uni-api / main.py
yym68686's picture
✨ Feature: Add feature: Add support for using angle brackets to enclose strings to set the string before the slash to the channel name.
6038b37
raw
history blame
22.5 kB
from log_config import logger
import re
import httpx
import secrets
import time as time_module
from contextlib import asynccontextmanager
from fastapi.middleware.cors import CORSMiddleware
from fastapi import FastAPI, HTTPException, Depends, Request
from fastapi.responses import StreamingResponse, JSONResponse
from fastapi.security import HTTPBearer, HTTPAuthorizationCredentials
from models import RequestModel, ImageGenerationRequest
from request import get_payload
from response import fetch_response, fetch_response_stream
from utils import error_handling_wrapper, post_all_models, load_config, safe_get, circular_list_encoder
from collections import defaultdict
from typing import List, Dict, Union
from urllib.parse import urlparse
@asynccontextmanager
async def lifespan(app: FastAPI):
# 启动时的代码
import os
TIMEOUT = float(os.getenv("TIMEOUT", 20))
timeout = httpx.Timeout(connect=15.0, read=TIMEOUT, write=30.0, pool=30.0)
default_headers = {
"User-Agent": "curl/7.68.0", # 模拟 curl 的 User-Agent
"Accept": "*/*", # curl 的默认 Accept 头
}
app.state.client = httpx.AsyncClient(
timeout=timeout,
headers=default_headers,
http2=True, # 禁用 HTTP/2
verify=True, # 保持 SSL 验证(如需禁用,设为 False,但不建议)
follow_redirects=True, # 自动跟随重定向
)
# app.state.client = httpx.AsyncClient(timeout=timeout)
app.state.config, app.state.api_keys_db, app.state.api_list = await load_config(app)
yield
# 关闭时的代码
await app.state.client.aclose()
app = FastAPI(lifespan=lifespan)
import asyncio
from time import time
from collections import defaultdict
from starlette.middleware.base import BaseHTTPMiddleware
from datetime import datetime
from datetime import timedelta
import json
import aiofiles
class StatsMiddleware(BaseHTTPMiddleware):
def __init__(self, app, exclude_paths=None, save_interval=3600, filename="stats.json"):
super().__init__(app)
self.request_counts = defaultdict(int)
self.request_times = defaultdict(float)
self.ip_counts = defaultdict(lambda: defaultdict(int))
self.request_arrivals = defaultdict(list)
self.channel_success_counts = defaultdict(int)
self.channel_failure_counts = defaultdict(int)
self.lock = asyncio.Lock()
self.exclude_paths = set(exclude_paths or [])
self.save_interval = save_interval
self.filename = filename
self.last_save_time = time()
# 启动定期保存和清理任务
asyncio.create_task(self.periodic_save_and_cleanup())
async def dispatch(self, request: Request, call_next):
arrival_time = datetime.now()
start_time = time()
response = await call_next(request)
process_time = time() - start_time
endpoint = f"{request.method} {request.url.path}"
client_ip = request.client.host
if request.url.path not in self.exclude_paths:
async with self.lock:
self.request_counts[endpoint] += 1
self.request_times[endpoint] += process_time
self.ip_counts[endpoint][client_ip] += 1
self.request_arrivals[endpoint].append(arrival_time)
return response
async def periodic_save_and_cleanup(self):
while True:
await asyncio.sleep(self.save_interval)
await self.save_stats()
await self.cleanup_old_data()
async def save_stats(self):
current_time = time()
if current_time - self.last_save_time < self.save_interval:
return
async with self.lock:
stats = {
"request_counts": dict(self.request_counts),
"request_times": dict(self.request_times),
"ip_counts": {k: dict(v) for k, v in self.ip_counts.items()},
"request_arrivals": {k: [t.isoformat() for t in v] for k, v in self.request_arrivals.items()},
"channel_success_counts": dict(self.channel_success_counts),
"channel_failure_counts": dict(self.channel_failure_counts),
"channel_success_percentages": self.calculate_success_percentages(),
"channel_failure_percentages": self.calculate_failure_percentages()
}
filename = self.filename
async with aiofiles.open(filename, mode='w') as f:
await f.write(json.dumps(stats, indent=2))
self.last_save_time = current_time
def calculate_success_percentages(self):
percentages = {}
for channel, success_count in self.channel_success_counts.items():
total_count = success_count + self.channel_failure_counts[channel]
if total_count > 0:
percentages[channel] = success_count / total_count * 100
else:
percentages[channel] = 0
sorted_percentages = dict(sorted(percentages.items(), key=lambda item: item[1], reverse=True))
