# Proxmox Kubernetes Cluster

Private HA K3s cluster on Proxmox, provisioned by Terraform, bootstrapped by Ansible, and reconciled by Flux.

## Architecture

| Component | Current Baseline |
|-----------|------------------|
| **Control plane** | 3 Proxmox VMs, VMIDs `200-202`, IPs `10.27.27.30-32`, 2 vCPU / 4 GiB / 32 GiB |
| **Workers** | 5 Proxmox VMs, VMIDs `210-214`, IPs `10.27.27.41-45`, 4 vCPU / 8 GiB / 64 GiB |
| **Kubernetes** | K3s `v1.34.6+k3s1`, HA embedded etcd, kube-vip API VIP `10.27.27.40` |
| **Proxmox** | Node `flex`, template VMID `9000`, datastore `Flash`, bridge `vmbr0` |
| **Storage** | Raw-manifest `nfs-subdir-external-provisioner`, `10.27.27.239:/TheFlash/k8s-nfs`, default StorageClass `flash-nfs` |
| **GitOps** | Flux source `platform` on branch `main`; `apps` Kustomization is intentionally suspended |
| **Private access** | Tailscale operator exposes Rancher, Grafana, and Prometheus; no public ingress baseline |
| **Runtime secrets** | Doppler service token bootstraps External Secrets Operator |

K3s is pinned because Rancher chart `2.13.3` requires Kubernetes `<1.35.0-0`.

## Prerequisites

- Terraform `>= 1.0`.
- Ansible with Python `jinja2` and `pyyaml`.
- `kubectl` for local verification.
- Proxmox API token for the `bpg/proxmox` provider.
- S3-compatible bucket for Terraform state, currently Backblaze B2.
- SSH key pair available to Terraform and Ansible, defaulting to `~/.ssh/infra` and `~/.ssh/infra.pub`.

Expected Proxmox inputs:

| Setting | Value |
|---------|-------|
| Endpoint | `https://100.105.0.115:8006/` |
| Node | `flex` |
| Clone source | Template VMID `9000` (`ubuntu-2404-k8s-template`) |
| Storage | `Flash` |

## Local Setup

Create local variables from the example:

```bash
cp terraform.tfvars.example terraform.tfvars
```

Important defaults in `terraform.tfvars.example`:

```hcl
proxmox_endpoint         = "https://100.105.0.115:8006/"
proxmox_api_token_id     = "terraform-prov@pve!k8s-cluster"
proxmox_api_token_secret = "your-proxmox-api-token-secret"

ssh_public_key  = "~/.ssh/infra.pub"
ssh_private_key = "~/.ssh/infra"

s3_access_key = "your-backblaze-key-id"
s3_secret_key = "your-backblaze-application-key"
s3_endpoint   = "https://s3.eu-central-003.backblazeb2.com"
s3_bucket     = "k8s-terraform-state"

tailscale_tailnet = "yourtailnet.ts.net"
kube_api_vip     = "10.27.27.40"
```

Initialize Terraform with backend credentials:

```bash
terraform -chdir=terraform init \
  -backend-config="endpoint=<s3-endpoint>" \
  -backend-config="bucket=<s3-bucket>" \
  -backend-config="region=auto" \
  -backend-config="access_key=<s3-access-key>" \
  -backend-config="secret_key=<s3-secret-key>" \
  -backend-config="skip_requesting_account_id=true"
```

## Common Commands

Terraform:

```bash
terraform -chdir=terraform fmt -recursive
terraform -chdir=terraform validate
terraform -chdir=terraform plan -var-file=../terraform.tfvars
terraform -chdir=terraform apply -var-file=../terraform.tfvars
```

Ansible setup:

```bash
ansible-galaxy collection install -r ansible/requirements.yml
cd ansible
python3 generate_inventory.py
ansible-playbook site.yml --syntax-check
```

Manual Ansible bootstrap uses the same extra vars as the deploy workflow:

```bash
cd ansible
ansible-playbook site.yml \
  -e "tailscale_auth_key=$TAILSCALE_AUTH_KEY" \
  -e "tailscale_tailnet=$TAILSCALE_TAILNET" \
  -e "tailscale_oauth_client_id=$TAILSCALE_OAUTH_CLIENT_ID" \
  -e "tailscale_oauth_client_secret=$TAILSCALE_OAUTH_CLIENT_SECRET" \
  -e "doppler_hetznerterra_service_token=$DOPPLER_HETZNERTERRA_SERVICE_TOKEN" \
  -e "tailscale_api_key=${TAILSCALE_API_KEY:-}" \
  -e "grafana_admin_password=${GRAFANA_ADMIN_PASSWORD:-}" \
  -e "cluster_name=k8s-cluster"
```

