sycommon-python-lib 0.2.5a7__tar.gz → 0.2.5a9__tar.gz

{sycommon_python_lib-0.2.5a7 → sycommon_python_lib-0.2.5a9}/PKG-INFO

@@ -1,15 +1,15 @@
 Metadata-Version: 2.4
 Name: sycommon-python-lib
-Version: 0.2.5a7
+Version: 0.2.5a9
 Summary: Add your description here
 Requires-Python: >=3.11
 Description-Content-Type: text/markdown
 Requires-Dist: aio-pika==9.6.2
 Requires-Dist: aiohttp==3.13.5
 Requires-Dist: aiomysql==0.3.2
-Requires-Dist: anyio==4.12.1
+Requires-Dist: anyio==4.13.0
 Requires-Dist: decorator==5.3.1
-Requires-Dist: deepagents==0.6.1
+Requires-Dist: deepagents==0.6.3
 Requires-Dist: elasticsearch==9.4.0
 Requires-Dist: fastapi==0.136.1
 Requires-Dist: jinja2==3.1.6
@@ -18,29 +18,30 @@ Requires-Dist: langchain==1.3.1
 Requires-Dist: langchain-core==1.4.0
 Requires-Dist: langchain-openai==1.2.1
 Requires-Dist: langfuse==4.6.1
-Requires-Dist: langgraph==1.2.0
+Requires-Dist: langgraph==1.2.1
 Requires-Dist: langgraph-checkpoint-postgres==3.1.0
 Requires-Dist: langgraph-checkpoint-redis==0.4.1
 Requires-Dist: ldap3==2.9.1
 Requires-Dist: loguru==0.7.3
-Requires-Dist: mysql-connector-python==9.6.0
+Requires-Dist: mysql-connector-python==9.7.0
 Requires-Dist: nacos-sdk-python<3.0,==2.0.11
 Requires-Dist: psutil==7.2.2
-Requires-Dist: pyxxl==0.4.6
+Requires-Dist: pyxxl==0.4.7
 Requires-Dist: pydantic==2.13.4
 Requires-Dist: python-dotenv==1.2.2
 Requires-Dist: python-multipart==0.0.29
 Requires-Dist: pyyaml==6.0.3
-Requires-Dist: redis==7.3.0
+Requires-Dist: redis==7.4.0
 Requires-Dist: sentry-sdk[fastapi]==2.60.0
-Requires-Dist: sqlalchemy[asyncio]==2.0.48
-Requires-Dist: starlette[full]==1.0.0
+Requires-Dist: sqlalchemy[asyncio]==2.0.49
+Requires-Dist: starlette[full]==1.0.1
 Requires-Dist: tiktoken==0.13.0
 Requires-Dist: uvicorn==0.47.0
 Requires-Dist: wecom-aibot-python-sdk==1.0.2
 Requires-Dist: twine==6.2.0
 Requires-Dist: minio==7.2.20
 Requires-Dist: langchain-mcp-adapters==0.2.2
+Requires-Dist: fastapi-mcp==0.4.0
 Requires-Dist: psycopg[binary,pool]==3.3.4
 
 # sycommon-python-lib

{sycommon_python_lib-0.2.5a7 → sycommon_python_lib-0.2.5a9}/pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "sycommon-python-lib"
-version = "0.2.5a7"
+version = "0.2.5a9"
 description = "Add your description here"
 readme = "README.md"
 requires-python = ">=3.11"
@@ -8,9 +8,9 @@ dependencies = [
     "aio-pika==9.6.2",
     "aiohttp==3.13.5",
     "aiomysql==0.3.2",
-    "anyio==4.12.1",
+    "anyio==4.13.0",
     "decorator==5.3.1",
-    "deepagents==0.6.1",
+    "deepagents==0.6.3",
     "elasticsearch==9.4.0",
     "fastapi==0.136.1",
     "jinja2==3.1.6",
@@ -19,35 +19,36 @@ dependencies = [
     "langchain-core==1.4.0",
     "langchain-openai==1.2.1",
     "langfuse==4.6.1",
-    "langgraph==1.2.0",
+    "langgraph==1.2.1",
     "langgraph-checkpoint-postgres==3.1.0",
     "langgraph-checkpoint-redis==0.4.1",
     "ldap3==2.9.1",
     "loguru==0.7.3",
-    "mysql-connector-python==9.6.0",
+    "mysql-connector-python==9.7.0",
     "nacos-sdk-python==2.0.11,<3.0",
     "psutil==7.2.2",
-    "pyxxl==0.4.6",
+    "pyxxl==0.4.7",
     "pydantic==2.13.4",
     "python-dotenv==1.2.2",
     "python-multipart==0.0.29",
     "pyyaml==6.0.3",
-    "redis==7.3.0",
+    "redis==7.4.0",
     "sentry-sdk[fastapi]==2.60.0",
-    "sqlalchemy[asyncio]==2.0.48",
-    "starlette[full]==1.0.0",
+    "sqlalchemy[asyncio]==2.0.49",
+    "starlette[full]==1.0.1",
     "tiktoken==0.13.0",
     "uvicorn==0.47.0",
     "wecom-aibot-python-sdk==1.0.2",
     "twine==6.2.0",
     "minio==7.2.20",
     "langchain-mcp-adapters==0.2.2",
+    "fastapi-mcp==0.4.0",
     "psycopg[binary,pool]==3.3.4",
 ]
 
