app.py

from flask import Flask, request, jsonify, Response, stream_with_context, render_template_string
from google.generativeai.types import StopCandidateException, generation_types
from google.api_core.exceptions import InvalidArgument, ResourceExhausted, Aborted, InternalServerError, ServiceUnavailable, PermissionDenied
import google.generativeai as genai
import json
import os
import re
import logging
import func
from datetime import datetime, timedelta
from apscheduler.schedulers.background import BackgroundScheduler
import time
import requests
from collections import deque
import random
import codecs
import json
from dataclasses import dataclass
from typing import Optional, Iterator, Dict, Any

app = Flask(__name__)

os.environ['TZ'] = 'Asia/Shanghai'

app = Flask(__name__)

app.secret_key = os.urandom(24)

formatter = logging.Formatter('%(message)s')
logger = logging.getLogger(__name__)
logger.setLevel(logging.INFO)
handler = logging.StreamHandler()
handler.setFormatter(formatter)
logger.addHandler(handler)

MAX_RETRIES = int(os.environ.get('MaxRetries', 3))
MAX_REQUESTS = int(os.environ.get('MaxRequests', 2))
LIMIT_WINDOW = int(os.environ.get('LimitWindow', 60))
RETRY_DELAY = 1
MAX_RETRY_DELAY = 16

request_counts = {}

api_key_blacklist = set()
api_key_blacklist_duration = 60


safety_settings = [
    {
        "category": "HARM_CATEGORY_HARASSMENT",
        "threshold": "BLOCK_NONE"
    },
    {
        "category": "HARM_CATEGORY_HATE_SPEECH",
        "threshold": "BLOCK_NONE"
    },
    {
        "category": "HARM_CATEGORY_SEXUALLY_EXPLICIT",
        "threshold": "BLOCK_NONE"
    },
    {
        "category": "HARM_CATEGORY_DANGEROUS_CONTENT",
        "threshold": "BLOCK_NONE"
    }
]
safety_settings_g2 = [
    {
        "category": "HARM_CATEGORY_HARASSMENT",
        "threshold": "OFF"
    },
    {
        "category": "HARM_CATEGORY_HATE_SPEECH",
        "threshold": "OFF"
    },
    {
        "category": "HARM_CATEGORY_SEXUALLY_EXPLICIT",
        "threshold": "OFF"
    },
    {
        "category": "HARM_CATEGORY_DANGEROUS_CONTENT",
        "threshold": "OFF"
    }
]
@dataclass
class GeneratedText:
    """用于存储生成的文本片段"""
    text: str
    finish_reason: Optional[str] = None


class ResponseWrapper:
    """处理非流式响应的包装类"""
    def __init__(self, data: Dict[Any, Any]):
        self._data = data
        self._text = self._extract_text()
        self._finish_reason = self._extract_finish_reason()

    def _extract_text(self) -> str:
        """从响应数据中提取文本"""
        try:
            return self._data['candidates'][0]['content']['parts'][0]['text']
        except (KeyError, IndexError):
            return ""

    def _extract_finish_reason(self) -> Optional[str]:
        """提取完成原因"""
        try:
            return self._data['candidates'][0].get('finishReason')
        except (KeyError, IndexError):
            return None

    @property
    def text(self) -> str:
        """获取响应文本"""
        return self._text

    @property
    def finish_reason(self) -> Optional[str]:
        """获取完成原因"""
        return self._finish_reason

class APIKeyManager:
    def __init__(self):
        self.api_keys = re.findall(r"AIzaSy[a-zA-Z0-9_-]{33}", os.environ.get('KeyArray'))
        self.current_index = random.randint(0, len(self.api_keys) - 1)

    def get_available_key(self):
        num_keys = len(self.api_keys)
        for _ in range(num_keys):
            if self.current_index >= num_keys:
                self.current_index = 0
            current_key = self.api_keys[self.current_index]
            self.current_index += 1

            if current_key not in api_key_blacklist:
                return current_key

        logger.error("所有API key都已耗尽或被暂时禁用，请重新配置或稍后重试")
        return None

    def show_all_keys(self):
        logger.info(f"当前可用API key个数: {len(self.api_keys)} ")
        for i, api_key in enumerate(self.api_keys):
            logger.info(f"API Key{i}: {api_key[:11]}...{api_key[-3:]}")

    def blacklist_key(self, key):
        logger.warning(f"{key[:11]} → 暂时禁用 {api_key_blacklist_duration} 秒")
        api_key_blacklist.add(key)

        scheduler.add_job(lambda: api_key_blacklist.discard(key), 'date', run_date=datetime.now() + timedelta(seconds=api_key_blacklist_duration))

key_manager = APIKeyManager()
key_manager.show_all_keys()
current_api_key = key_manager.get_available_key()

def switch_api_key():
    global current_api_key
    key = key_manager.get_available_key()
    if key:
      current_api_key = key
      logger.info(f"API key 替换为 → {current_api_key[:11]}...{current_api_key[-3:]}")
    else:
      logger.error("API key 替换失败，所有API key都已耗尽或被暂时禁用，请重新配置或稍后重试")

logger.info(f"当前 API key: {current_api_key[:11]}...{current_api_key[-3:]}")