Flux/Kustomize verification:

```bash
kubectl kustomize infrastructure/addons/<addon>
kubectl kustomize infrastructure/addons
kubectl kustomize clusters/prod/flux-system
```

Refresh kubeconfig after rebuilds:

```bash
scripts/refresh-kubeconfig.sh 10.27.27.30
export KUBECONFIG=$(pwd)/outputs/kubeconfig
kubectl get nodes
```

Run the tailnet smoke check from cp1:

```bash
ssh ubuntu@10.27.27.30 'bash -s' < scripts/smoke-check-tailnet-services.sh
```

## Gitea CI/CD

The supported full rebuild path is the Gitea deploy workflow.

| Workflow | Trigger | Purpose |
|----------|---------|---------|
| `.gitea/workflows/deploy.yml` | PR to `main`, push to `main`, manual dispatch | PRs run Terraform plan; pushes run Terraform apply, Ansible bootstrap, Flux bootstrap, addon gates, health checks, and tailnet smoke checks |
| `.gitea/workflows/destroy.yml` | Manual dispatch with `confirm: destroy` | Terraform destroy with retries; no Rancher backup gate |
| `.gitea/workflows/dashboards.yml` | Grafana content changes or manual dispatch | Fast Grafana datasource/dashboard update through `ansible/dashboards.yml` |

Deploy and destroy share `concurrency.group: prod-cluster` so they do not run at the same time.

Deploy sequence on push to `main`:

1. Terraform fmt/init/validate/plan/apply.
2. Cleanup/retry around known transient Proxmox clone and disk-update failures.
3. Generate Ansible inventory from Terraform outputs.
4. Prepare critical image archives with `skopeo` on the runner.
5. Run `ansible/site.yml` to bootstrap nodes, K3s, kube-vip, prerequisite secrets, and kubeconfig.
6. Apply Flux CRDs/controllers and the `clusters/prod/flux-system` graph.
7. Gate cert-manager, External Secrets, Tailscale, NFS, Rancher, and observability.
8. Run post-deploy health checks and Tailscale service smoke checks.

Required Gitea secrets:

| Secret | Description |
|--------|-------------|
| `PROXMOX_ENDPOINT` | Proxmox API endpoint, for example `https://100.105.0.115:8006/` |
| `PROXMOX_API_TOKEN_ID` | Proxmox API token ID |
| `PROXMOX_API_TOKEN_SECRET` | Proxmox API token secret |
| `S3_ACCESS_KEY` | S3/Backblaze access key for Terraform state |
| `S3_SECRET_KEY` | S3/Backblaze secret key for Terraform state |
| `S3_ENDPOINT` | S3 endpoint, for example `https://s3.eu-central-003.backblazeb2.com` |
| `S3_BUCKET` | Terraform state bucket, for example `k8s-terraform-state` |
| `TAILSCALE_AUTH_KEY` | Tailscale auth key for node bootstrap |
| `TAILSCALE_TAILNET` | Tailnet domain, for example `silverside-gopher.ts.net` |
| `TAILSCALE_OAUTH_CLIENT_ID` | Tailscale OAuth client ID for the Kubernetes operator |
| `TAILSCALE_OAUTH_CLIENT_SECRET` | Tailscale OAuth client secret for the Kubernetes operator |
| `TAILSCALE_API_KEY` | Optional API key used to delete stale offline reserved devices before service proxies exist |
| `DOPPLER_HETZNERTERRA_SERVICE_TOKEN` | Doppler service token for runtime cluster secrets |
| `GRAFANA_ADMIN_PASSWORD` | Optional Grafana admin password |
| `SSH_PUBLIC_KEY` | SSH public key content |
| `SSH_PRIVATE_KEY` | SSH private key content |

## GitOps Graph

Flux entrypoint:

```text
clusters/prod/flux-system/
├── gotk-components.yaml
├── gitrepository-platform.yaml
├── kustomization-infrastructure.yaml
└── kustomization-apps.yaml  # suspend: true
```

