s to the on-prem honey RKE2 cluster (3 nodes: honey, bumble, sting) with OpenEBS ZFS storage. The Civo cluster was decommissioned in April 2026. The Civo section below is retained as a historical reference deployment type.

GloriousFlywheel has been tested on two Kubernetes distributions: RKE2 for on-premise deployments and Civo k3s for cloud deployments. Both support the full stack (Attic cache, GitLab runners, runner dashboard).

RKE2 (On-Premise)

RKE2 is Rancher’s next-generation Kubernetes distribution, designed for security and compliance (FIPS 140-2 by default). It runs well on bare metal or VMs behind a corporate firewall.

Cluster requirements

• 3+ nodes (control plane + workers), or single-node for development
• Longhorn or another CSI driver for persistent volumes
• Ingress controller (nginx-ingress bundled with RKE2)
• Minimum 4 CPU, 8 GB RAM per worker for a full runner pool

Provisioning

Install RKE2 on each node per the RKE2 docs. For managed provisioning, use Rancher to create and lifecycle the cluster.

Storage

Longhorn is recommended. Install via Helm:

helm repo add longhorn https://charts.longhorn.io
helm install longhorn longhorn/longhorn --namespace longhorn-system --create-namespace

Network access

On-premise clusters typically require one of:

• Direct kubeconfig access (same network)
• VPN or Tailscale mesh
• SSH tunnel / SOCKS proxy through a jump host

See Cluster Access for details.

GitLab Agent

Register a GitLab Agent for CI/CD access:

1. Create an agent config in your overlay repo under k8s/agent/
2. Install the agent via Helm
3. Reference the agent path in your CI pipeline

Civo k3s (Cloud) — decommissioned April 2026

Civo provided managed k3s clusters with fast provisioning (under 2 minutes). This deployment type has been decommissioned; the section is retained for historical reference only.

Cluster requirements

• Civo account with API key
• Medium nodes (2 CPU, 4 GB) minimum; large (4 CPU, 8 GB) recommended for runners
• Civo CNI (Cilium) and Traefik ingress (default)

Provisioning

Via Civo CLI:

civo kubernetes create my-cluster \
  --size g4s.kube.medium \
  --nodes 3 \
  --region NYC1 \
  --save --merge --wait

Or via OpenTofu using the Civo provider:

resource "civo_kubernetes_cluster" "cluster" {
  name        = "my-cluster"
  region      = "NYC1"
  node_pool {
    size  = "g4s.kube.medium"
    count = 3
  }
}

Storage

Civo provides a default storage class backed by Civo Volumes. No additional setup needed.

Network access

Civo clusters have public API endpoints by default. For private clusters, use Tailscale:

1. Install Tailscale operator on the cluster
2. Connect from your local machine via Tailscale mesh
3. Access cluster services through Tailscale DNS

See Cluster Access for details.

GitLab Agent

Same agent registration process as RKE2. The Civo cluster’s public endpoint simplifies agent connectivity.

Comparison