Arquitectura

Section titled “Arquitectura”
Gateway (:8008/metrics)
    |
    +-- Prometheus scrape (15s)
    |
    +-- runtime/analytics/runtime_analytics.py
            |                       |
            |  +-- health_score.py  |  (lee gateway metrics + cluster state)
            |  +-- session_metrics  |  (sesiones activas)
            |  +-- routing_metrics  |  (rutas por modelo)
            |
            v
    Live API (:8084/api/analytics)
            |
            +-- /api/analytics (JSON)
            |
            v
    Astro Portal
    +-- /ops (Command Center)
    +-- /status/history

Componentes

Section titled “Componentes”

runtime/analytics/runtime_analytics.py

Section titled “runtime/analytics/runtime_analytics.py”
  • Lee metricas del gateway via :8008/metrics
  • Lee estado del cluster via cluster_state.json
  • Retorna metricas agregadas: requests, streams, errores, routing, latencia, nodos

runtime/analytics/health_score.py

Section titled “runtime/analytics/health_score.py”
  • Score 0-100 evaluando 6 factores
  • Penalizaciones: GPUs offline (-30), Gateway (-15), Router (-10), Prometheus (-10), Docker (-5), Latencia (-20)
  • Niveles: perfect (90+), good (70+), degraded (50+), poor (30+), critical (<30)

runtime/analytics/session_metrics.py

Section titled “runtime/analytics/session_metrics.py”
  • Sesiones activas, totales, duracion media

runtime/analytics/routing_metrics.py

Section titled “runtime/analytics/routing_metrics.py”
  • Rutas por tarea, failovers, ultimas 20 rutas

API Endpoint

Section titled “API Endpoint”
GET /api/analytics


Response:
{
  "health": {"score": 85, "level": "good", "reasons": ["1 GPU(s) offline"]},
  "metrics": {"requests_total": 18, "events_per_minute": 0, ...},
  "sessions": {"total_sessions": 0, "active_sessions": 0, ...},
  "routing": {"total_routes": 0, "failovers": 0, ...},
  "event_bus": {"total_events": 0, "active_listeners": 0, ...},
  "timestamp": 1747336800.0
}

Integracion con Traefik

Section titled “Integracion con Traefik”

El endpoint /api/analytics se sirve a traves de Traefik para el blog privado:

ai-lab-docs-api:
  rule: "Host(`blog-ai-lab.labrazahome.com`) && PathPrefix(`/api/`)"
  service: ai-lab-docs-api
  priority: 100

El priority: 100 es necesario para que esta ruta tenga prioridad sobre la ruta principal de docs.

Persistencia post-reboot

Section titled “Persistencia post-reboot”

Todos los componentes necesarios arrancan automaticamente:

Terminal window
# Servicios verificados
systemctl is-active ailab-live-api   # Live API con analytics endpoint
systemctl is-active ailab-docs        # Astro portal
systemctl is-active ailab-gateway     # Gateway con metricas
systemctl is-active ailab-router      # Router API
systemctl is-active ailab-traefik     # Proxy inverso (via systemd)
systemctl is-active ailab-runner      # GitHub Actions runner

Ademas, cron @reboot ejecuta scripts/startup.sh para verificar stacks Docker.

Correccion de Metricas (2026-05-15)

Section titled “Correccion de Metricas (2026-05-15)”

Ver Runtime Analytics — Correccion para el diagnostico de los bugs de health score, nodos online y gateway health check.