mirror of
https://github.com/snailyp/gemini-balance.git
synced 2026-05-21 08:09:41 +08:00
9a7a1d7c2f
- 为 Gemini 聊天（流式/非流式）、OpenAI 图像聊天（流式/非流式）和 embedding 服务的 API 调用实现全面的数据库日志记录。日志包括请求详情、成功/失败状态、状态码、延迟和错误消息。 - 重构 Gemini 流式聊天服务 (`stream_generate_content`) 以整合使用 `KeyManager` 的重试逻辑，与非流式实现保持一致，包括失败时的 API 密钥切换。 - 增强重试处理器 (`RetryHandler`) 的日志记录，以提高密钥切换和失败场景下的清晰度。 - 确保 `api_key` 正确传递给 OpenAI 图像聊天完成。 - 改进 embedding 服务中的错误处理，区分 `APIStatusError` 和通用异常，并将错误记录到数据库。 - 为 embedding 服务日志添加请求负载截断。 - 修复 Gemini `_build_payload` 中使用正确的 `model` 变量获取 `THINKING_BUDGET_MAP` 的错误。 - 移除 `ImageCreateService` 中未使用的 `paid_key` 类变量。
83 lines
3.4 KiB
Python
83 lines
3.4 KiB
Python
import datetime
|
|
import time
|
|
import re # For potential status code parsing from generic errors
|
|
from typing import List, Union
|
|
|
|
import openai
|
|
from openai import APIStatusError # Import specific error type
|
|
from openai.types import CreateEmbeddingResponse
|
|
|
|
from app.config.config import settings
|
|
from app.log.logger import get_embeddings_logger
|
|
from app.database.services import add_error_log, add_request_log # Import DB logging functions
|
|
|
|
logger = get_embeddings_logger()
|
|
|
|
|
|
class EmbeddingService:
|
|
|
|
async def create_embedding(
|
|
self, input_text: Union[str, List[str]], model: str, api_key: str
|
|
) -> CreateEmbeddingResponse:
|
|
"""Create embeddings using OpenAI API with database logging"""
|
|
start_time = time.perf_counter()
|
|
request_datetime = datetime.datetime.now()
|
|
is_success = False
|
|
status_code = None
|
|
response = None
|
|
error_log_msg = ""
|
|
# Prepare request message for logging (truncate if list or long string)
|
|
if isinstance(input_text, list):
|
|
request_msg_log = {"input_truncated": [str(item)[:100] + "..." if len(str(item)) > 100 else str(item) for item in input_text[:5]]}
|
|
if len(input_text) > 5:
|
|
request_msg_log["input_truncated"].append("...")
|
|
else:
|
|
request_msg_log = {"input_truncated": input_text[:1000] + "..." if len(input_text) > 1000 else input_text}
|
|
|
|
|
|
try:
|
|
client = openai.OpenAI(api_key=api_key, base_url=settings.BASE_URL)
|
|
response = client.embeddings.create(input=input_text, model=model)
|
|
is_success = True
|
|
status_code = 200 # Assume 200 OK on success
|
|
return response
|
|
except APIStatusError as e:
|
|
is_success = False
|
|
status_code = e.status_code
|
|
error_log_msg = f"OpenAI API error: {e}"
|
|
logger.error(f"Error creating embedding (APIStatusError): {error_log_msg}")
|
|
raise e # Re-raise the specific error
|
|
except Exception as e:
|
|
is_success = False
|
|
error_log_msg = f"Generic error: {e}"
|
|
logger.error(f"Error creating embedding (Exception): {error_log_msg}")
|
|
# Try to parse status code from generic error (less reliable)
|
|
match = re.search(r"status code (\d+)", str(e))
|
|
if match:
|
|
status_code = int(match.group(1))
|
|
else:
|
|
status_code = 500 # Default if parsing fails
|
|
raise e # Re-raise the generic error
|
|
finally:
|
|
end_time = time.perf_counter()
|
|
latency_ms = int((end_time - start_time) * 1000)
|
|
if not is_success:
|
|
# Log error to database if it failed
|
|
await add_error_log(
|
|
gemini_key=api_key, # Using gemini_key parameter name for consistency
|
|
model_name=model,
|
|
error_type="openai-embedding",
|
|
error_log=error_log_msg,
|
|
error_code=status_code,
|
|
request_msg=request_msg_log
|
|
)
|
|
# Log request outcome to database regardless of success/failure
|
|
await add_request_log(
|
|
model_name=model,
|
|
api_key=api_key,
|
|
is_success=is_success,
|
|
status_code=status_code,
|
|
latency_ms=latency_ms,
|
|
request_time=request_datetime
|
|
)
|