 [tool.setuptools]
 packages = {find = {where = ["src"], exclude = ["*.__pycache__", "*.__pycache__.*"]}}
-package-data = {command = ["templates/**/*", "templates/**/.*"]}
+package-data = {command = ["templates/**/*", "templates/**/.*"], sycommon = ["llm/tiktoken_cache/*"]}
 
 [build-system]
 requires = ["setuptools"]

{sycommon_python_lib-0.2.5a7 → sycommon_python_lib-0.2.5a9}/src/sycommon/agent/mcp/tool_loader.py

@@ -6,13 +6,16 @@
 
 import re
 import hashlib
-from typing import List, Optional, Callable, Awaitable
+from typing import List, Optional, Callable, Awaitable, TYPE_CHECKING
 
 from langchain_core.tools import BaseTool
 
 from sycommon.logging.kafka_log import SYLogger
 from sycommon.agent.mcp.models import MCPServerConfig
 
+if TYPE_CHECKING:
+    from langchain_mcp_adapters.interceptors import MCPToolCallRequest
+
 
 def sanitize_name(name: str) -> str:
     """将名称转为合法的标识符，中文等非ASCII字符做 transliterate"""
@@ -37,6 +40,8 @@ def wrap_tool_with_error_handler(
         on_tool_success: 异步回调 (server_config_id, tool_name) -> None
         on_tool_error: 异步回调 (server_config_id, tool_name, error_msg) -> None
     """
+    is_content_and_artifact = getattr(
+        tool, 'response_format', '') == 'content_and_artifact'
     original_coroutine = tool.coroutine
     tool_name = tool.name
 
@@ -52,7 +57,8 @@ def wrap_tool_with_error_handler(
         except Exception as e:
             err_type = type(e).__name__
             err_msg = str(e)[:500]
-            SYLogger.warning(f"[MCP] 工具 '{tool_name}' 调用失败 ({err_type}): {err_msg}")
+            SYLogger.warning(
+                f"[MCP] 工具 '{tool_name}' 调用失败 ({err_type}): {err_msg}")
 
             if on_tool_error:
                 try:
@@ -66,6 +72,8 @@ def wrap_tool_with_error_handler(
                 f"可能原因: MCP 服务器未启动、网络不可达或连接超时。\n"
                 f"请尝试不使用该工具继续完成任务，或联系管理员检查 MCP 服务 '{server_name}'。"
             )
+            if is_content_and_artifact:
+                return friendly_msg, None
             return [friendly_msg]
 
     tool.coroutine = _safe_coroutine
@@ -74,13 +82,10 @@ def wrap_tool_with_error_handler(
 
