tamar-model-client 0.1.15__tar.gz → 0.1.16__tar.gz

{tamar_model_client-0.1.15 → tamar_model_client-0.1.16}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: tamar-model-client
-Version: 0.1.15
+Version: 0.1.16
 Summary: A Python SDK for interacting with the Model Manager gRPC service
 Home-page: http://gitlab.tamaredge.top/project-tap/AgentOS/model-manager-client
 Author: Oscar Ou
@@ -273,13 +273,13 @@ async def main():
     )
 
     # 发送请求并获取响应
-    response = await client.invoke(request_data)
-    if response.error:
-        print(f"错误: {response.error}")
-    else:
-        print(f"响应: {response.content}")
-        if response.usage:
-            print(f"Token 使用情况: {response.usage}")
+    async for r in await client.invoke(model_request):
+        if r.error:
+            print(f"错误: {r.error}")
+        else:
+            print(f"响应: {r.content}")
+            if r.usage:
+                print(f"Token 使用情况: {r.usage}")
 
 
 # 运行异步示例
@@ -531,7 +531,7 @@ python make_grpc.py
 ### 部署到 pip
 ```bash
 python setup.py sdist bdist_wheel
-twine check dist/*
+twine upload dist/*
 
 ```
 

{tamar_model_client-0.1.15 → tamar_model_client-0.1.16}/README.md

@@ -243,13 +243,13 @@ async def main():
     )
 
     # 发送请求并获取响应
-    response = await client.invoke(request_data)
-    if response.error:
-        print(f"错误: {response.error}")
-    else:
-        print(f"响应: {response.content}")
-        if response.usage:
-            print(f"Token 使用情况: {response.usage}")
+    async for r in await client.invoke(model_request):
+        if r.error:
+            print(f"错误: {r.error}")
+        else:
+            print(f"响应: {r.content}")
+            if r.usage:
+                print(f"Token 使用情况: {r.usage}")
 
 
 # 运行异步示例
@@ -501,7 +501,7 @@ python make_grpc.py
 ### 部署到 pip
 ```bash
 python setup.py sdist bdist_wheel
-twine check dist/*
+twine upload dist/*
 
 ```
 

{tamar_model_client-0.1.15 → tamar_model_client-0.1.16}/setup.py

@@ -2,7 +2,7 @@ from setuptools import setup, find_packages
 
 setup(
     name="tamar-model-client",
-    version="0.1.15",
+    version="0.1.16",
     description="A Python SDK for interacting with the Model Manager gRPC service",
     author="Oscar Ou",
     author_email="oscar.ou@tamaredge.ai",

{tamar_model_client-0.1.15 → tamar_model_client-0.1.16}/tamar_model_client/async_client.py

@@ -203,6 +203,37 @@ class AsyncTamarModelClient:
                     logger.error(f"❌ Non-retryable gRPC error: {e}", exc_info=True)
                     raise
 
+    async def _retry_request_stream(self, func, *args, **kwargs):
+        retry_count = 0
+        while retry_count < self.max_retries:
+            try:
+                return func(*args, **kwargs)
+            except (grpc.aio.AioRpcError, grpc.RpcError) as e:
+                # 对于取消的情况进行指数退避重试
+                if isinstance(e, grpc.aio.AioRpcError) and e.code() == grpc.StatusCode.CANCELLED:
+                    retry_count += 1
+                    logger.warning(f"❌ RPC cancelled, retrying {retry_count}/{self.max_retries}...")
+                    if retry_count < self.max_retries:
+                        delay = self.retry_delay * (2 ** (retry_count - 1))
+                        await asyncio.sleep(delay)
+                    else:
+                        logger.error("❌ Max retry reached for CANCELLED")
+                        raise
+                # 针对其他 RPC 错误类型，如暂时的连接问题、服务器超时等
+                elif isinstance(e, grpc.RpcError) and e.code() in {grpc.StatusCode.UNAVAILABLE,
+                                                                   grpc.StatusCode.DEADLINE_EXCEEDED}:
+                    retry_count += 1
+                    logger.warning(f"❌ gRPC error {e.code()}, retrying {retry_count}/{self.max_retries}...")
+                    if retry_count < self.max_retries:
+                        delay = self.retry_delay * (2 ** (retry_count - 1))
+                        await asyncio.sleep(delay)
+                    else:
+                        logger.error(f"❌ Max retry reached for {e.code()}")
+                        raise
+                else:
+                    logger.error(f"❌ Non-retryable gRPC error: {e}", exc_info=True)
+                    raise
+
     def _build_auth_metadata(self, request_id: str) -> list:
         # if not self.jwt_token and self.jwt_handler:
         # 更改为每次请求都生成一次token
@@ -263,25 +294,23 @@ class AsyncTamarModelClient:
             logger.info(f"🚀 Retrying connection (attempt {retry_count}/{self.max_retries}) after {delay:.2f}s delay...")
             await asyncio.sleep(delay)
 