return sorted_percentages
def calculate_failure_percentages(self):
percentages = {}
for channel, failure_count in self.channel_failure_counts.items():
total_count = failure_count + self.channel_success_counts[channel]
if total_count > 0:
percentages[channel] = failure_count / total_count * 100
else:
percentages[channel] = 0
sorted_percentages = dict(sorted(percentages.items(), key=lambda item: item[1], reverse=True))
return sorted_percentages
async def cleanup_old_data(self):
cutoff_time = datetime.now() - timedelta(hours=24)
async with self.lock:
for endpoint in list(self.request_arrivals.keys()):
self.request_arrivals[endpoint] = [
t for t in self.request_arrivals[endpoint] if t > cutoff_time
]
if not self.request_arrivals[endpoint]:
del self.request_arrivals[endpoint]
self.request_counts.pop(endpoint, None)
self.request_times.pop(endpoint, None)
self.ip_counts.pop(endpoint, None)
async def cleanup(self):
await self.save_stats()
# 配置 CORS 中间件
app.add_middleware(
CORSMiddleware,
allow_origins=["*"], # 允许所有来源
allow_credentials=True,
allow_methods=["*"], # 允许所有 HTTP 方法
allow_headers=["*"], # 允许所有头部字段
)
app.add_middleware(StatsMiddleware, exclude_paths=["/stats", "/generate-api-key"])
# 在 process_request 函数中更新成功和失败计数
async def process_request(request: Union[RequestModel, ImageGenerationRequest], provider: Dict, endpoint=None):
url = provider['base_url']
parsed_url = urlparse(url)
# print("parsed_url", parsed_url)
engine = None
if parsed_url.netloc == 'generativelanguage.googleapis.com':
engine = "gemini"
elif parsed_url.netloc == 'aiplatform.googleapis.com':
engine = "vertex"
elif parsed_url.netloc == 'api.cloudflare.com':
engine = "cloudflare"
elif parsed_url.netloc == 'api.anthropic.com' or parsed_url.path.endswith("v1/messages"):
engine = "claude"
elif parsed_url.netloc == 'openrouter.ai':
engine = "openrouter"
else:
engine = "gpt"
if "claude" not in provider['model'][request.model] \
and "gpt" not in provider['model'][request.model] \
and "gemini" not in provider['model'][request.model] \
and parsed_url.netloc != 'api.cloudflare.com':
engine = "openrouter"
if "claude" in provider['model'][request.model] and engine == "vertex":
engine = "vertex-claude"
if "gemini" in provider['model'][request.model] and engine == "vertex":
engine = "vertex-gemini"
if endpoint == "/v1/images/generations":
engine = "dalle"
request.stream = False
if provider.get("engine"):
engine = provider["engine"]
logger.info(f"provider: {provider['provider']:<10} model: {request.model:<10} engine: {engine}")
url, headers, payload = await get_payload(request, engine, provider)
# logger.info(json.dumps(headers, indent=4, ensure_ascii=False))
# logger.info(json.dumps(payload, indent=4, ensure_ascii=False))
try:
if request.stream:
model = provider['model'][request.model]
generator = fetch_response_stream(app.state.client, url, headers, payload, engine, model)
wrapped_generator = await error_handling_wrapper(generator, status_code=500)
response = StreamingResponse(wrapped_generator, media_type="text/event-stream")
else:
response = await anext(fetch_response(app.state.client, url, headers, payload))
# 更新成功计数
async with app.middleware_stack.app.lock:
app.middleware_stack.app.channel_success_counts[provider['provider']] += 1
return response
except (Exception, HTTPException, asyncio.CancelledError, httpx.ReadError) as e:
# 更新失败计数
async with app.middleware_stack.app.lock:
app.middleware_stack.app.channel_failure_counts[provider['provider']] += 1
raise e
def weighted_round_robin(weights):
provider_names = list(weights.keys())
current_weights = {name: 0 for name in provider_names}
num_selections = total_weight = sum(weights.values())
weighted_provider_list = []
for _ in range(num_selections):
max_ratio = -1
selected_letter = None
for name in provider_names:
current_weights[name] += weights[name]
ratio = current_weights[name] / weights[name]
if ratio > max_ratio:
max_ratio = ratio
selected_letter = name
weighted_provider_list.append(selected_letter)
current_weights[selected_letter] -= total_weight
return weighted_provider_list
import asyncio
class ModelRequestHandler:
def __init__(self):
self.last_provider_index = -1
def get_matching_providers(self, model_name, token):
config = app.state.config
# api_keys_db = app.state.api_keys_db
api_list = app.state.api_list
api_index = api_list.index(token)
if not safe_get(config, 'api_keys', api_index, 'model'):
