# =============================================================================
# AI Software Factory — Worker (BullMQ) Deployment + HPA
#
# Consumes job queues from Redis (search indexing, file pipeline, automation
# execution, email, webhooks). 2 replicas by default.
#
# HPA NOTE — custom queue-depth metric:
#   The autoscaler below scales on the custom metric `queue_depth` (the
#   EXPOSED name of the OTel gauge the worker itself emits — TASK-0135:
#   packages/observability `queue_depth{queue}`, scraped by Prometheus at
#   /metrics:4003 per queue every 15 s).
#   For the metric to reach the autoscaler you need:
#     1. The Prometheus scrape target (infrastructure/observability/prometheus/
#        scrape.yaml, job `factory-worker`) — shipped with this task.
#     2. The Prometheus Adapter (k8s) with an `APIService` mapping the custom
#        metric, e.g. configmap adapter rules:
#        ---
#        rules:
#          - seriesQuery: queue_depth{queue=~".+"}
#            resources: {service: worker}
#            name: {as: queueDepth}
#            metricsQuery: |
#              sum by (queue) (queue_depth{namespace="{{.Namespace}}",service_name="factory-worker"})
#        (adjust the metricsQuery to your exporter's labels)
#   Without the adapter, replace the metric with a CPU target (commented at
#   the bottom) so the HPA still functions.
# =============================================================================
apiVersion: apps/v1
kind: Deployment
metadata:
  name: worker
  labels:
    app.kubernetes.io/name: worker
    app.kubernetes.io/component: jobs
spec:
  replicas: 2
  selector:
    matchLabels:
      app.kubernetes.io/name: worker
  strategy:
    type: RollingUpdate
    rollingUpdate:
      maxUnavailable: 0
      maxSurge: 1
  template:
    metadata:
      labels:
        app.kubernetes.io/name: worker
      annotations:
        checksum/config: "factory-config"
    spec:
      containers:
        - name: worker
          image: ghcr.io/ai-software-factory/worker:TAG
          imagePullPolicy: IfNotPresent
          # Workers have no HTTP port; liveness is a process heartbeat file
          # maintained by the worker (written each loop tick). Adjust the
          # path if your worker uses a different heartbeat mechanism.
          envFrom:
            - configMapRef:
                name: factory-config
            - secretRef:
                name: factory-secrets
          resources:
            requests:
              cpu: 250m
              memory: 512Mi
            limits:
              cpu: "1"
              memory: 1Gi
          livenessProbe:
            exec:
              command:
                - sh
                - -c
                - "test -f /tmp/worker.heartbeat && [ $(( $(date +%s) - $(cat /tmp/worker.heartbeat) )) -lt 90 ]"
            initialDelaySeconds: 30
            periodSeconds: 30
            timeoutSeconds: 5
            failureThreshold: 3
          securityContext:
            allowPrivilegeEscalation: false
            runAsNonRoot: true
            runAsUser: 1000
            runAsGroup: 1000
            seccompProfile:
              type: RuntimeDefault
      terminationGracePeriodSeconds: 60 # let in-flight jobs finish
---
apiVersion: autoscaling/v2
kind: HorizontalPodAutoscaler
metadata:
  name: worker
spec:
  scaleTargetRef:
    apiVersion: apps/v1
    kind: Deployment
    name: worker
  minReplicas: 2
  maxReplicas: 8
  metrics:
    # Preferred: scale on backlog. Requires the Prometheus Adapter setup
    # documented in the header of this file.
    - type: Pods
      pods:
        metric:
          name: queue_depth
        target:
          type: AverageValue
          averageValue: "50" # 50 pending jobs per worker pod → scale up
  # Fallback (uncomment if the custom metric is not wired yet):
  # - type: Resource
  #   resource:
  #     name: cpu
  #     target:
  #       type: Utilization
  #       averageUtilization: 70