# cloud-config.yaml
clouds:
  aliyun:
    provider: aliyun
    region: cn-hangzhou
    instance_type: ecs.g7.xlarge
    monthly_cost: 389
    status: active
    
  aws:
    provider: aws
    region: ap-southeast-1
    instance_type: t5.medium
    monthly_cost: 245
    status: active
    
  digitalocean:
    provider: digitalocean
    region: sgp1
    instance_type: c-2
    monthly_cost: 192
    status: active

# ai_scheduler/agent.py
import asyncio
import json
from typing import Dict, List, Optional
from langchain.agents import initialize_agent, AgentType
from langchain.chat_models import ChatOpenAI
from langchain.tools import Tool

class CloudResourceScheduler:
    """AI 驱动的多云资源调度器"""
    
    def __init__(self, llm_model="gpt-4o"):
        self.llm = ChatOpenAI(model=llm_model, temperature=0.1)
        self.clouds = {}
        self.metrics_history = []
        
    async def fetch_cloud_prices(self, cloud_name: str) -> Dict:
        """获取指定云厂商的实时价格"""
        # 实际项目中对接各云 API
        price_map = {
            "aliyun": self._fetch_aliyun_pricing,
            "aws": self._fetch_aws_pricing,
            "digitalocean": self._fetch_do_pricing,
        }
        if cloud_name in price_map:
            return await price_map[cloud_name]()
        raise ValueError(f"Unsupported cloud: {cloud_name}")
    
    async def _fetch_aliyun_pricing(self) -> Dict:
        """获取阿里云价格信息"""
        # 调用阿里云 OpenAPI
        return {
            "instances": [
                {"type": "ecs.g7.xlarge", "price_per_hour": 1.82, "currency": "CNY"},
                {"type": "ecs.c7.2xlarge", "price_per_hour": 2.95, "currency": "CNY"},
            ],
            "spot_discount": 0.65,  # 竞价实例折扣 35%
        }
    
    async def _fetch_aws_pricing(self) -> Dict:
        """获取 AWS 价格信息"""
        return {
            "instances": [
                {"type": "t5.medium", "price_per_hour": 0.026, "currency": "USD"},
                {"type": "m6i.xlarge", "price_per_hour": 0.192, "currency": "USD"},
            ],
            "spot_discount": 0.70,
        }
    
    async def _fetch_do_pricing(self) -> Dict:
        """获取 DigitalOcean 价格信息"""
        return {
            "instances": [
                {"type": "c-2", "price_per_hour": 0.265, "currency": "USD"},
                {"type": "c-4", "price_per_hour": 0.53, "currency": "USD"},
            ],
            "spot_discount": 0.81,
        }
    
    async def predict_load(self, hours_ahead: int = 24) -> Dict:
        """预测未来负载"""
        # 使用简单的线性回归 + 周期模式
        # 实际项目中使用 Prophet 或 LSTM
        base_cpu = 35.0
        trend = 0.5  # 每日增长趋势
        daily_cycle = lambda h: 20 * abs((h % 24) - 12) / 12
        
        predictions = {}
        for h in range(hours_ahead):
            cpu_pct = min(100, base_cpu + trend * (h / 24) + daily_cycle(h))
            predictions[h] = {
                "cpu_percent": round(cpu_pct, 1),
                "memory_gb": round(cpu_pct * 0.064, 1),  # 假设 16GB 总内存
                "bandwidth_mbps": round(cpu_pct * 0.1, 1),
            }
        return predictions
    
    def optimize_costs(self, predictions: Dict, prices: Dict[str, Dict]) -> Dict:
        """AI 驱动的成本优化决策"""
        # 构建优化问题
        best_config = None
        min_cost = float('inf')
        
        for cloud_name, cloud_data in prices.items():
            for inst in cloud_data.get("instances", []):
                # 计算每小时成本（考虑竞价折扣）
                spot_price = inst["price_per_hour"] * (1 - cloud_data.get("spot_discount", 0))
                
                # 估算月度成本
                monthly_cost = spot_price * 24 * 30
                
                # 转换为 CNY（简化汇率）
                exchange_rate = 7.2 if inst["currency"] == "USD" else 1.0
                monthly_cny = monthly_cost * exchange_rate
                
                if monthly_cny < min_cost:
                    min_cost = monthly_cny
                    best_config = {
                        "cloud": cloud_name,
                        "instance_type": inst["type"],
                        "hourly_cost": spot_price,
                        "monthly_cost_cny": round(monthly_cny, 2),
                    }
        
        return best_config
    
    async def generate_report(self) -> str:
        """生成 AI 运维日报"""
        # 获取当前数据
        prices = {}
        for cloud in self.clouds.keys():
            prices[cloud] = await self.fetch_cloud_prices(cloud)
        
        predictions = await self.predict_load(24)
        optimization = self.optimize_costs(predictions, prices)
        