-    async def _stream_inner(self, model_request, metadata, invoke_timeout) -> AsyncIterator[ModelResponse]:
-        """Inner function to handle the actual streaming gRPC call."""
-        async for response in self.stub.Invoke(model_request, metadata=metadata, timeout=invoke_timeout):
+    async def _stream(self, request, metadata, invoke_timeout) -> AsyncIterator[ModelResponse]:
+        async for response in self.stub.Invoke(request, metadata=metadata, timeout=invoke_timeout):
             yield ModelResponse(
                 content=response.content,
                 usage=json.loads(response.usage) if response.usage else None,
-                raw_response=json.loads(response.raw_response) if response.raw_response else None,
                 error=response.error or None,
+                raw_response=json.loads(response.raw_response) if response.raw_response else None,
+                request_id=response.request_id if response.request_id else None,
             )
 
-    async def _stream(self, model_request, metadata, invoke_timeout) -> AsyncIterator[ModelResponse]:
-        return await self._retry_request(self._stream_inner, model_request, metadata, invoke_timeout)
-
     async def _invoke_request(self, request, metadata, invoke_timeout):
         async for response in self.stub.Invoke(request, metadata=metadata, timeout=invoke_timeout):
             return ModelResponse(
                 content=response.content,
                 usage=json.loads(response.usage) if response.usage else None,
                 error=response.error or None,
+                raw_response=json.loads(response.raw_response) if response.raw_response else None,
                 request_id=response.request_id if response.request_id else None,
             )
 
@@ -373,7 +402,7 @@ class AsyncTamarModelClient:
         try:
             invoke_timeout = timeout or self.default_invoke_timeout
             if model_request.stream:
-                return await self._stream(request, metadata, invoke_timeout)
+                return await self._retry_request_stream(self._stream, request, metadata, invoke_timeout)
             else:
                 return await self._retry_request(self._invoke_request, request, metadata, invoke_timeout)
         except grpc.RpcError as e:

{tamar_model_client-0.1.15 → tamar_model_client-0.1.16}/tamar_model_client/sync_client.py

@@ -247,20 +247,16 @@ class TamarModelClient:
             logger.info(f"🚀 Retrying connection (attempt {retry_count}/{self.max_retries}) after {delay:.2f}s delay...")
             time.sleep(delay)  # Blocking sleep in sync version
 
-    def _stream_inner(self, model_request, metadata, invoke_timeout) -> Iterator[ModelResponse]:
-        """Inner function to handle the actual streaming gRPC call."""
-        response = self.stub.Invoke(model_request, metadata=metadata, timeout=invoke_timeout)
-        for res in response:
+    def _stream(self, request, metadata, invoke_timeout) -> Iterator[ModelResponse]:
+        for response in self.stub.Invoke(request, metadata=metadata, timeout=invoke_timeout):
             yield ModelResponse(
-                content=res.content,
-                usage=json.loads(res.usage) if res.usage else None,
-                raw_response=json.loads(res.raw_response) if res.raw_response else None,
-                error=res.error or None,
+                content=response.content,
+                usage=json.loads(response.usage) if response.usage else None,
+                error=response.error or None,
+                raw_response=json.loads(response.raw_response) if response.raw_response else None,
+                request_id=response.request_id if response.request_id else None,
             )
 
-    def _stream(self, model_request, metadata, invoke_timeout) -> Iterator[ModelResponse]:
-        return self._retry_request(self._stream_inner, model_request, metadata, invoke_timeout)
-
     def _invoke_request(self, request, metadata, invoke_timeout):
         response = self.stub.Invoke(request, metadata=metadata, timeout=invoke_timeout)
         for response in response:
@@ -268,6 +264,7 @@ class TamarModelClient:
                 content=response.content,
                 usage=json.loads(response.usage) if response.usage else None,
                 error=response.error or None,
+                raw_response=json.loads(response.raw_response) if response.raw_response else None,
                 request_id=response.request_id if response.request_id else None,
             )
 
@@ -358,7 +355,7 @@ class TamarModelClient:
         try:
             invoke_timeout = timeout or self.default_invoke_timeout
             if model_request.stream:
-                return self._stream(request, metadata, invoke_timeout)
+                return self._retry_request(self._stream, request, metadata, invoke_timeout)
             else:
                 return self._retry_request(self._invoke_request, request, metadata, invoke_timeout)
         except grpc.RpcError as e:

{tamar_model_client-0.1.15 → tamar_model_client-0.1.16}/tamar_model_client.egg-info/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: tamar-model-client
-Version: 0.1.15
+Version: 0.1.16
 Summary: A Python SDK for interacting with the Model Manager gRPC service
 Home-page: http://gitlab.tamaredge.top/project-tap/AgentOS/model-manager-client
 Author: Oscar Ou
@@ -273,13 +273,13 @@ async def main():
     )
 
     # 发送请求并获取响应
-    response = await client.invoke(request_data)
-    if response.error:
-        print(f"错误: {response.error}")
-    else:
-        print(f"响应: {response.content}")
-        if response.usage:
-            print(f"Token 使用情况: {response.usage}")
+    async for r in await client.invoke(model_request):
+        if r.error:
+            print(f"错误: {r.error}")
+        else:
+            print(f"响应: {r.content}")
+            if r.usage:
+                print(f"Token 使用情况: {r.usage}")
 
 
 # 运行异步示例
@@ -531,7 +531,7 @@ python make_grpc.py
 ### 部署到 pip
 ```bash
 python setup.py sdist bdist_wheel
-twine check dist/*
+twine upload dist/*
 
 ```