# =============================================================================
# Backend NestJS
# =============================================================================
---
apiVersion: apps/v1
kind: Deployment
metadata:
  name: xpeditis-backend
  namespace: xpeditis-prod
  labels:
    app.kubernetes.io/name: xpeditis-backend
    app.kubernetes.io/component: api
    app.kubernetes.io/part-of: xpeditis
spec:
  replicas: 2
  revisionHistoryLimit: 5
  strategy:
    type: RollingUpdate
    rollingUpdate:
      # Aucun pod retire avant qu'un remplacant ne soit pret : zero coupure.
      maxUnavailable: 0
      maxSurge: 1
  selector:
    matchLabels:
      app.kubernetes.io/name: xpeditis-backend
  template:
    metadata:
      labels:
        app.kubernetes.io/name: xpeditis-backend
        app.kubernetes.io/component: api
        app.kubernetes.io/part-of: xpeditis
      annotations:
        # Force le redemarrage des pods quand la configuration change : sans
        # cela, modifier le ConfigMap ne produit aucun effet visible.
        # La valeur est recalculee par scripts/deploy.sh.
        xpeditis.com/config-checksum: "PLACEHOLDER"
    spec:
      imagePullSecrets:
        - name: regcred
      # 2 replicas sur 1 noeud : la contrainte est "preferred", sinon le second
      # pod resterait indefiniment en Pending. Elle deviendra effective le jour
      # ou un second noeud sera ajoute (phase 2 du plan de charge).
      topologySpreadConstraints:
        - maxSkew: 1
          topologyKey: kubernetes.io/hostname
          whenUnsatisfiable: ScheduleAnyway
          labelSelector:
            matchLabels:
              app.kubernetes.io/name: xpeditis-backend
      securityContext:
        runAsNonRoot: true
        runAsUser: 1001
        runAsGroup: 1001
        fsGroup: 1001
        seccompProfile:
          type: RuntimeDefault
      # 60 s : laisse le temps aux requetes en cours et aux connexions
      # WebSocket de se fermer proprement.
      terminationGracePeriodSeconds: 60
      initContainers:
        - name: seed-rates
          image: rg.fr-par.scw.cloud/weworkstudio/xpeditis-backend:latest
          command: [sh, -ec, 'cp -R /app/src/infrastructure/storage/csv-storage/rates/. /rates/']
          securityContext:
            allowPrivilegeEscalation: false
            readOnlyRootFilesystem: true
            capabilities:
              drop: ["ALL"]
          resources:
            requests: { cpu: 50m, memory: 64Mi }
            limits: { cpu: 200m, memory: 128Mi }
          volumeMounts:
            - name: rates
              mountPath: /rates
      containers:
        - name: backend
          # Le tag est remplace au deploiement (kubectl set image).
          image: rg.fr-par.scw.cloud/weworkstudio/xpeditis-backend:latest
          imagePullPolicy: IfNotPresent
          ports:
            - name: http
              containerPort: 4000
              protocol: TCP
          envFrom:
            - configMapRef:
                name: xpeditis-backend-config
            - secretRef:
                name: xpeditis-backend-secrets
          env:
            - name: POD_NAME
              valueFrom:
                fieldRef:
                  fieldPath: metadata.name

          # Le demarrage inclut l'attente de PostgreSQL puis la verification
          # des migrations : la sonde de demarrage laisse jusqu'a 150 s avant
          # de declarer le pod perdu, sans penaliser les redemarrages rapides.
          startupProbe:
            httpGet:
              path: /api/v1/health/live
              port: http
            initialDelaySeconds: 10
            periodSeconds: 5
            failureThreshold: 30
            timeoutSeconds: 3

          livenessProbe:
            httpGet:
              path: /api/v1/health/live
              port: http
            periodSeconds: 20
            timeoutSeconds: 5
            failureThreshold: 3