        # 使用 LLM 生成自然语言报告
        prompt = f"""你是一个专业的 IT 运维顾问。请根据以下数据生成一份简洁的 AI 调度日报：

当前多云资源配置与价格：
{json.dumps(prices, indent=2, ensure_ascii=False)}

未来24小时负载预测：
{json.dumps({k: v for k, v in list(predictions.items())[:5]}, indent=2, ensure_ascii=False)}

成本优化建议：
{json.dumps(optimization, indent=2, ensure_ascii=False)}

请按以下格式输出：
1. 📊 当前成本概览
2. 🔮 负载趋势预测
3. 💡 优化建议
4. ⚠️ 风险提示
"""
        
        response = self.llm.predict(prompt)
        return response

# crontab -e
# 每 15 分钟检查一次云价格
*/15 * * * * /opt/ai-scheduler/bin/check-pricing.sh >> /var/log/scheduler.log 2>&1

# 每小时生成一次负载预测
0 * * * * /opt/ai-scheduler/bin/predict-load.sh >> /var/log/scheduler.log 2>&1

# 每天上午 9 点生成运维日报
0 9 * * * /opt/ai-scheduler/bin/daily-report.sh >> /var/log/scheduler.log 2>&1

# 每周一上午 10 点执行深度成本优化
0 10 * * 1 /opt/ai-scheduler/bin/weekly-optimize.sh >> /var/log/scheduler.log 2>&1

# migration_rules.py
"""定义自动迁移的策略和条件"""

MIGRATION_RULES = {
    "cost_threshold": {
        "description": "当成本超过阈值时触发迁移",
        "max_monthly_cost_cny": 500,
        "action": "switch_to_cheapest_cloud",
    },
    "performance_threshold": {
        "description": "当性能低于 SLA 时触发扩容",
        "min_cpu_available": 20,  # 至少保留 20% CPU
        "max_response_time_ms": 200,
        "action": "upgrade_instance_or_migrate",
    },
    "availability_threshold": {
        "description": "当可用性低于目标时触发故障转移",
        "target_uptime": 99.95,
        "max_downtime_minutes": 4,
        "action": "failover_to_backup_cloud",
    },
    "spot_interruption": {
        "description": "竞价实例被回收时的处理策略",
        "warning_time_minutes": 2,
        "action": "graceful_migration_to_on_demand",
    },
}

# smart_reporter.py
from langchain.prompts import PromptTemplate

def generate_explanatory_report(decisions: dict) -> str:
    """生成可解释的 AI 运维报告"""
    
    template = PromptTemplate(
        input_variables=["cost_savings", "recommendations", "risks"],
        template="""
## 📊 本周 AI 调度报告

### 成本节约总结
本周通过智能调度共节约 ¥{cost_savings}，主要来源于：
- 跨云比价迁移：减少 ¥XXX
- 竞价实例利用：减少 ¥XXX
- 闲置资源回收：减少 ¥XXX

### 智能决策记录
{recommendations}

### 风险提示
{risks}

### 下周预测
根据历史数据和当前趋势，预计下周：
- 负载将在周三达到峰值
- 建议提前扩容或准备弹性资源
- 某云厂商可能有价格调整，持续关注
        """
    )
    
    return template.format(**decisions)

## 📊 本周 AI 调度报告

### 成本节约总结
本周通过智能调度共节约 ¥432，主要来源于：
- 跨云比价迁移：减少 ¥218
- 竞价实例利用：减少 ¥156
- 闲置资源回收：减少 ¥58

### 智能决策记录
- 周一 09:00：检测到阿里云存储费用异常增长，建议迁移至 S3
- 周三 14:30：促销活动期间自动扩容至 AWS，活动结束后释放
- 周五 22:00：识别出 2 台低负载开发机，建议合并

### 风险提示
- 腾讯云 ap-sea 区域实例供应紧张，建议提前准备备用方案
- 某云厂商即将调价，预计涨幅 10%-15%

# 使用 Vault 管理云厂商 API 密钥
vault kv put secret/cloud/aliyun \
  access_key_id="AK..." \
  access_key_secret="SK..."
  
vault kv put secret/cloud/aws \
  aws_access_key_id="AKIA..." \
  aws_secret_access_key="wJalr..."

{
  "Version": "2012-10-17",
  "Statement": [
    {
      "Effect": "Allow",
      "Action": [
        "ec2:Describe*",
        "ec2:StartInstances",
        "ec2:StopInstances",
        "ec2:TerminateInstances",
        "ec2:RunInstances"
      ],
      "Resource": "*"
    }
  ]
}