VictoriaMetrics — Stack de Observabilidad NutriFlow

📦

docker-compose.yml — Single-node + vmagent + vmalert

Listo para producción NutriFlow

# docker-compose.ymlservices:
  victoriametrics:
    image: victoriametrics/victoria-metrics:v1.96.0
    command:
      - '-storageDataPath=/victoria-metrics-data'
      - '-retentionPeriod=90d'            # 90 días para NutriFlow
      - '-httpListenAddr=:8428'
      - '-search.maxUniqueTimeseries=2000000'  # 2M series para escala actual
      - '-dedup.minScrapeInterval=15s'
    ports:  ["8428:8428"]
    volumes:  [vm_data:/victoria-metrics-data]
    restart: always

  vmagent:
    image: victoriametrics/vmagent:v1.96.0
    command:
      - '-promscrape.config=/etc/vmagent/scrape.yml'
      - '-remoteWrite.url=http://victoriametrics:8428/api/v1/write'
      - '-remoteWrite.tmpDataPath=/vmagent-remotewrite-data'
      - '-promscrape.maxScrapeSize=64MB'
    volumes:
      - ./vmagent-scrape.yml:/etc/vmagent/scrape.yml:ro
      - vmagent_data:/vmagent-remotewrite-data

  vmalert:
    image: victoriametrics/vmalert:v1.96.0
    command:
      - '-datasource.url=http://victoriametrics:8428'
      - '-notifier.url=http://alertmanager:9093'
      - '-rule=/etc/vmalert/rules/*.yml'
      - '-evaluationInterval=30s'
    volumes:  [./alert-rules:/etc/vmalert/rules:ro]

  alertmanager:
    image: prom/alertmanager:v0.26.0
    volumes:  [./alertmanager.yml:/etc/alertmanager/config.yml:ro]
    ports:  ["9093:9093"]

volumes:  { vm_data: {}, vmagent_data: {} }

🔍

vmagent-scrape.yml

Kubernetes + servicios estáticos

global:
  scrape_interval: 15s
  external_labels:
    cluster: nutriflow-prod
    env: production

scrape_configs:
  - job_name: nutriflow-services
    static_configs:
      - targets:
          - api-gateway:8080
          - plan-service:8080
          - billing-service:8080
        labels:
          team: backend
          product: nutriflow
    metrics_path: /metrics
    relabel_configs:
      - source_labels: [__address__]
        regex: '([^:]+):.*'
        target_label: service

  - job_name: node-exporter
    static_configs:
      - targets: ['node-exporter:9100']
        labels:
          environment: production

  - job_name: kubernetes-pods
    kubernetes_sd_configs:
      - role: pod
    relabel_configs:
      - source_labels:
          [__meta_kubernetes_pod_annotation_prometheus_io_scrape]
        action: keep
        regex: true
      - source_labels:
          [__meta_kubernetes_namespace]
        target_label: namespace

🔔

alert-rules/service-alerts.yml

Umbrales NutriFlow

🚨

HighErrorRate

error_rate > 0.03 (umbral: 3% — más estricto que default)

for: 3m severity: critical → PagerDuty

⚡

P99LatencySpike — plan-service

quantile_over_time(0.99, duration[5m]) > 2.5

for: 5m severity: critical → Slack #ops

⚠️

HighMemoryUsage

mem_used / mem_total > 0.85

for: 10m severity: warning → Slack #infra

💳

BillingServiceDown

up{service="billing-service"} == 0

for: 1m severity: critical → PagerDuty inmediato

📐

Queries MetricsQL — Dashboards Grafana NutriFlow

MetricsQL extiende PromQL con funciones avanzadas

Request Rate por servicio (panel principal) PromQL compatible

sum(rate(http_requests_total{product="nutriflow"}[5m])) by (service)

Muestra RPS de api-gateway, plan-service y billing-service en tiempo real

Latencia P99 por servicio PromQL compatible

quantile_over_time(0.99, http_request_duration_seconds{product="nutriflow"}[5m]) by (service)

Percentil 99 de latencia — detecta cuellos de botella en plan-service (llamadas IA)

Top 3 endpoints con mayor error rate MetricsQL exclusivo

topk_avg(3, sum(rate(http_requests_total{status=~"5..",product="nutriflow"}[5m])) by (endpoint) / sum(rate(http_requests_total{product="nutriflow"}[5m])) by (endpoint) )

topk_avg (MetricsQL) evita spikes temporales que engañan a topk() estándar

Total acumulado de planes generados (últimas 24h) MetricsQL exclusivo

running_sum(increase(plans_generated_total{service="plan-service"}[1h]))

running_sum no existe en PromQL — muestra curva acumulativa de planes IA del día

CPU median suavizado (todos los nodos) MetricsQL exclusivo

range_median(cpu_usage_percent{cluster="nutriflow-prod",instance=~"gke-node-.*"})

range_median elimina outliers transitorios — mejor que avg() para capacity planning

⚡

VictoriaMetrics vs Prometheus

Benchmark para escala NutriFlow

Métrica	Prometheus	VictoriaMetrics
RAM para 800k series	~6.4 GB	~1.8 GB
Disk (90d · 2M pts/min)	~420 GB	~95 GB
Velocidad de ingesta	450k/s	2.1M/s
Tiempo query 90d range	~8.2s	~0.9s
HA / Dedup nativo	❌ Requiere Thanos	✅ -dedup flag
Retención configurable	⚠️ Limitado	✅ -retentionPeriod=90d

📝

Recording rules — Optimización Grafana

Pre-compute para dashboards rápidos

# alert-rules/recording-rules.ymlgroups:
  - name: nutriflow-precomputed
    interval: 30s
    rules:
      # RPS por servicio — panel principal Grafana
      - record: service:request_rate:5m
        expr: |
          sum(rate(http_requests_total[5m]))
          by (service)

      # Error rate — usado en alertas y SLO panel
      - record: service:error_rate:5m
        expr: |
          sum(rate(http_requests_total
            {status=~"5.."}[5m])) by (service)
          /
          sum(rate(http_requests_total[5m])) by (service)

      # P95 latencia billing-service (crítico)
      - record: billing:p95_latency:5m
        expr: |
          quantile_over_time(0.95,
            http_request_duration_seconds
              {service="billing-service"}[5m])