llm-proxy/app.py

"""
大模型API中转系统
兼容OpenAI API格式，支持多上游提供商优先级调度
"""

from flask import Flask, request, jsonify, Response, stream_with_context
from flask_cors import CORS
import requests
import json
import time
import logging
from datetime import datetime
from pathlib import Path
import sys

# 添加配置路径
sys.path.insert(0, str(Path(__file__).parent))
from config.settings import (
    get_providers, get_model_aliases, SERVER_CONFIG, 
    LOG_CONFIG, RETRY_CONFIG
)

app = Flask(__name__)
CORS(app)

# 配置缓存时间（秒）
CONFIG_CACHE_TTL = 5
_last_config_load = 0
_cached_providers = []
_cached_aliases = {}

def refresh_config():
    """动态刷新配置（支持后台管理修改）"""
    global _last_config_load, _cached_providers, _cached_aliases, provider_status
    
    current_time = time.time()
    if current_time - _last_config_load > CONFIG_CACHE_TTL:
        _cached_providers = get_providers()
        _cached_aliases = get_model_aliases()
        _last_config_load = current_time
        
        # 更新提供商状态缓存（新增的提供商）
        for provider in _cached_providers:
            if provider['name'] not in provider_status:
                provider_status[provider['name']] = {
                    'available': True,
                    'last_check': None,
                    'error_count': 0,
                    'last_error': None,
                }

# 配置日志
log_dir = Path(__file__).parent / LOG_CONFIG['log_dir']
log_dir.mkdir(exist_ok=True)

logging.basicConfig(
    level=logging.INFO,
    format='%(asctime)s - %(name)s - %(levelname)s - %(message)s',
    handlers=[
        logging.FileHandler(log_dir / 'proxy.log', encoding='utf-8'),
        logging.StreamHandler()
    ]
)
logger = logging.getLogger(__name__)

# 提供商状态缓存
provider_status = {}

# 初始化
refresh_config()


def get_provider_for_model(model_name):
    """根据模型名获取提供商"""
    refresh_config()
    
    # 解析别名
    resolved_model = _cached_aliases.get(model_name, model_name)
    
    # auto模式：按优先级选择可用提供商
    if resolved_model == 'auto':
        return get_available_provider()
    
    # 查找支持该模型的提供商
    sorted_providers = sorted(_cached_providers, key=lambda x: x['priority'])
    
    for provider in sorted_providers:
        if not provider['enabled']:
            continue
        if not provider_status.get(provider['name'], {}).get('available', True):
            continue
        if resolved_model in provider['models']:
            return provider, resolved_model
    
    # 如果没找到精确匹配，尝试模糊匹配
    for provider in sorted_providers:
        if not provider['enabled']:
            continue
        if not provider_status.get(provider['name'], {}).get('available', True):
            continue
        for m in provider['models']:
            if resolved_model.lower() in m.lower() or m.lower() in resolved_model.lower():
                return provider, m
    
    return None, None


def get_available_provider():
    """获取可用的提供商（按优先级）"""
    refresh_config()
    
    sorted_providers = sorted(_cached_providers, key=lambda x: x['priority'])
    
    for provider in sorted_providers:
        if provider['enabled'] and provider_status.get(provider['name'], {}).get('available', True):
            return provider, provider['default_model']
    
    # 如果都不可用，返回第一个尝试（让错误信息传递）
    if sorted_providers:
        return sorted_providers[0], sorted_providers[0]['default_model']
    
    return None, None


def mark_provider_error(provider_name, error):
    """标记提供商错误"""
    if provider_name in provider_status:
        provider_status[provider_name]['error_count'] += 1
        provider_status[provider_name]['last_error'] = str(error)
        provider_status[provider_name]['last_check'] = datetime.now()
        
        # 连续错误超过阈值，暂时标记不可用
        if provider_status[provider_name]['error_count'] >= 3:
            provider_status[provider_name]['available'] = False
            logger.warning(f"Provider {provider_name} marked as unavailable due to errors")


def mark_provider_success(provider_name):
    """标记提供商成功"""
    if provider_name in provider_status:
        provider_status[provider_name]['error_count'] = 0
        provider_status[provider_name]['available'] = True
        provider_status[provider_name]['last_check'] = datetime.now()


def proxy_request(provider, model, request_data, stream=False):
    """转发请求到上游提供商"""
    url = f"{provider['base_url'].rstrip('/')}/chat/completions"
    
    headers = {
        "Content-Type": "application/json",
        "Authorization": f"Bearer {provider['api_key']}"
    }
    