raise HTTPException(status_code=404, detail="No matching model found")
provider_rules = []
for model in config['api_keys'][api_index]['model']:
if "/" in model:
if model.startswith("<") and model.endswith(">"):
model = model[1:-1]
# 处理带斜杠的模型名
for provider in config['providers']:
if model in provider['model'].keys():
provider_rules.append(provider['provider'] + "/" + model)
else:
provider_name = model.split("/")[0]
model_name_split = "/".join(model.split("/")[1:])
models_list = []
for provider in config['providers']:
if provider['provider'] == provider_name:
models_list.extend(list(provider['model'].keys()))
# print("models_list", models_list)
# print("model_name", model_name)
# print("model", model)
if (model_name_split and model_name in models_list) or (model_name_split == "*" and model_name in models_list):
provider_rules.append(provider_name)
else:
for provider in config['providers']:
if model in provider['model'].keys():
provider_rules.append(provider['provider'] + "/" + model)
provider_list = []
# print("provider_rules", provider_rules)
for item in provider_rules:
for provider in config['providers']:
# print("provider", provider, provider['provider'] == item, item)
if "/" in item:
if provider['provider'] == item.split("/")[0]:
if model_name in provider['model'].keys() and "/".join(item.split("/")[1:]) == model_name:
provider_list.append(provider)
elif provider['provider'] == item:
if model_name in provider['model'].keys():
provider_list.append(provider)
else:
pass
# if provider['provider'] == item:
# if "/" in item:
# if item.split("/")[1] == model_name:
# provider_list.append(provider)
# else:
# if model_name in provider['model'].keys():
# provider_list.append(provider)
# import json
# for provider in provider_list:
# print(json.dumps(provider, indent=4, ensure_ascii=False, default=circular_list_encoder))
return provider_list
async def request_model(self, request: Union[RequestModel, ImageGenerationRequest], token: str, endpoint=None):
config = app.state.config
# api_keys_db = app.state.api_keys_db
api_list = app.state.api_list
model_name = request.model
matching_providers = self.get_matching_providers(model_name, token)
# import json
# print("matching_providers", json.dumps(matching_providers, indent=4, ensure_ascii=False))
if not matching_providers:
raise HTTPException(status_code=404, detail="No matching model found")
# 检查是否启用轮询
api_index = api_list.index(token)
weights = safe_get(config, 'api_keys', api_index, "weights")
if weights:
# 步骤 1: 提取 matching_providers 中的所有 provider 值
providers = set(provider['provider'] for provider in matching_providers)
weight_keys = set(weights.keys())
# 步骤 3: 计算交集
intersection = providers.intersection(weight_keys)
weights = dict(filter(lambda item: item[0] in intersection, weights.items()))
weighted_provider_name_list = weighted_round_robin(weights)
new_matching_providers = []
for provider_name in weighted_provider_name_list:
for provider in matching_providers:
if provider['provider'] == provider_name:
new_matching_providers.append(provider)
matching_providers = new_matching_providers
# import json
# print("matching_providers", json.dumps(matching_providers, indent=4, ensure_ascii=False, default=circular_list_encoder))
use_round_robin = True
auto_retry = True
if safe_get(config, 'api_keys', api_index, "preferences", "USE_ROUND_ROBIN") == False:
use_round_robin = False
if safe_get(config, 'api_keys', api_index, "preferences", "AUTO_RETRY") == False:
auto_retry = False
return await self.try_all_providers(request, matching_providers, use_round_robin, auto_retry, endpoint)
# 在 try_all_providers 函数中处理失败的情况
async def try_all_providers(self, request: Union[RequestModel, ImageGenerationRequest], providers: List[Dict], use_round_robin: bool, auto_retry: bool, endpoint: str = None):
num_providers = len(providers)
start_index = self.last_provider_index + 1 if use_round_robin else 0
for i in range(num_providers + 1):
self.last_provider_index = (start_index + i) % num_providers
provider = providers[self.last_provider_index]
try:
response = await process_request(request, provider, endpoint)
return response
except (Exception, HTTPException, asyncio.CancelledError, httpx.ReadError) as e:
logger.error(f"Error with provider {provider['provider']}: {str(e)}")
if auto_retry:
continue
else:
raise HTTPException(status_code=500, detail="Error: Current provider response failed!")
