api/health_manager.py

import asyncio
import logging
import time
from enum import Enum
from typing import Dict, List, Any, Optional
from pydantic import BaseModel

_logger = logging.getLogger("api.health_manager")

class HealthStatus(str, Enum):
    HEALTHY = "healthy"
    DEGRADED = "degraded"
    DOWN = "down"

class ComponentHealth(BaseModel):
    id: str
    type: str # "worker" | "provider" | "service"
    status: HealthStatus = HealthStatus.HEALTHY
    failure_count: int = 0
    last_check: float = 0.0
    latency: float = 0.0
    error_message: Optional[str] = None

class HealthManager:
    """
    ARCH-P5.1: Health Manager
    Gestisce il monitoraggio, il Circuit Breaker e il Traffic Management.
    """
    def __init__(self):
        self.components: Dict[str, ComponentHealth] = {}
        self._lock = asyncio.Lock()
        self.failure_threshold = 5 # Numero di errori prima di aprire il circuit
        self.recovery_timeout = 60 # Secondi prima di riprovare un componente DOWN

    async def record_success(self, component_id: str, latency: float = 0.0, component_type: str = "worker"):
        """Registra un'operazione riuscita per un componente."""
        async with self._lock:
            if component_id not in self.components:
                self.components[component_id] = ComponentHealth(id=component_id, type=component_type)
            
            c = self.components[component_id]
            c.status = HealthStatus.HEALTHY
            c.failure_count = 0
            c.last_check = time.time()
            c.latency = latency
            c.error_message = None

    async def record_failure(self, component_id: str, error: str, component_type: str = "worker"):
        """Registra un fallimento e attiva il circuit breaker se necessario."""
        async with self._lock:
            if component_id not in self.components:
                self.components[component_id] = ComponentHealth(id=component_id, type=component_type)
            
            c = self.components[component_id]
            c.failure_count += 1
            c.last_check = time.time()
            c.error_message = error
            
            if c.failure_count >= self.failure_threshold:
                if c.status != HealthStatus.DOWN:
                    _logger.warning(f"Circuit Breaker APERTO per {component_id}: {error}")
                    c.status = HealthStatus.DOWN
            elif c.failure_count >= 2:
                c.status = HealthStatus.DEGRADED

    async def is_healthy(self, component_id: str) -> bool:
        """Verifica se un componente è sano (o se è tempo di riprovare)."""
        async with self._lock:
            if component_id not in self.components:
                return True
            
            c = self.components[component_id]
            if c.status == HealthStatus.DOWN:
                # Half-open state: riprova dopo il timeout
                if time.time() - c.last_check > self.recovery_timeout:
                    _logger.info(f"Circuit Breaker HALF-OPEN per {component_id} (tentativo di recovery)")
                    return True
                return False
            return True

    async def get_status(self) -> Dict[str, Any]:
        """Ritorna lo stato aggregato di salute del sistema."""
        async with self._lock:
            return {
                "ts": time.time(),
                "components": {k: v.dict() for k, v in self.components.items()},
                "summary": {
                    "healthy": sum(1 for c in self.components.values() if c.status == HealthStatus.HEALTHY),
                    "degraded": sum(1 for c in self.components.values() if c.status == HealthStatus.DEGRADED),
                    "down": sum(1 for c in self.components.values() if c.status == HealthStatus.DOWN),
                }
            }

# Singleton instance
health_manager = HealthManager()