    # 构建请求数据
    data = request_data.copy()
    data['model'] = model
    
    try:
        if stream:
            # 流式请求
            response = requests.post(
                url,
                headers=headers,
                json=data,
                stream=True,
                timeout=provider.get('timeout', 120)
            )
            return response
        else:
            # 非流式请求
            response = requests.post(
                url,
                headers=headers,
                json=data,
                timeout=provider.get('timeout', 120)
            )
            return response
            
    except requests.exceptions.Timeout:
        mark_provider_error(provider['name'], "Timeout")
        raise Exception(f"Provider {provider['name']} timeout")
    except requests.exceptions.ConnectionError:
        mark_provider_error(provider['name'], "Connection error")
        raise Exception(f"Provider {provider['name']} connection error")
    except Exception as e:
        mark_provider_error(provider['name'], str(e))
        raise


def stream_response(response):
    """流式响应生成器"""
    try:
        for line in response.iter_lines():
            if line:
                yield line + b'\n'
    except Exception as e:
        logger.error(f"Stream error: {e}")
        yield b'data: {"error": "' + str(e).encode() + b'"}\n\n'


# ============ API 路由 ============

@app.route('/')
def index():
    """首页"""
    return jsonify({
        "name": "LLM Proxy",
        "version": "1.0.0",
        "description": "OpenAI-compatible LLM API Proxy",
        "endpoints": {
            "chat": "/v1/chat/completions",
            "models": "/v1/models",
            "health": "/health",
            "status": "/status"
        }
    })


@app.route('/v1/models', methods=['GET'])
def list_models():
    """列出可用模型"""
    refresh_config()
    
    models_list = []
    added_models = set()
    
    for provider in _cached_providers:
        if not provider['enabled']:
            continue
        for model in provider['models']:
            if model not in added_models:
                models_list.append({
                    "id": model,
                    "object": "model",
                    "created": int(time.time()),
                    "owned_by": provider['name'],
                })
                added_models.add(model)
    
    # 添加auto模型
    if "auto" not in added_models:
        models_list.insert(0, {
            "id": "auto",
            "object": "model",
            "created": int(time.time()),
            "owned_by": "proxy",
            "description": "Auto-select available model by priority"
        })
    
    return jsonify({
        "object": "list",
        "data": models_list
    })


@app.route('/v1/chat/completions', methods=['POST'])
def chat_completions():
    """聊天完成API"""
    try:
        data = request.get_json()
        
        if not data:
            return jsonify({"error": "Invalid request body"}), 400
        
        model = data.get('model', 'auto')
        stream = data.get('stream', False)
        
        # 获取提供商
        provider, resolved_model = get_provider_for_model(model)
        
        if not provider:
            return jsonify({
                "error": {
                    "message": f"No available provider for model: {model}",
                    "type": "invalid_request_error"
                }
            }), 400
        
        logger.info(f"Request: model={model} -> provider={provider['name']}, resolved_model={resolved_model}, stream={stream}")
        
        # 重试逻辑
        last_error = None
        tried_providers = []
        
        for attempt in range(RETRY_CONFIG['max_retries']):
            try:
                response = proxy_request(provider, resolved_model, data, stream)
                
                if response.status_code == 200:
                    mark_provider_success(provider['name'])
                    
                    if stream:
                        # 流式响应
                        return Response(
                            stream_with_context(stream_response(response)),
                            content_type='text/event-stream',
                            headers={
                                'Cache-Control': 'no-cache',
                                'Connection': 'keep-alive',
                            }
                        )
                    else:
                        # 非流式响应
                        return jsonify(response.json())
                
                elif response.status_code == 429:
                    # 速率限制，尝试下一个提供商
                    mark_provider_error(provider['name'], "Rate limit")
                    tried_providers.append(provider['name'])
                    