GEMINI_MODELS = [
    {"id": "gemini-1.5-flash-8b-latest"},
    {"id": "gemini-1.5-flash-8b-exp-0924"},
    {"id": "gemini-1.5-flash-latest"},
    {"id": "gemini-1.5-flash-exp-0827"},
    {"id": "gemini-1.5-pro-latest"},
    {"id": "gemini-1.5-pro-exp-0827"},
    {"id": "learnlm-1.5-pro-experimental"},
    {"id": "gemini-exp-1114"},
    {"id": "gemini-exp-1121"},
    {"id": "gemini-exp-1206"},
    {"id": "gemini-2.0-flash-exp"},
    {"id": "gemini-2.0-flash-thinking-exp-1219"},
    {"id": "gemini-2.0-pro-exp"}
]

@app.route('/')
def index():
    main_content = "Moonfanz Reminiproxy v2.3.0 2025-01-11"
    html_template = """
<!DOCTYPE html>
<html>
<head>
<meta charset="utf-8">
<script>
function copyToClipboard(text) {
  var textarea = document.createElement("textarea");
  textarea.textContent = text;
  textarea.style.position = "fixed";
  document.body.appendChild(textarea);
  textarea.select();
  try {
    return document.execCommand("copy");
  } catch (ex) {
    console.warn("Copy to clipboard failed.", ex);
    return false;
  } finally {
    document.body.removeChild(textarea);
  }
}
function copyLink(event) {
  event.preventDefault();
  const url = new URL(window.location.href);
  const link = url.protocol + '//' + url.host + '/hf/v1';
  copyToClipboard(link);
  alert('链接已复制: ' + link);
}
</script>
</head>
<body>
{{ main_content }}<br/><br/>完全开源、免费且禁止商用<br/><br/>点击复制反向代理: <a href="v1" onclick="copyLink(event)">Copy Link</a><br/>聊天来源选择"自定义(兼容 OpenAI)"<br/>将复制的网址填入到自定义端点<br/>将设置password填入自定义API秘钥<br/><br/><br/>
</body>
</html>
    """
    return render_template_string(html_template, main_content=main_content)

def is_within_rate_limit(api_key):
    now = datetime.now()
    if api_key not in request_counts:
        request_counts[api_key] = deque()

    while request_counts[api_key] and request_counts[api_key][0] < now - timedelta(seconds=LIMIT_WINDOW):
        request_counts[api_key].popleft()

    if len(request_counts[api_key]) >= MAX_REQUESTS:
        earliest_request_time = request_counts[api_key][0]
        wait_time = (earliest_request_time + timedelta(seconds=LIMIT_WINDOW)) - now
        return False, wait_time.total_seconds()
    else:
        return True, 0

def increment_request_count(api_key):
    now = datetime.now()
    if api_key not in request_counts:
        request_counts[api_key] = deque()
    request_counts[api_key].append(now)

def handle_api_error(error, attempt):
    if attempt > MAX_RETRIES:
        logger.error(f"{MAX_RETRIES} 次尝试后仍然失败，请修改预设或输入")
        return 0, jsonify({
                'error': {
                    'message': f"{MAX_RETRIES} 次尝试后仍然失败，请修改预设或输入",
                    'type': 'max_retries_exceeded'
                }
        })

    if isinstance(error, InvalidArgument):
        logger.error(f"{current_api_key[:11]} → 无效，可能已过期或被删除")
        key_manager.blacklist_key(current_api_key)
        switch_api_key()
        return 0, None

    elif isinstance(error, ResourceExhausted):
        delay = min(RETRY_DELAY * (2 ** attempt), MAX_RETRY_DELAY)
        logger.warning(f"{current_api_key[:11]} → 429 官方资源耗尽 → {delay} 秒后重试...")
        key_manager.blacklist_key(current_api_key)
        switch_api_key()
        time.sleep(delay)
        return 0, None

    elif isinstance(error, Aborted):
        delay = min(RETRY_DELAY * (2 ** attempt), MAX_RETRY_DELAY)
        logger.warning(f"{current_api_key[:11]} → 操作被中止 → {delay} 秒后重试...")
        switch_api_key()
        time.sleep(delay)
        return 0, None