Active infrastructure addons from `infrastructure/addons/kustomization.yaml`:

- `addon-nfs-storage`
- `addon-external-secrets`
- `addon-cert-manager`
- `addon-tailscale-operator`
- `addon-tailscale-proxyclass`
- `traefik` HelmRelease manifests applied directly by the top-level infrastructure Kustomization
- `addon-observability`
- `addon-observability-content`
- `addon-rancher`
- `addon-rancher-config`

Chart/source strategy:

- Vendored charts are intentional: `cert-manager`, `traefik`, `kube-prometheus-stack`, `tailscale-operator`, and `rancher` live under `infrastructure/charts/`.
- External Secrets, Loki, and Promtail use Flux `OCIRepository` sources.
- NFS storage is raw Kubernetes manifests, not a Helm chart.
- Rancher backup/restore is not part of the current live graph.

Doppler bootstrap details:

- `ansible/roles/doppler-bootstrap` creates the `external-secrets` namespace and the Doppler token secret only.
- The deploy workflow creates `ClusterSecretStore/doppler-hetznerterra` after ESO CRDs and webhook endpoints exist.
- The checked-in `infrastructure/addons/external-secrets/clustersecretstore-doppler-hetznerterra.yaml` is not included by the addon kustomization.

## Access URLs

| Service | URL |
|---------|-----|
| Rancher | `https://rancher.silverside-gopher.ts.net/` |
| Grafana | `http://grafana.silverside-gopher.ts.net/` |
| Prometheus | `http://prometheus.silverside-gopher.ts.net:9090/` |

Fallback port-forward from a tailnet-connected machine:

```bash
export KUBECONFIG=$(pwd)/outputs/kubeconfig
kubectl -n observability port-forward svc/kube-prometheus-stack-grafana 3000:80
kubectl -n observability port-forward svc/kube-prometheus-stack-prometheus 9090:9090
```

Grafana user is `admin`; password comes from the `GRAFANA_ADMIN_PASSWORD` Doppler secret or the workflow-provided fallback.

## Operations

Scale workers by updating `terraform.tfvars` counts, IP lists, and VMID lists together. If node names or VMIDs change, also update the hard-coded retry cleanup target map in `.gitea/workflows/deploy.yml`.

Upgrade K3s by changing the role defaults in `ansible/roles/k3s-server/defaults/main.yml` and `ansible/roles/k3s-agent/defaults/main.yml`. Check Rancher chart compatibility before moving to a Kubernetes minor outside `<1.35.0-0`.

Destroy through the Gitea `Destroy` workflow with `confirm: destroy`, or locally with:

```bash
terraform -chdir=terraform destroy -var-file=../terraform.tfvars
```

## Troubleshooting

Check K3s from cp1:

```bash
ssh ubuntu@10.27.27.30 'sudo k3s kubectl get nodes -o wide'
ssh ubuntu@10.27.27.30 'sudo journalctl -u k3s -n 120 --no-pager'
```

Check Flux and Rancher:

```bash
kubectl -n flux-system get gitrepositories,kustomizations,helmreleases,ocirepositories
kubectl -n flux-system describe helmrelease rancher
kubectl -n cattle-system get pods,deploy -o wide
```

Check Tailscale services:

```bash
kubectl -n tailscale-system get pods
kubectl -n cattle-system get svc rancher-tailscale
kubectl -n observability get svc grafana-tailscale prometheus-tailscale
kubectl -n cattle-system describe svc rancher-tailscale | grep TailscaleProxyReady
kubectl -n observability describe svc grafana-tailscale | grep TailscaleProxyReady
kubectl -n observability describe svc prometheus-tailscale | grep TailscaleProxyReady
```

If local `kubectl` falls back to `localhost:8080`, refresh `outputs/kubeconfig` with `scripts/refresh-kubeconfig.sh 10.27.27.30`.

## Security Notes

- Never commit `terraform.tfvars`, kubeconfigs, private keys, `outputs/`, or real secret values.
- Terraform/bootstrap/CI secrets stay in Gitea Actions secrets.
- Runtime cluster secrets are sourced from Doppler through External Secrets.
- This repo does not manage Proxmox/LAN firewalls or public ingress.

## License

MIT