Files

duynguyen c2c265c898 docs: never store argocd admin password hash in git

Drop .env/envsubst/bake-hash-into-values.yaml flow entirely. Chart's
configs.secret block is removed for good, so the rendered manifest
never declares admin.password/admin.passwordMtime. Password hash is
set once directly on the live argocd-secret via kubectl patch, and
survives every self-heal sync since ArgoCD never owns those fields.

2026-07-14 18:52:46 +07:00

7.7 KiB

Raw Blame History

K8s Cluster Bootstrap — Platform Services

Bootstraps platform services onto the k8s cluster using ArgoCD app-of-apps pattern. Run this after cluster-init finishes provisioning and configuring nodes.

Architecture

Git repo (cluster-bootstrap)
  └── ArgoCD watches bootstrap/apps/ → syncs all Applications

Bootstrap order (sync waves):
  Wave -1 → argocd          (self-managed, once bootstrapped)
  Wave 0 → metallb          (LoadBalancer IPs)
  Wave 1 → metallb-config   (IPAddressPool + L2Advertisement)
  Wave 2 → envoy-gateway    (HTTP gateway controller)
  Wave 2 → nfs-provisioner  (dynamic PVC provisioner from xpen NAS)
  Wave 3 → envoy-gateway-config (GatewayClass + EnvoyProxy + Gateway)

External access:
  MetalLB assigns 192.168.1.30 to Envoy Gateway LoadBalancer service
  All HTTP traffic → Envoy Gateway (192.168.1.30:80) → HTTPRoutes → services
  DNS: *.fireflylab.local → 192.168.1.30  (configure in your local DNS/router)

Prerequisites

k8s cluster running (see cluster-init repo)
kubectl configured on client machine (kubeconfig at ~/.kube/config)
Client machine can reach 192.168.1.31 (master01)
xpen NAS NFS export accessible from all k8s nodes

Phase 1 — Install Helm

curl https://raw.githubusercontent.com/helm/helm/main/scripts/get-helm-3 | bash

helm version

Phase 2 — Install tools

sudo dnf install -y httpd-tools

htpasswd is used in Phase 3.1 to generate the admin password hash. No .env file, no envsubst — the password never touches Git.

Phase 3 — Install ArgoCD

Add Helm repo:

helm repo add argo https://argoproj.github.io/argo-helm

helm repo update

Install ArgoCD:

helm install argocd argo/argo-cd -n argocd --create-namespace -f manifests/argocd/values.yaml

values.yaml deliberately has no configs.secret block — the chart auto-generates a random admin password, stored in argocd-initial-admin-secret. That's replaced with your own in 3.1.

Wait for ArgoCD to be ready:

kubectl wait --for=condition=available deployment/argocd-server -n argocd --timeout=120s

Check pods are up:

kubectl get pods -n argocd

UI login isn't needed for bootstrap (Phase 4 applies the root Application via kubectl). Only port-forward if you want to inspect ArgoCD manually: kubectl port-forward svc/argocd-server -n argocd 8080:443, then open https://localhost:8080 with admin / your chosen password.

3.1 Set your own admin password directly on the Secret

Set it once, straight on argocd-secret (the Secret ArgoCD actually reads) — never in values.yaml, never committed to Git:

read -s -p "ArgoCD admin password: " PW; echo
HASH=$(htpasswd -nbBC 12 "" "$PW" | tr -d ':\n' | sed 's/$2y/$2a/')
kubectl patch secret argocd-secret -n argocd --type merge -p \
  "{\"stringData\":{\"admin.password\":\"$HASH\",\"admin.passwordMtime\":\"$(date -u +%FT%TZ)\"}}"
unset PW HASH

Why this stays stable across self-management: values.yaml never declares configs.secret.argocdServerAdminPassword, so the Helm chart's rendered manifest never includes admin.password/admin.passwordMtime. Once ArgoCD self-manages (wave -1, selfHeal: true), it only reconciles fields present in its own rendered output — it has no opinion on keys it never declared, so your patched hash survives every sync, forever. If that block is ever added back to values.yaml, the next self-heal overwrites it — don't add it.

