微信小程序页面迁移校验之前 P5任务处理之前
This commit is contained in:
273
apps/backend/app/ai/bailian_client.py
Normal file
273
apps/backend/app/ai/bailian_client.py
Normal file
@@ -0,0 +1,273 @@
|
||||
"""百炼 API 统一封装层。
|
||||
|
||||
使用 openai Python SDK（百炼兼容 OpenAI 协议），提供流式和非流式两种调用模式。
|
||||
所有 AI 应用通过此客户端统一调用阿里云通义千问。
|
||||
"""
|
||||
|
||||
from __future__ import annotations
|
||||
|
||||
import asyncio
|
||||
import copy
|
||||
import json
|
||||
import logging
|
||||
from datetime import datetime
|
||||
from typing import Any, AsyncGenerator
|
||||
|
||||
import openai
|
||||
|
||||
logger = logging.getLogger(__name__)
|
||||
|
||||
|
||||
# ── 异常类 ──────────────────────────────────────────────────────────
|
||||
|
||||
|
||||
class BailianApiError(Exception):
|
||||
"""百炼 API 调用失败（重试耗尽后）。"""
|
||||
|
||||
def __init__(self, message: str, status_code: int | None = None):
|
||||
super().__init__(message)
|
||||
self.status_code = status_code
|
||||
|
||||
|
||||
class BailianJsonParseError(Exception):
|
||||
"""百炼 API 返回的 JSON 解析失败。"""
|
||||
|
||||
def __init__(self, message: str, raw_content: str = ""):
|
||||
super().__init__(message)
|
||||
self.raw_content = raw_content
|
||||
|
||||
|
||||
class BailianAuthError(BailianApiError):
|
||||
"""百炼 API Key 无效（HTTP 401）。"""
|
||||
|
||||
def __init__(self, message: str = "API Key 无效或已过期"):
|
||||
super().__init__(message, status_code=401)
|
||||
|
||||
|
||||
# ── 客户端 ──────────────────────────────────────────────────────────
|
||||
|
||||
|
||||
class BailianClient:
|
||||
"""百炼 API 统一封装层。
|
||||
|
||||
使用 openai.AsyncOpenAI 客户端，base_url 指向百炼端点。
|
||||
提供流式（chat_stream）和非流式（chat_json）两种调用模式。
|
||||
"""
|
||||
|
||||
# 重试配置
|
||||
MAX_RETRIES = 3
|
||||
BASE_INTERVAL = 1 # 秒
|
||||
|
||||
def __init__(self, api_key: str, base_url: str, model: str):
|
||||
"""初始化百炼客户端。
|
||||
|
||||
Args:
|
||||
api_key: 百炼 API Key（环境变量 BAILIAN_API_KEY）
|
||||
base_url: 百炼 API 端点（环境变量 BAILIAN_BASE_URL）
|
||||
model: 模型标识，如 qwen-plus（环境变量 BAILIAN_MODEL）
|
||||
"""
|
||||
self.model = model
|
||||
self._client = openai.AsyncOpenAI(
|
||||
api_key=api_key,
|
||||
base_url=base_url,
|
||||
)
|
||||
|
||||
async def chat_stream(
|
||||
self,
|
||||
messages: list[dict],
|
||||
*,
|
||||
temperature: float = 0.7,
|
||||
max_tokens: int = 2000,
|
||||
) -> AsyncGenerator[str, None]:
|
||||
"""流式调用，逐 chunk yield 文本。用于应用 1 SSE。
|
||||
|
||||
Args:
|
||||
messages: 消息列表
|
||||
temperature: 温度参数，默认 0.7
|
||||
max_tokens: 最大 token 数，默认 2000
|
||||
|
||||
Yields:
|
||||
文本 chunk
|
||||
"""
|
||||
messages = self._inject_current_time(messages)
|
||||
response = await self._call_with_retry(
|
||||
model=self.model,
|
||||
messages=messages,
|
||||
temperature=temperature,
|
||||
max_tokens=max_tokens,