    elif isinstance(error, InternalServerError):
        delay = min(RETRY_DELAY * (2 ** attempt), MAX_RETRY_DELAY)
        logger.warning(f"{current_api_key[:11]} → 500 服务器内部错误 → {delay} 秒后重试...")
        switch_api_key()
        time.sleep(delay)
        return 0, None

    elif isinstance(error, ServiceUnavailable):
        delay = min(RETRY_DELAY * (2 ** attempt), MAX_RETRY_DELAY)
        logger.warning(f"{current_api_key[:11]} → 503 服务不可用 → {delay} 秒后重试...")
        switch_api_key()
        time.sleep(delay)
        return 0, None

    elif isinstance(error, PermissionDenied):
        logger.error(f"{current_api_key[:11]} → 403 权限被拒绝，该 API KEY 可能已经被官方封禁")
        key_manager.blacklist_key(current_api_key)
        switch_api_key()
        return 0, None

    elif isinstance(error, StopCandidateException):
        logger.warning(f"AI输出内容被Gemini官方阻挡，代理没有得到有效回复")
        switch_api_key()
        return 0, None

    elif isinstance(error, generation_types.BlockedPromptException):
        try:
            full_reason_str = str(error.args[0])
            logger.error(f"{full_reason_str}")
            if "block_reason:" in full_reason_str:
                start_index = full_reason_str.find("block_reason:") + len("block_reason:")
                block_reason_str = full_reason_str[start_index:].strip()

                if block_reason_str == "SAFETY":
                    logger.warning(f"用户输入因安全原因被阻止")
                    return 1, None
                elif block_reason_str == "BLOCKLIST":
                    logger.warning(f"用户输入因包含阻止列表中的术语而被阻止")
                    return 1, None
                elif block_reason_str == "PROHIBITED_CONTENT":
                    logger.warning(f"用户输入因包含禁止内容而被阻止")
                    return 1, None
                elif block_reason_str == "OTHER":
                    logger.warning(f"用户输入因未知原因被阻止")
                    return 1, None
                else:
                    logger.warning(f"用户输入被阻止，原因未知: {block_reason_str}")
                    return 1, None
            else:
                logger.warning(f"用户输入被阻止，原因未知: {full_reason_str}")
                return 1, None

        except (IndexError, AttributeError) as e:
            logger.error(f"获取提示原因失败↙\n{e}")
            logger.error(f"提示被阻止↙\n{error}")
            return 2, None

    else:
        logger.error(f"该模型还未发布，暂时不可用，请更换模型或未来一段时间再试")
        logger.error(f"证明↙\n{error}")
        return 2, None

@app.route('/hf/v1/chat/completions', methods=['POST'])
def chat_completions():
    is_authenticated, auth_error, status_code = func.authenticate_request(request)
    if not is_authenticated:
        return auth_error if auth_error else jsonify({'error': '未授权'}), status_code if status_code else 401

    request_data = request.get_json()
    messages = request_data.get('messages', [])
    model = request_data.get('model', 'gemini-2.0-flash-exp')
    temperature = request_data.get('temperature', 1)
    max_tokens = request_data.get('max_tokens', 8192)
    stream = request_data.get('stream', False)
    hint = "流式" if stream else "非流"
    logger.info(f"\n{model} [{hint}] → ...")
    is_thinking = 'thinking' in model
    api_version = 'v1alpha' if is_thinking else 'v1beta'
    response_type = 'streamGenerateContent' if stream else 'generateContent'
    is_SSE = '&alt=sse' if stream else ''

    gemini_history, system_instruction, error_response = func.process_messages_for_gemini(messages)

    if error_response:
        logger.error(f"处理输入消息时出错↙\n {error_response}")
        return jsonify(error_response), 400

    def do_request(current_api_key, attempt):
        isok, time_remaining = is_within_rate_limit(current_api_key)
        if not isok:
            logger.warning(f"暂时超过限额，该API key将在 {time_remaining} 秒后启用...")
            switch_api_key()
            return 0, None

        increment_request_count(current_api_key)


        url = f"https://generativelanguage.googleapis.com/{api_version}/models/{model}:{response_type}?key={current_api_key}{is_SSE}"
        headers = {
            "Content-Type": "application/json",
        }

        data = {
            "contents": gemini_history,
            "generationConfig": {
                "temperature": temperature,
                "maxOutputTokens": max_tokens,
            },
            "safetySettings": safety_settings_g2 if model == 'gemini-2.0-flash-exp' else safety_settings,
        }
        if system_instruction:
            data["system_instruction"] = system_instruction

        try:
            response = requests.post(url, headers=headers, json=data, stream=True)
            response.raise_for_status()

            if stream:

                return 1, response
            else:

                return 1, ResponseWrapper(response.json())
        except requests.exceptions.RequestException as e:
            return handle_api_error(e, attempt)

    def generate_stream(response):
        buffer = b""  
        try:
            for line in response.iter_lines():
                if not line:
                    continue
                try:
                    if line.startswith(b'data: '):
                        line = line[6:]

                    buffer += line

                    try:
                        data = json.loads(buffer.decode('utf-8'))
                        buffer = b""
                        if 'candidates' in data and data['candidates']:
                            candidate = data['candidates'][0]
                            if 'content' in candidate:
                                content = candidate['content']
                                if 'parts' in content and content['parts']:
                                    text = content['parts'][0].get('text', '')
                                    finish_reason = candidate.get('finishReason')