          # LIMITE CONNUE : /health/ready renvoie toujours "ready" sans tester
          # PostgreSQL ni Redis (health.controller.ts). Un pod incapable de
          # joindre la base sera donc declare pret. Correctif recommande apres
          # la mise en ligne : @nestjs/terminus avec des indicateurs reels.
          readinessProbe:
            httpGet:
              path: /api/v1/health/ready
              port: http
            initialDelaySeconds: 5
            periodSeconds: 10
            timeoutSeconds: 3
            failureThreshold: 3

          resources:
            requests:
              cpu: 300m
              memory: 512Mi
            limits:
              cpu: 1500m
              memory: 1536Mi

          securityContext:
            allowPrivilegeEscalation: false
            capabilities:
              drop: ["ALL"]
            readOnlyRootFilesystem: true

          volumeMounts:
            - name: rates
              mountPath: /app/src/infrastructure/storage/csv-storage/rates
            - name: logs
              mountPath: /app/logs
            - name: tmp
              mountPath: /tmp
          lifecycle:
            preStop:
              exec:
                # Laisse Traefik retirer le pod de son pool avant que le
                # processus ne commence a refuser des connexions.
                command: ["sh", "-c", "sleep 10"]

      volumes:
        - name: rates
          emptyDir:
            sizeLimit: 1Gi
        - name: logs
          emptyDir:
            sizeLimit: 512Mi
        - name: tmp
          emptyDir:
            sizeLimit: 256Mi

---
apiVersion: v1
kind: Service
metadata:
  name: xpeditis-backend
  namespace: xpeditis-prod
  labels:
    app.kubernetes.io/name: xpeditis-backend
  annotations:
    # Sessions collantes : obligatoires pour Socket.IO. La negociation
    # long-polling echoue si deux requetes d'un meme handshake atterrissent sur
    # des replicas differents.
    #
    # ATTENTION -- limite non resolue par ce reglage : le gateway
    # (notifications.gateway.ts) garde la carte userId -> sockets EN MEMOIRE et
    # n'utilise pas @socket.io/redis-adapter. Une notification emise par le
    # replica A n'atteint donc pas un utilisateur connecte au replica B.
    # Cf. docs/mise-en-prod/15-exploitation-incidents.md, "Points de vigilance".
    traefik.ingress.kubernetes.io/service.sticky.cookie: "true"
    traefik.ingress.kubernetes.io/service.sticky.cookie.name: "xpd_be"
    traefik.ingress.kubernetes.io/service.sticky.cookie.secure: "true"
    traefik.ingress.kubernetes.io/service.sticky.cookie.httponly: "true"
    traefik.ingress.kubernetes.io/service.sticky.cookie.samesite: "lax"
spec:
  type: ClusterIP
  selector:
    app.kubernetes.io/name: xpeditis-backend
  ports:
    - name: http
      port: 4000
      targetPort: http
      protocol: TCP

---
apiVersion: policy/v1
kind: PodDisruptionBudget
metadata:
  name: xpeditis-backend
  namespace: xpeditis-prod
spec:
  minAvailable: 1
  selector:
    matchLabels:
      app.kubernetes.io/name: xpeditis-backend

---
apiVersion: autoscaling/v2
kind: HorizontalPodAutoscaler
metadata:
  name: xpeditis-backend
  namespace: xpeditis-prod
spec:
  scaleTargetRef:
    apiVersion: apps/v1
    kind: Deployment
    name: xpeditis-backend
  minReplicas: 2
  # Plafond a 4 : au-dela, le CPX41 sature. Passer a 3 noeuds avant d'augmenter
  # (cf. 15-exploitation-incidents.md, section montee en charge).
  maxReplicas: 4
  metrics:
    - type: Resource
      resource:
        name: cpu
        target:
          type: Utilization
          averageUtilization: 70
    - type: Resource
      resource:
        name: memory
        target:
          type: Utilization
          averageUtilization: 80
  behavior:
    scaleUp:
      stabilizationWindowSeconds: 60
    scaleDown:
      # Descente lente : evite de retirer un pod juste avant un nouveau pic.
      stabilizationWindowSeconds: 600