|
||||
stream=True,
|
||||
)
|
||||
async for chunk in response:
|
||||
if chunk.choices and chunk.choices[0].delta.content:
|
||||
yield chunk.choices[0].delta.content
|
||||
|
||||
async def chat_json(
|
||||
self,
|
||||
messages: list[dict],
|
||||
*,
|
||||
temperature: float = 0.3,
|
||||
max_tokens: int = 4000,
|
||||
) -> tuple[dict, int]:
|
||||
"""非流式调用，返回解析后的 JSON dict 和 tokens_used。
|
||||
|
||||
用于应用 2-8 的结构化输出。使用 response_format={"type": "json_object"}
|
||||
确保返回合法 JSON。
|
||||
|
||||
Args:
|
||||
messages: 消息列表
|
||||
temperature: 温度参数，默认 0.3（结构化输出用低温度）
|
||||
max_tokens: 最大 token 数，默认 4000
|
||||
|
||||
Returns:
|
||||
(parsed_json_dict, tokens_used) 元组
|
||||
|
||||
Raises:
|
||||
BailianJsonParseError: 响应内容无法解析为 JSON
|
||||
BailianApiError: API 调用失败（重试耗尽后）
|
||||
"""
|
||||
messages = self._inject_current_time(messages)
|
||||
response = await self._call_with_retry(
|
||||
model=self.model,
|
||||
messages=messages,
|
||||
temperature=temperature,
|
||||
max_tokens=max_tokens,
|
||||
stream=False,
|
||||
response_format={"type": "json_object"},
|
||||
)
|
||||
raw_content = response.choices[0].message.content or ""
|
||||
tokens_used = response.usage.total_tokens if response.usage else 0
|
||||
|
||||
try:
|
||||
parsed = json.loads(raw_content)
|
||||
except (json.JSONDecodeError, TypeError) as e:
|
||||
logger.error("百炼 API 返回非法 JSON: %s", raw_content[:500])
|
||||
raise BailianJsonParseError(
|
||||
f"JSON 解析失败: {e}",
|
||||
raw_content=raw_content,
|
||||
) from e
|
||||
|
||||
return parsed, tokens_used
|
||||
|
||||
def _inject_current_time(self, messages: list[dict]) -> list[dict]:
|
||||
"""纯函数：在首条消息的 content（JSON 字符串）中注入 current_time 字段。
|
||||
|
||||
- 深拷贝输入，不修改原始 messages
|
||||
- 首条消息 content 尝试解析为 JSON，注入 current_time
|
||||
- 如果首条消息 content 不是 JSON，则包装为 JSON
|
||||
- 其余消息不变
|
||||
- current_time 格式：ISO 8601 精确到秒，如 2026-03-08T14:30:00
|
||||
|
||||
Args:
|
||||
messages: 原始消息列表
|
||||
|
||||
Returns:
|
||||
注入 current_time 后的新消息列表
|
||||
"""
|
||||
if not messages:
|
||||
return []
|
||||
|
||||
result = copy.deepcopy(messages)
|
||||
first = result[0]
|
||||
content = first.get("content", "")
|
||||
now_str = datetime.now().strftime("%Y-%m-%dT%H:%M:%S")
|
||||
|
||||
try:
|
||||
parsed = json.loads(content)
|
||||
if isinstance(parsed, dict):
|
||||
parsed["current_time"] = now_str
|
||||
else:
|
||||
# content 是合法 JSON 但不是 dict（如数组、字符串），包装为 dict
|
||||
parsed = {"original_content": parsed, "current_time": now_str}
|
||||
except (json.JSONDecodeError, TypeError):
|
||||
# content 不是 JSON，包装为 dict
|
||||
parsed = {"content": content, "current_time": now_str}
|
||||
|
||||
first["content"] = json.dumps(parsed, ensure_ascii=False)
|
||||
return result
|
||||