 def _make_user_id_interceptor(user_id: str):
     """创建 MCP 工具调用拦截器，注入 X-User-Id header"""
-    from langchain_mcp_adapters.interceptors import MCPToolCallRequest, MCPToolCallResult
-    from typing import Callable, Awaitable
-
     async def inject_user_id(
-        request: MCPToolCallRequest,
-        handler: Callable[[MCPToolCallRequest], Awaitable[MCPToolCallResult]],
-    ) -> MCPToolCallResult:
+        request: "MCPToolCallRequest",
+        handler: Callable[["MCPToolCallRequest"], Awaitable],
+    ):
         headers = dict(request.headers or {})
         headers["X-User-Id"] = user_id
         request = request.override(headers=headers)
@@ -94,7 +99,8 @@ async def load_mcp_tools(
     user_id: str = None,
     on_tool_success: Optional[Callable[[str, str], Awaitable[None]]] = None,
     on_tool_error: Optional[Callable[[str, str, str], Awaitable[None]]] = None,
-    on_batch_available: Optional[Callable[[str, list], Awaitable[None]]] = None,
+    on_batch_available: Optional[Callable[[
+        str, list], Awaitable[None]]] = None,
     on_server_failure: Optional[Callable[[str, str], Awaitable[None]]] = None,
 ) -> List[BaseTool]:
     """加载 MCP 工具列表
@@ -135,33 +141,32 @@ async def load_mcp_tools(
             client = MultiServerMCPClient(client_config)
 
             async with client.session(key) as session:
-                from langchain_mcp_adapters.tools import _list_all_tools
+                from langchain_mcp_adapters.tools import _list_all_tools, convert_mcp_tool_to_langchain_tool
                 raw_tools = await _list_all_tools(session)
 
-            original_names = []
-            for tool in raw_tools:
-                original_name = tool.name
-                original_names.append(original_name)
-
-                from langchain_mcp_adapters.tools import convert_mcp_tool_to_langchain_tool
-                lc_tool = convert_mcp_tool_to_langchain_tool(
-                    session,
-                    tool,
-                    connection=client_config[key],
-                    tool_interceptors=interceptors if interceptors else None,
-                    server_name=config.name,
-                    tool_name_prefix=False,
-                )
-
-                lc_tool.name = f"mcp__{key}__{original_name}"
-                if lc_tool.description and not lc_tool.description.startswith("[MCP"):
-                    lc_tool.description = f"[MCP:{config.name}] {lc_tool.description}"
-                wrap_tool_with_error_handler(
-                    lc_tool, config.id, config.name, original_name,
-                    on_tool_success=on_tool_success,
-                    on_tool_error=on_tool_error,
-                )
-                all_tools.append(lc_tool)
+                original_names = []
+                for tool in raw_tools:
+                    original_name = tool.name
+                    original_names.append(original_name)
+
+                    lc_tool = convert_mcp_tool_to_langchain_tool(
+                        None,
+                        tool,
+                        connection=client_config[key],
+                        tool_interceptors=interceptors if interceptors else None,
+                        server_name=config.name,
+                        tool_name_prefix=False,
+                    )
+
+                    lc_tool.name = f"mcp__{key}__{original_name}"
+                    if lc_tool.description and not lc_tool.description.startswith("[MCP"):
+                        lc_tool.description = f"[MCP:{config.name}] {lc_tool.description}"
+                    wrap_tool_with_error_handler(
+                        lc_tool, config.id, config.name, original_name,
+                        on_tool_success=on_tool_success,
+                        on_tool_error=on_tool_error,
+                    )
+                    all_tools.append(lc_tool)
 
             if on_batch_available:
                 try:
@@ -169,10 +174,12 @@ async def load_mcp_tools(
                 except Exception as e:
                     SYLogger.warning(f"[MCP] 写入工具状态失败: {e}")
 
-            SYLogger.info(f"[MCP] 服务器 '{config.name}' 加载了 {len(raw_tools)} 个工具")
+            SYLogger.info(
+                f"[MCP] 服务器 '{config.name}' 加载了 {len(raw_tools)} 个工具")
 
         except Exception as e:
-            SYLogger.warning(f"[MCP] 服务器 '{config.name}' ({config.server_url}) 连接失败，跳过: {e}")
+            SYLogger.warning(
+                f"[MCP] 服务器 '{config.name}' ({config.server_url}) 连接失败，跳过: {e}")
             if on_server_failure:
                 try:
                     await on_server_failure(config.id, str(e)[:300])

{sycommon_python_lib-0.2.5a7 → sycommon_python_lib-0.2.5a9}/src/sycommon/llm/struct_token.py

@@ -10,10 +10,55 @@ from sycommon.llm.llm_logger import LLMLogger
 from langfuse import Langfuse, LangfuseSpan, propagate_attributes
 from typing import Dict, List, Optional, Any
 import tiktoken
+import os
+import asyncio
 import threading
 import warnings
 warnings.filterwarnings("ignore", message="Pydantic serializer warnings")
 