raise HTTPException(status_code=500, detail=f"All providers failed: {request.model}")
model_handler = ModelRequestHandler()
def parse_rate_limit(limit_string):
# 定义时间单位到秒的映射
time_units = {
's': 1, 'sec': 1, 'second': 1,
'm': 60, 'min': 60, 'minute': 60,
'h': 3600, 'hr': 3600, 'hour': 3600,
'd': 86400, 'day': 86400,
'mo': 2592000, 'month': 2592000,
'y': 31536000, 'year': 31536000
}
# 使用正则表达式匹配数字和单位
match = re.match(r'^(\d+)/(\w+)$', limit_string)
if not match:
raise ValueError(f"Invalid rate limit format: {limit_string}")
count, unit = match.groups()
count = int(count)
# 转换单位到秒
if unit not in time_units:
raise ValueError(f"Unknown time unit: {unit}")
seconds = time_units[unit]
return (count, seconds)
class InMemoryRateLimiter:
def __init__(self):
self.requests = defaultdict(list)
async def is_rate_limited(self, key: str, limit: int, period: int) -> bool:
now = time_module.time()
self.requests[key] = [req for req in self.requests[key] if req > now - period]
if len(self.requests[key]) >= limit:
return True
self.requests[key].append(now)
return False
rate_limiter = InMemoryRateLimiter()
async def get_user_rate_limit(api_index: str = None):
# 这里应该实现根据 token 获取用户速率限制的逻辑
# 示例: 返回 (次数, 秒数)
config = app.state.config
raw_rate_limit = safe_get(config, 'api_keys', api_index, "preferences", "RATE_LIMIT")
if not api_index or not raw_rate_limit:
return (60, 60)
rate_limit = parse_rate_limit(raw_rate_limit)
return rate_limit
security = HTTPBearer()
async def rate_limit_dependency(request: Request, credentials: HTTPAuthorizationCredentials = Depends(security)):
token = credentials.credentials if credentials else None
api_list = app.state.api_list
try:
api_index = api_list.index(token)
except ValueError:
print("error: Invalid or missing API Key:", token)
api_index = None
token = None
limit, period = await get_user_rate_limit(api_index)
# 使用 IP 地址和 token(如果有)作为限制键
client_ip = request.client.host
rate_limit_key = f"{client_ip}:{token}" if token else client_ip
if await rate_limiter.is_rate_limited(rate_limit_key, limit, period):
raise HTTPException(status_code=429, detail="Too many requests")
def verify_api_key(credentials: HTTPAuthorizationCredentials = Depends(security)):
api_list = app.state.api_list
token = credentials.credentials
if token not in api_list:
raise HTTPException(status_code=403, detail="Invalid or missing API Key")
return token
def verify_admin_api_key(credentials: HTTPAuthorizationCredentials = Depends(security)):
api_list = app.state.api_list
token = credentials.credentials
if token not in api_list:
raise HTTPException(status_code=403, detail="Invalid or missing API Key")
for api_key in app.state.api_keys_db:
if api_key['api'] == token:
if api_key.get('role') != "admin":
raise HTTPException(status_code=403, detail="Permission denied")
return token
@app.post("/v1/chat/completions", dependencies=[Depends(rate_limit_dependency)])
async def request_model(request: Union[RequestModel, ImageGenerationRequest], token: str = Depends(verify_api_key)):
return await model_handler.request_model(request, token)
@app.options("/v1/chat/completions", dependencies=[Depends(rate_limit_dependency)])
async def options_handler():
return JSONResponse(status_code=200, content={"detail": "OPTIONS allowed"})
@app.get("/v1/models", dependencies=[Depends(rate_limit_dependency)])
async def list_models(token: str = Depends(verify_api_key)):
models = post_all_models(token, app.state.config, app.state.api_list)
return JSONResponse(content={
"object": "list",
"data": models
})
@app.post("/v1/images/generations", dependencies=[Depends(rate_limit_dependency)])
async def images_generations(
request: ImageGenerationRequest,
token: str = Depends(verify_api_key)
):
return await model_handler.request_model(request, token, endpoint="/v1/images/generations")
@app.get("/generate-api-key", dependencies=[Depends(rate_limit_dependency)])
def generate_api_key():
api_key = "sk-" + secrets.token_urlsafe(36)
return JSONResponse(content={"api_key": api_key})
# 在 /stats 路由中返回成功和失败百分比
@app.get("/stats", dependencies=[Depends(rate_limit_dependency)])
async def get_stats(request: Request, token: str = Depends(verify_admin_api_key)):
middleware = app.middleware_stack.app
if isinstance(middleware, StatsMiddleware):
async with middleware.lock:
stats = {
"request_counts": dict(middleware.request_counts),
"request_times": dict(middleware.request_times),
"ip_counts": {k: dict(v) for k, v in middleware.ip_counts.items()},
"request_arrivals": {k: [t.isoformat() for t in v] for k, v in middleware.request_arrivals.items()},
"channel_success_counts": dict(middleware.channel_success_counts),
"channel_failure_counts": dict(middleware.channel_failure_counts),
"channel_success_percentages": middleware.calculate_success_percentages(),
"channel_failure_percentages": middleware.calculate_failure_percentages()
}
return JSONResponse(content=stats)
return {"error": "StatsMiddleware not found"}
# async def on_fetch(request, env):
# import asgi
# return await asgi.fetch(app, request, env)
if __name__ == '__main__':
import uvicorn
uvicorn.run(
"__main__:app",
host="0.0.0.0",
port=8000,
reload=True,
ws="none",
# log_level="warning"
)