                    # 尝试下一个提供商
                    next_provider, next_model = get_available_provider()
                    if next_provider and next_provider['name'] not in tried_providers:
                        provider = next_provider
                        resolved_model = next_model
                        continue
                    
                    return jsonify(response.json()), response.status_code
                
                else:
                    last_error = response.json() if response.headers.get('content-type', '').startswith('application/json') else {"error": response.text}
                    return jsonify(last_error), response.status_code
                    
            except Exception as e:
                last_error = str(e)
                logger.error(f"Attempt {attempt + 1} failed: {e}")
                tried_providers.append(provider['name'])
                
                # 尝试下一个提供商
                next_provider, next_model = get_available_provider()
                if next_provider and next_provider['name'] not in tried_providers:
                    provider = next_provider
                    resolved_model = next_model
                    time.sleep(RETRY_CONFIG['retry_delay'])
                    continue
        
        # 所有重试都失败
        return jsonify({
            "error": {
                "message": f"All providers failed. Last error: {last_error}",
                "type": "api_error"
            }
        }), 503
        
    except Exception as e:
        logger.error(f"Unexpected error: {e}")
        return jsonify({
            "error": {
                "message": str(e),
                "type": "internal_error"
            }
        }), 500


@app.route('/v1/embeddings', methods=['POST'])
def embeddings():
    """嵌入API（简单转发）"""
    refresh_config()
    
    try:
        data = request.get_json()
        model = data.get('model', 'text-embedding-ada-002')
        
        # 使用第一个可用提供商
        if _cached_providers:
            provider = _cached_providers[0]
            
            url = f"{provider['base_url'].rstrip('/')}/embeddings"
            headers = {
                "Content-Type": "application/json",
                "Authorization": f"Bearer {provider['api_key']}"
            }
            
            response = requests.post(url, headers=headers, json=data, timeout=60)
            return jsonify(response.json()), response.status_code
        else:
            return jsonify({"error": "No providers available"}), 503
        
    except Exception as e:
        return jsonify({"error": str(e)}), 500


@app.route('/health', methods=['GET'])
def health():
    """健康检查"""
    available_count = sum(1 for s in provider_status.values() if s['available'])
    total_count = len(provider_status)
    
    return jsonify({
        "status": "healthy" if available_count > 0 else "degraded",
        "providers": {
            "available": available_count,
            "total": total_count,
        },
        "timestamp": datetime.now().isoformat()
    })


@app.route('/status', methods=['GET'])
def status():
    """详细状态"""
    refresh_config()
    
    providers_detail = []
    
    for provider in _cached_providers:
        status_info = provider_status.get(provider['name'], {})
        providers_detail.append({
            "name": provider['name'],
            "priority": provider['priority'],
            "enabled": provider['enabled'],
            "available": status_info.get('available', True),
            "error_count": status_info.get('error_count', 0),
            "last_error": status_info.get('last_error'),
            "models": provider['models'],
        })
    
    return jsonify({
        "version": "1.0.0",
        "uptime": time.time(),
        "providers": providers_detail,
        "model_aliases": _cached_aliases,
    })


# 兼容 OpenAI SDK 的其他端点
@app.route('/v1/engines', methods=['GET'])
def list_engines():
    """兼容旧版 engines 端点"""
    return list_models()


@app.route('/v1/engines/<model>/completions', methods=['POST'])
def engine_completions(model):
    """兼容旧版 completions 端点"""
    data = request.get_json()
    data['model'] = model
    return chat_completions()


if __name__ == '__main__':
    refresh_config()
    
    print("=" * 60)
    print("大模型API中转系统")
    print("=" * 60)
    print(f"访问地址: http://localhost:{SERVER_CONFIG['port']}")
    print(f"API端点: http://localhost:{SERVER_CONFIG['port']}/v1/chat/completions")
    print("=" * 60)
    print("上游提供商:")
    for p in sorted(_cached_providers, key=lambda x: x['priority']):
        print(f"  [{p['priority']}] {p['name']}: {p['base_url']}")
        print(f"       模型: {', '.join(p['models'])}")
    print("=" * 60)
    print("支持的模型别名:")
    for alias, target in _cached_aliases.items():
        print(f"  {alias} -> {target}")
    print("=" * 60)
    
    app.run(
        host=SERVER_CONFIG['host'],
        port=SERVER_CONFIG['port'],
        debug=SERVER_CONFIG['debug']
    )