- Agent tracks consecutive reports above cpu_threshold (default 90%)
- After renice_consecutive reports (default 2 = 60s sustained), runs
renice -n 19 on the top CPU process
- Reports the action back in the heartbeat payload
- Backend sends Telegram alert when renice is applied
- All params configurable in agent.yml and install.sh flags
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Filter by RAM first (cheap), then call CPUPercent on just the 10 winners.
Reduces /proc reads from hundreds to 10 per interval.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
CPUPercent() on every process reads /proc/<pid>/stat for all processes,
causing ~19% CPU spike on servers with many containers. Sorting by RAM
is a single /proc read per process and much cheaper.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Backend:
- Servidor model: AgentToken, AgentLastSeen, MetricasJson nuevos campos
- AgentHeartbeat: endpoint público POST /agent/heartbeat (auth por token)
- GenerateAgentToken: endpoint protegido POST /app/servidor/:id/agent-token
- Ruta pública /agent/install.sh sirve el script de instalación
Agente Go (agent/):
- agent/main.go: binario independiente con gopsutil
- Recolecta RAM, CPU, Disco, Uptime, OS, Load average
- Lee agent.yml o flags --api-url / --token
- Envía POST a /agent/heartbeat cada N segundos
- Instala como servicio systemd via install.sh
- agent/go.mod: módulo independiente (usite-agent)
- agent/agent.yml.sample: configuración de ejemplo
- agent/install.sh: instalador en 1 curl para Linux (systemd)
Frontend servidor_dashboard.html:
- Cards: badge ● En línea / ● Fuera / ○ Sin agente
- Barras de progreso live: RAM%, CPU%, Disco%
- Valores: GB usado/total, núcleos, uptime, OS real
- Tiempo desde último reporte
- Botón 'Agente' en cada card → modal de configuración
- Modal agente: estado, token con copy, comando curl 1 línea,
instalación manual/avanzada con collapse, métricas actuales
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
