geo/backend/app/services/health_checker.py

"""详细健康检查服务"""
import time
from dataclasses import dataclass
from typing import Optional

import redis.asyncio as aioredis
from sqlalchemy import text
from sqlalchemy.ext.asyncio import AsyncSession


@dataclass
class HealthCheckResult:
    """健康检查结果"""
    name: str
    healthy: bool
    latency_ms: Optional[float] = None
    message: Optional[str] = None
    details: Optional[dict] = None


class HealthChecker:
    """健康检查服务"""

    def __init__(self, db: AsyncSession, redis_url: str):
        self.db = db
        self.redis_url = redis_url

    async def check_database(self) -> HealthCheckResult:
        """检查数据库连接"""
        start = time.perf_counter()
        try:
            await self.db.execute(text("SELECT 1"))
            latency = (time.perf_counter() - start) * 1000

            return HealthCheckResult(
                name="database",
                healthy=True,
                latency_ms=round(latency, 2),
                message="Connection OK",
            )
        except Exception as e:
            latency = (time.perf_counter() - start) * 1000
            return HealthCheckResult(
                name="database",
                healthy=False,
                latency_ms=round(latency, 2),
                message=f"Connection failed: {str(e)}",
            )

    async def check_redis(self) -> HealthCheckResult:
        """检查Redis连接"""
        start = time.perf_counter()
        try:
            from app.core.redis import get_redis
            redis = await get_redis()
            await redis.ping()

            latency = (time.perf_counter() - start) * 1000
            return HealthCheckResult(
                name="redis",
                healthy=True,
                latency_ms=round(latency, 2),
                message="Connection OK",
            )
        except Exception as e:
            latency = (time.perf_counter() - start) * 1000
            return HealthCheckResult(
                name="redis",
                healthy=False,
                latency_ms=round(latency, 2),
                message=f"Connection failed: {str(e)}",
            )

    async def check_llm_providers(self) -> HealthCheckResult:
        """检查LLM服务提供商"""
        from app.config import settings
        from app.services.llm.factory import LLMFactory

        providers = {}
        all_healthy = True

        # 检查默认provider
        try:
            provider_name = getattr(settings, 'DEFAULT_LLM_PROVIDER', 'openai')
            provider = LLMFactory.create(provider_name)
            providers[provider_name] = {
                "healthy": True,
                "available": True,
            }
        except Exception as e:
            providers[getattr(settings, 'DEFAULT_LLM_PROVIDER', 'openai')] = {
                "healthy": False,
                "error": str(e),
            }
            all_healthy = False

        # 检查所有已注册的provider
        for name in LLMFactory.list_providers():
            if name not in providers:
                try:
                    provider = LLMFactory.create(name)
                    providers[name] = {
                        "healthy": True,
                        "available": True,
                    }
                except Exception as e:
                    providers[name] = {
                        "healthy": False,
                        "error": str(e),
                    }
                    all_healthy = False

        return HealthCheckResult(
            name="llm_providers",
            healthy=all_healthy,
            message="All providers healthy" if all_healthy else "Some providers unhealthy",
            details={"providers": providers},
        )

    async def check_storage(self) -> HealthCheckResult:
        """检查存储（本地文件系统）"""
        import os

        storage_path = "/data/documents"

        try:
            if os.path.exists(storage_path):
                # 检查读写权限
                test_file = os.path.join(storage_path, ".health_check")
                with open(test_file, "w") as f:
                    f.write("ok")
                os.remove(test_file)

                return HealthCheckResult(
                    name="storage",
                    healthy=True,
                    message=f"Storage path {storage_path} is writable",
                    details={"path": storage_path},
                )
            else:
                return HealthCheckResult(
                    name="storage",
                    healthy=True,
                    message=f"Storage path {storage_path} does not exist (will be created)",
                    details={"path": storage_path, "created": True},
                )
        except Exception as e:
            return HealthCheckResult(
                name="storage",
                healthy=False,
                message=f"Storage check failed: {str(e)}",
            )

    async def check_all(self) -> dict:
        """执行所有健康检查"""
        import asyncio

        # 并行执行所有检查
        checks = [
            self.check_database(),
            self.check_redis(),
            self.check_llm_providers(),
            self.check_storage(),
        ]

        results = await asyncio.gather(*checks)

        # 汇总结果
        all_healthy = all(r.healthy for r in results)

        return {
            "status": "healthy" if all_healthy else "degraded",
            "timestamp": time.time(),
            "checks": {
                r.name: {
                    "healthy": r.healthy,
                    "latency_ms": r.latency_ms,
                    "message": r.message,
                    "details": r.details,
                }
                for r in results
            },
        }