+# 设置 tiktoken 缓存目录为本地内置文件，避免首次使用时从远程下载 BPE 词表
+os.environ["TIKTOKEN_CACHE_DIR"] = os.path.join(os.path.dirname(__file__), "tiktoken_cache")
+
+# 模块级 encoding 单例，所有实例共享，只初始化一次
+_cached_encoding = None
+_cached_model_name = None
+_encoding_lock = threading.Lock()
+
+
+def _init_encoding_sync(model_name: str):
+    """同步初始化 tiktoken encoding（线程安全）"""
+    global _cached_encoding, _cached_model_name
+    with _encoding_lock:
+        if _cached_encoding is not None and _cached_model_name == model_name:
+            return _cached_encoding
+
+        try:
+            _cached_encoding = tiktoken.encoding_for_model(model_name)
+            _cached_model_name = model_name
+        except KeyError:
+            try:
+                _cached_encoding = tiktoken.get_encoding("cl100k_base")
+            except Exception as e:
+                SYLogger.error(f"tiktoken 初始化失败，将使用字符数近似估算: {e}")
+        except Exception as e:
+            SYLogger.error(f"tiktoken 初始化失败，将使用字符数近似估算: {e}")
+
+    return _cached_encoding
+
+
+def _get_shared_encoding(model_name: str):
+    """获取模块级共享的 tiktoken encoding（同步路径，快速路径无锁）"""
+    if _cached_encoding is not None and _cached_model_name == model_name:
+        return _cached_encoding
+    return _init_encoding_sync(model_name)
+
+
+async def _aget_shared_encoding(model_name: str):
+    """获取模块级共享的 tiktoken encoding（异步路径，不阻塞事件循环）"""
+    if _cached_encoding is not None and _cached_model_name == model_name:
+        return _cached_encoding
+    return await asyncio.to_thread(_init_encoding_sync, model_name)
+
 
 class StructuredRunnableWithToken(Runnable):
     """
@@ -49,32 +94,15 @@ class StructuredRunnableWithToken(Runnable):
         self.max_compression_attempts = max_compression_attempts  # 最大压缩尝试次数
         self.is_native_mode = is_native_mode  # 是否为原生模式
 
-        # 初始化 Tokenizer（5秒超时保护）
+        # 初始化 Tokenizer（延迟到首次使用时初始化，避免在异步上下文中阻塞事件循环）
         self.encoding = None
-
-        def _init_encoding():
-            nonlocal encoding_result
-            try:
-                encoding_result = tiktoken.encoding_for_model(model_name)
-            except KeyError:
-                try:
-                    encoding_result = tiktoken.get_encoding("cl100k_base")
-                except Exception as e:
-                    SYLogger.error(f"tiktoken get_encoding 失败: {e}")
-            except Exception as e:
-                SYLogger.error(f"tiktoken encoding_for_model 失败: {e}")
-
-        encoding_result = None
-        t = threading.Thread(target=_init_encoding, daemon=True)
-        t.start()
-        t.join(timeout=5)
-        if t.is_alive():
-            SYLogger.error("tiktoken 初始化超时(5s)，将使用字符数近似估算")
-        elif encoding_result is not None:
-            self.encoding = encoding_result
+        self._encoding_initialized = False
 
     def _encode_text(self, text: str) -> int:
         """计算文本的 token 数量，encoding 不可用时按字符数近似估算"""
+        if not self._encoding_initialized:
+            self.encoding = _get_shared_encoding(self.model_name)
+            self._encoding_initialized = True
         if self.encoding:
             return len(self.encoding.encode(text))
         # 中文约 1.5 字符/token，英文约 4 字符/token，取中间值 2
@@ -466,10 +494,15 @@ class StructuredRunnableWithToken(Runnable):
                 user_id=user_id
             )
 
-            # 1. 适配输入
+            # 1. 确保 encoding 已初始化（异步路径，不阻塞事件循环）
+            if not self._encoding_initialized:
+                self.encoding = await _aget_shared_encoding(self.model_name)
+                self._encoding_initialized = True
+
+            # 2. 适配输入
             adapted_input = self._adapt_input(input)
 
-            # 2. 检查并执行上下文压缩 (仅在异步模式且开启时)
+            # 3. 检查并执行上下文压缩 (仅在异步模式且开启时)
             if self.enable_compression:
                 max_tokens = int(self.llmConfig.maxTokens * self.threshold_ratio)
                 current_tokens = self._count_tokens(adapted_input)