|
||||
async def _call_with_retry(self, **kwargs: Any) -> Any:
|
||||
"""带指数退避的重试封装。
|
||||
|
||||
重试策略：
|
||||
- 最多重试 MAX_RETRIES 次（默认 3 次）
|
||||
- 间隔：BASE_INTERVAL × 2^(n-1)，即 1s → 2s → 4s
|
||||
- HTTP 4xx：不重试，直接抛出（401 → BailianAuthError）
|
||||
- HTTP 5xx / 超时：重试
|
||||
|
||||
Args:
|
||||
**kwargs: 传递给 openai client 的参数
|
||||
|
||||
Returns:
|
||||
API 响应对象
|
||||
|
||||
Raises:
|
||||
BailianAuthError: API Key 无效（HTTP 401）
|
||||
BailianApiError: API 调用失败（重试耗尽后）
|
||||
"""
|
||||
is_stream = kwargs.get("stream", False)
|
||||
last_error: Exception | None = None
|
||||
|
||||
for attempt in range(self.MAX_RETRIES):
|
||||
try:
|
||||
if is_stream:
|
||||
# 流式调用：返回 async iterator
|
||||
return await self._client.chat.completions.create(**kwargs)
|
||||
else:
|
||||
return await self._client.chat.completions.create(**kwargs)
|
||||
|
||||
except openai.AuthenticationError as e:
|
||||
# 401：API Key 无效，不重试
|
||||
logger.error("百炼 API 认证失败: %s", e)
|
||||
raise BailianAuthError(str(e)) from e
|
||||
|
||||
except openai.BadRequestError as e:
|
||||
# 400：请求参数错误，不重试
|
||||
logger.error("百炼 API 请求参数错误: %s", e)
|
||||
raise BailianApiError(str(e), status_code=400) from e
|
||||
|
||||
except openai.RateLimitError as e:
|
||||
# 429：限流，不重试（属于 4xx）
|
||||
logger.error("百炼 API 限流: %s", e)
|
||||
raise BailianApiError(str(e), status_code=429) from e
|
||||
|
||||
except openai.PermissionDeniedError as e:
|
||||
# 403：权限不足，不重试
|
||||
logger.error("百炼 API 权限不足: %s", e)
|
||||
raise BailianApiError(str(e), status_code=403) from e
|
||||
|
||||
except openai.NotFoundError as e:
|
||||
# 404：资源不存在，不重试
|
||||
logger.error("百炼 API 资源不存在: %s", e)
|
||||
raise BailianApiError(str(e), status_code=404) from e
|
||||
|
||||
except openai.UnprocessableEntityError as e:
|
||||
# 422：不可处理，不重试
|
||||
logger.error("百炼 API 不可处理的请求: %s", e)
|
||||
raise BailianApiError(str(e), status_code=422) from e
|
||||
|
||||
except (openai.InternalServerError, openai.APIConnectionError, openai.APITimeoutError) as e:
|
||||
# 5xx / 超时 / 连接错误：重试
|
||||
last_error = e
|
||||
if attempt < self.MAX_RETRIES - 1:
|
||||
wait_time = self.BASE_INTERVAL * (2 ** attempt)
|
||||
logger.warning(
|
||||
"百炼 API 调用失败（第 %d/%d 次），%ds 后重试: %s",
|
||||
attempt + 1,
|
||||
self.MAX_RETRIES,
|
||||
wait_time,
|
||||
e,
|
||||
)
|
||||
await asyncio.sleep(wait_time)
|
||||
else:
|
||||
logger.error(
|
||||
"百炼 API 调用失败，已达最大重试次数 %d: %s",
|
||||
self.MAX_RETRIES,
|
||||
e,
|
||||
)
|
||||
|
||||
# 重试耗尽
|
||||
status_code = getattr(last_error, "status_code", None)
|
||||
raise BailianApiError(
|
||||
f"百炼 API 调用失败（重试 {self.MAX_RETRIES} 次后）: {last_error}",
|
||||
status_code=status_code,
|
||||
) from last_error
|
||||
Reference in New Issue
Block a user