apiVersion: apps/v1
kind: StatefulSet
metadata:
  name: vault
  namespace: vault
spec:
  serviceName: vault-internal
  replicas: 3
  selector:
    matchLabels:
      app.kubernetes.io/name: vault
  template:
    metadata:
      labels:
        app.kubernetes.io/name: vault
    spec:
      serviceAccountName: vault # Allows k8s auth for autounsealing
      terminationGracePeriodSeconds: 10
      affinity:
        # 1. Pin the pods strictly to K3s server nodes
        # nodeAffinity:
        #   requiredDuringSchedulingIgnoredDuringExecution:
        #     nodeSelectorTerms:
        #     - matchExpressions:
        #       - key: node-role.kubernetes.io/control-plane
        #         operator: Exists

        # 2. Force those pods onto different physical server nodes
        podAntiAffinity:
          requiredDuringSchedulingIgnoredDuringExecution:
            - labelSelector:
                matchExpressions:
                  - key: app.kubernetes.io/name
                    operator: In
                    values: ["vault"]
              topologyKey: "kubernetes.io/hostname"
      securityContext:
        runAsUser: 100
        runAsGroup: 1000
        fsGroup: 1000
      initContainers:
      - name: fix-raft-permissions
        image: busybox:latest
        command: ["sh", "-c", "chown -R 100:1000 /vault/data && chmod -R 700 /vault/data"]
        volumeMounts:
        - name: data
          mountPath: /vault/data
        securityContext:
          runAsUser: 0
          privileged: true

      - name: fetch-transit-token
        image: hashicorp/vault:latest
        command: ["sh", "-c"]
        args:
          - |
            set -e
            K8S_TOKEN=$(cat /var/run/secrets/kubernetes.io/serviceaccount/token)
            TRANSIT_TOKEN=$(vault write -field=token auth/kubernetes/login \
              role=vault-unsealer \
              jwt=$K8S_TOKEN)
            
            sed "s/REPLACE_WITH_TOKEN/$TRANSIT_TOKEN/" /etc/vault/extraconfig-from-values.hcl > /vault/runtime/vault.hcl
        securityContext:
          allowPrivilegeEscalation: false
          capabilities:
            add: ["IPC_LOCK"]
        env:
          - name: VAULT_ADDR
            value: "http://vault-keykeeper.vault.svc.cluster.local:80"
        volumeMounts:
          - name: config
            mountPath: /etc/vault
          - name: config-runtime
            mountPath: /vault/runtime
      containers:
      - name: vault
        image: hashicorp/vault:latest
        args: ["server", "-config=/vault/runtime/vault.hcl"]
        securityContext:
          allowPrivilegeEscalation: false
          capabilities:
            add: ["IPC_LOCK"]
        env:
        - name: POD_NAME
          valueFrom:
            fieldRef:
              fieldPath: metadata.name
        - name: VAULT_API_ADDR
          value: "http://$(POD_NAME).vault-internal.vault.svc.cluster.local:8200"
        - name: VAULT_CLUSTER_ADDR
          value: "https://$(POD_NAME).vault-internal.vault.svc.cluster.local:8201"
        - name: SKIP_SETCAP
          value: "true"
        - name: VAULT_ADDR
          value: "http://127.0.0.1:8200"
        - name: VAULT_K8S_NAMESPACE
          valueFrom:
            fieldRef:
              fieldPath: metadata.namespace
        - name: VAULT_K8S_POD_NAME
          valueFrom:
            fieldRef:
              fieldPath: metadata.name
        ports:
        - containerPort: 8200
          name: http
        - containerPort: 8201
          name: https-internal
        volumeMounts:
        - name: config
          mountPath: /etc/vault
        - name: data
          mountPath: /vault/data
        - name: config-runtime
          mountPath: /vault/runtime
        livenessProbe:
          httpGet:
            path: /v1/sys/health?standbyok=true&sealedok=true
            port: 8200
            scheme: HTTP
          initialDelaySeconds: 30
          periodSeconds: 10
          failureThreshold: 3
          timeoutSeconds: 5
        readinessProbe:
          httpGet:
            path: /v1/sys/health?standbyok=true
            port: 8200
          initialDelaySeconds: 10
          periodSeconds: 5
      volumes:
        - name: config
          configMap:
            name: vault-config
        - name: config-runtime
          emptyDir: {}
  volumeClaimTemplates:
  - metadata:
      name: data
    spec:
      accessModes: [ "ReadWriteOnce" ]
      storageClassName: longhorn
      resources:
        requests:
          storage: 1Gi