                                    if text:
                                        data = {
                                            'choices': [{
                                                'delta': {
                                                    'content': text
                                                },
                                                'finish_reason': finish_reason,
                                                'index': 0
                                            }],
                                            'object': 'chat.completion.chunk'
                                        }
                                        yield f"data: {json.dumps(data)}\n\n"

                    except json.JSONDecodeError:
                        logger.debug(f"JSONDecodeError, buffer now: {buffer}")
                        continue

                except Exception as e:
                    logger.error(f"Stream error during processing: {e}, Raw data line: {line}")
                    yield f"data: {json.dumps({'error': str(e)})}\n\n"

            yield f"data: {json.dumps({'choices': [{'delta': {}, 'finish_reason': 'stop', 'index': 0}]})}\n\n"

        except Exception as e:
            logger.error(f"Stream error: {e}")
            yield f"data: {json.dumps({'error': str(e)})}\n\n"

    attempt = 0
    success = 0
    response = None
    for attempt in range(1, MAX_RETRIES + 1):
        logger.info(f"第 {attempt}/{MAX_RETRIES} 次尝试 ...")
        success, response = do_request(current_api_key, attempt)

        if success == 1:
            break
        elif success == 2:

            logger.error(f"{model} 很可能暂时不可用，请更换模型或未来一段时间再试")
            response = {
                'error': {
                    'message': f'{model} 很可能暂时不可用，请更换模型或未来一段时间再试',
                    'type': 'internal_server_error'
                }
            }
            return jsonify(response), 503

    else:
        logger.error(f"{MAX_RETRIES} 次尝试均失败，请调整配置，或等待官方恢复，或向Moonfanz反馈")
        response = {
            'error': {
                'message': f'{MAX_RETRIES} 次尝试均失败，请调整配置或向Moonfanz反馈',
                'type': 'internal_server_error'
            }
        }
        return jsonify(response), 500 if response is not None else 503

    if stream:
        return Response(
                stream_with_context(generate_stream(response)),
                mimetype='text/event-stream'
            )
    else:
        try:
            text_content = response.text
        except (AttributeError, IndexError, TypeError, ValueError) as e:
            if "response.candidates" in str(e) or "response.text" in str(e):
                logger.error(f"用户输入被AI安全过滤器阻止")
                return jsonify({
                    'error': {
                        'message': '用户输入被AI安全过滤器阻止',
                        'type': 'prompt_blocked_error',
                        'details': str(e)
                    }
                }), 400
            else:
                return jsonify({
                    'error': {
                        'message': 'AI响应处理失败',
                        'type': 'response_processing_error'
                    }
                }), 500

        response_data = {
            'id': 'chatcmpl-xxxxxxxxxxxx',  
            'object': 'chat.completion',
            'created': int(datetime.now().timestamp()),
            'model': model,
            'choices': [{
                'index': 0,
                'message': {
                    'role': 'assistant',
                    'content': text_content
                },
                'finish_reason': 'stop'
            }],
            'usage': {
                'prompt_tokens': 0,
                'completion_tokens': 0,
                'total_tokens': 0
            }
        }
        logger.info(f"200!")
        return jsonify(response_data)

@app.route('/hf/v1/models', methods=['GET'])
def list_models():
    response = {"object": "list", "data": GEMINI_MODELS}
    return jsonify(response)

def keep_alive():
    try:
        response = requests.get("http://127.0.0.1:7860/", timeout=10)
        response.raise_for_status()  
        print(f"Keep alive ping successful: {response.status_code} at {time.ctime()}")
    except requests.exceptions.RequestException as e:
        print(f"Keep alive ping failed: {e} at {time.ctime()}")

if __name__ == '__main__':
    scheduler = BackgroundScheduler()

    scheduler.add_job(keep_alive, 'interval', hours=12)
    scheduler.start()
    logger.info(f"Reminiproxy v2.3.0 启动")
    logger.info(f"最大尝试次数/MaxRetries: {MAX_RETRIES}")
    logger.info(f"最大请求次数/MaxRequests: {MAX_REQUESTS}")
    logger.info(f"请求限额窗口/LimitWindow: {LIMIT_WINDOW} 秒")

    app.run(debug=True, host='0.0.0.0', port=int(os.environ.get('PORT', 7860)))