To rotate later: repeat the same kubectl patch with a new hash and a new admin.passwordMtime (ArgoCD only accepts the change if the Mtime also changes).

Phase 4 — Apply the root bootstrap Application

bootstrap-app.yaml (repo root) is the one seed manifest — it lives in Git, not created ad-hoc via UI/CLI. Per GitOps practice, every Application ArgoCD runs must be recreatable from Git; a UI-created app that's stored nowhere can't survive a rebuild. This is the only manifest ever applied by hand — everything downstream (including ArgoCD managing itself, wave -1) is reached by ArgoCD syncing bootstrap/apps/ from here.

kubectl apply -f bootstrap-app.yaml

ArgoCD syncs all child Applications in wave order automatically, including self-managing itself via bootstrap/apps/argocd.yaml (wave -1).

Monitor progress:

kubectl get applications -n argocd

kubectl get pods -n metallb-system

kubectl get pods -n envoy-gateway-system

kubectl get pods -n nfs-provisioner

Phase 5 — Verify Envoy Gateway has external IP

kubectl get svc -n envoy-gateway-system

EXTERNAL-IP should be 192.168.1.30 (assigned by MetalLB).

If it stays <pending>, check MetalLB:

kubectl get ipaddresspool -n metallb-system

kubectl get l2advertisement -n metallb-system

Phase 6 — Apply ArgoCD HTTPRoute

Once Envoy Gateway has the external IP, expose ArgoCD via hostname:

kubectl apply -f manifests/argocd/httproute.yaml

ArgoCD UI now accessible at http://argocd.fireflylab.local — no more port-forward needed.

Phase 7 — Verify StorageClasses

kubectl get storageclass

Expected:

NAME          PROVISIONER                                     RECLAIMPOLICY
nfs-delete    cluster.local/nfs-subdir-external-provisioner  Delete
nfs-retain    cluster.local/nfs-subdir-external-provisioner  Retain

Test PVC provisioning:

kubectl apply -f - <<'EOF'
apiVersion: v1
kind: PersistentVolumeClaim
metadata:
  name: nfs-test
  namespace: default
spec:
  accessModes: [ReadWriteOnce]
  storageClassName: nfs-delete
  resources:
    requests:
      storage: 1Gi
EOF

kubectl get pvc nfs-test

Status should be Bound. Clean up:

kubectl delete pvc nfs-test

Full verification

kubectl get applications -n argocd

kubectl get pods -n metallb-system

kubectl get pods -n envoy-gateway-system

kubectl get pods -n nfs-provisioner

kubectl get svc -n envoy-gateway-system

kubectl get storageclass

All Applications should be Synced / Healthy.

File reference

cluster-bootstrap/
├── README.md
├── bootstrap-app.yaml                 # root seed Application — the only manifest applied by hand
├── bootstrap/
│   └── apps/                         # ArgoCD Application CRDs
│       ├── argocd.yaml               # wave -1 — self-managed ArgoCD
│       ├── metallb.yaml              # wave 0 — Helm chart
│       ├── metallb-config.yaml       # wave 1 — IPAddressPool + L2Advertisement
│       ├── envoy-gateway.yaml        # wave 2 — Helm chart
│       ├── nfs-provisioner.yaml      # wave 2 — Helm chart
│       └── envoy-gateway-config.yaml # wave 3 — GatewayClass + EnvoyProxy + Gateway
└── manifests/
    ├── argocd/
    │   ├── values.yaml               # ArgoCD Helm values
    │   └── httproute.yaml            # ArgoCD HTTPRoute (applied after Envoy is up)
    ├── metallb/
    │   └── values.yaml
    ├── metallb-config/
    │   ├── ipaddresspool.yaml        # IP pool: 192.168.1.30/32
    │   └── l2advertisement.yaml
    ├── envoy-gateway/
    │   └── values.yaml
    ├── envoy-gateway-config/
    │   ├── gatewayclass.yaml
    │   ├── envoy-proxy.yaml          # DaemonSet, LoadBalancer service
    │   └── gateway.yaml             # HTTP :80 listener
    └── nfs-provisioner/
        └── values.yaml              # ⚠ fill in nfs.server + nfs.path before push

7.7 KiB Raw Blame History