# =============================================================================
# Prometheus main config (prometheus.yml) — TASK-0135, phase-15 §Observability
#
# Source of truth. The monitoring-namespace ConfigMap
# (infrastructure/kubernetes/observability/kustomization.yaml,
# configMapGenerator key `prometheus.yml`) is generated FROM this file — do
# not hand-edit the rendered ConfigMap.
#
# Targets — every app serves /metrics (the worker's is its ONLY port, 4003):
#   factory-api        api.factory.svc.cluster.local:4000/metrics
#   factory-worker     worker pods:4003/metrics  (pod SD — the worker has no
#                       ClusterIP Service; its health server binds the pod IP)
#   factory-realtime   realtime.factory.svc.cluster.local:4001/metrics
#                       (the app's local ws_* registry + OTel text, merged)
#   factory-web        web.factory.svc.cluster.local:3000/api/metrics
#                       (Next.js route — the path differs from the others)
#   node-exporter      expected-present daemonset (pod SD by container name;
#                       the DiskUsageHigh alert reads it — no data until a
#                       node exporter is installed in the cluster)
#
# Label strategy: each job sets a `service_name` static label, so EVERY
# series from that target carries service_name="factory-<app>" (including the
# app-local realtime ws_* series, which carry no label of their own). The
# `target_info{service_name=...}` gauge (always emitted by the OTel exporter)
# is the belt-and-braces proof of identity.
#
# Naming contract (classic text format, exporter-prometheus 0.57.0):
# counters expose with a `_total` suffix that is NOT in the instrument name —
# http_request_count → http_request_count_total; ws_messages_total /
# storage_ops_total / ai_tokens_total / error_total already end in _total and
# are unchanged. Recording rules + alerts use the EXPOSED names.
# =============================================================================
global:
  scrape_interval: 15s
  evaluation_interval: 15s

rule_files:
  # sibling files, mounted into /etc/prometheus/rules/ by the Deployment
  - /etc/prometheus/rules/recording-rules.yaml
  - /etc/prometheus/rules/alerts.yaml

alerting:
  alertmanagers:
    - static_configs:
        - targets: ["alertmanager.monitoring.svc.cluster.local:9093"]

scrape_configs:
  # --- the four factory apps --------------------------------------------------
  - job_name: factory-api
    metrics_path: /metrics
    static_configs:
      - targets: ["api.factory.svc.cluster.local:4000"]
        labels:
          service_name: factory-api

  # The worker exposes /metrics only on its pod IP (no ClusterIP Service in
  # worker.yaml) → pod SD, keep the worker container, pin port 4003.
  - job_name: factory-worker
    metrics_path: /metrics
    kubernetes_sd_configs:
      - role: pod
    relabel_configs:
      - source_labels: [__meta_kubernetes_namespace]
        regex: "factory"
        action: keep
      - source_labels: [__meta_kubernetes_pod_container_name]
        regex: "worker"
        action: keep
      - source_labels: [__meta_kubernetes_pod_ip]
        target_label: __address__
        replacement: "$1:4003"
      - target_label: service_name
        replacement: factory-worker

  - job_name: factory-realtime
    metrics_path: /metrics
    static_configs:
      - targets: ["realtime.factory.svc.cluster.local:4001"]
        labels:
          service_name: factory-realtime

  # web serves the exposition at /api/metrics (Next.js app route).
  - job_name: factory-web
    metrics_path: /api/metrics
    static_configs:
      - targets: ["web.factory.svc.cluster.local:3000"]
        labels:
          service_name: factory-web

  # --- node exporter (expected-present; see header) ---------------------------
  - job_name: node-exporter
    kubernetes_sd_configs:
      - role: pod
    relabel_configs:
      - source_labels: [__meta_kubernetes_pod_container_name]
        regex: "node-exporter"
        action: keep
      - source_labels: [__meta_kubernetes_pod_node_name]
        target_label: node
      - target_label: service_name
        replacement: node-exporter

  # --- self -------------------------------------------------------------------
  - job_name: prometheus
    static_configs:
      - targets: ["localhost:9090"]