Merge pull request #89 from Serverless-Devs/copilot/fix-terminated-sandbox-issue

fix: raise ResourceNotExistError in delete_sandbox when data plane returns "sandbox not found"

Author: OhYee

agentrun/sandbox/client.py

@@ -717,7 +717,11 @@ async def delete_sandbox_async(
             Sandbox: 停止后的 Sandbox 对象
 
         Raises:
-            ResourceNotExistError: Sandbox 不存在
+            ResourceNotExistError: Sandbox 不存在（包括 HTTP 404 与数据面业务层
+                not-found 两种情形）。调用方可 catch 此异常实现幂等删除。
+                Sandbox does not exist (covers both HTTP 404 and data-plane
+                business-level not-found).  Callers can catch this exception
+                for idempotent delete logic.
             ClientError: 客户端错误
             ServerError: 服务器错误
         """
@@ -728,11 +732,24 @@ async def delete_sandbox_async(
 
             # 判断返回结果是否成功
             if result.get("code") != "SUCCESS":
+                message = result.get("message", "")
+                # 数据面报告 sandbox 不存在时，与 HTTP 404 路径保持一致，
+                # 统一抛出 ResourceNotExistError，方便调用方幂等处理。
+                # When the data plane reports sandbox not found, raise
+                # ResourceNotExistError for consistency with the HTTP 404 path.
+                # Callers can catch ResourceNotExistError to implement idempotent
+                # deletion (e.g. when TERMINATED instances still appear in list
+                # results but have already been removed from the data plane).
+                # Note: long-term the server should return a stable error_code
+                # (e.g. SandboxNotFound) so the SDK can match on that instead
+                # of a message string.
+                if "sandbox not found" in message.lower():
+                    raise ResourceNotExistError("Sandbox", sandbox_id)
                 raise ClientError(
                     status_code=0,
                     message=(
                         "Failed to stop sandbox:"
-                        f" {result.get('message', 'Unknown error')}"
+                        f" {message or 'Unknown error'}"
                     ),
                 )
 
@@ -757,7 +774,11 @@ def delete_sandbox(
             Sandbox: 停止后的 Sandbox 对象
 
         Raises:
-            ResourceNotExistError: Sandbox 不存在
+            ResourceNotExistError: Sandbox 不存在（包括 HTTP 404 与数据面业务层
+                not-found 两种情形）。调用方可 catch 此异常实现幂等删除。
+                Sandbox does not exist (covers both HTTP 404 and data-plane
+                business-level not-found).  Callers can catch this exception
+                for idempotent delete logic.
             ClientError: 客户端错误
             ServerError: 服务器错误
         """
@@ -768,11 +789,24 @@ def delete_sandbox(
 
             # 判断返回结果是否成功
             if result.get("code") != "SUCCESS":
+                message = result.get("message", "")
+                # 数据面报告 sandbox 不存在时，与 HTTP 404 路径保持一致，
+                # 统一抛出 ResourceNotExistError，方便调用方幂等处理。
+                # When the data plane reports sandbox not found, raise
+                # ResourceNotExistError for consistency with the HTTP 404 path.
+                # Callers can catch ResourceNotExistError to implement idempotent
+                # deletion (e.g. when TERMINATED instances still appear in list
+                # results but have already been removed from the data plane).
+                # Note: long-term the server should return a stable error_code
+                # (e.g. SandboxNotFound) so the SDK can match on that instead
+                # of a message string.
+                if "sandbox not found" in message.lower():
+                    raise ResourceNotExistError("Sandbox", sandbox_id)
                 raise ClientError(
                     status_code=0,
                     message=(
                         "Failed to stop sandbox:"
-                        f" {result.get('message', 'Unknown error')}"
+                        f" {message or 'Unknown error'}"
                     ),
                 )
 

tests/unittests/sandbox/test_client.py

@@ -810,6 +810,119 @@ def test_delete_sandbox_not_exist(
         with pytest.raises(ResourceNotExistError):
             client.delete_sandbox("nonexistent")
 
+    @patch("agentrun.sandbox.client.SandboxControlAPI")
+    @patch("agentrun.sandbox.client.SandboxDataAPI")
+    def test_delete_sandbox_not_found_in_response_raises_resource_not_exist(
+        self, mock_data_api_class, mock_control_api_class
+    ):
+        """数据面业务层返回 not-found 时，与 HTTP 404 路径统一抛 ResourceNotExistError。
+
+        Callers can catch ResourceNotExistError for idempotent deletion when the
+        control plane still lists a TERMINATED sandbox but the data plane has
+        already removed it (e.g. ``except ResourceNotExistError: pass``).
+        """
+        mock_data_api = MagicMock()
+        mock_data_api.delete_sandbox.return_value = {
+            "code": "FAILED",
+            "message": "sandbox not found",
+        }
+        mock_data_api_class.return_value = mock_data_api
+
+        client = SandboxClient()
+        with pytest.raises(ResourceNotExistError):
+            client.delete_sandbox("sandbox-123")
+
+    @patch("agentrun.sandbox.client.SandboxControlAPI")
+    @patch("agentrun.sandbox.client.SandboxDataAPI")
+    def test_delete_sandbox_not_found_case_insensitive(
+        self, mock_data_api_class, mock_control_api_class
+    ):
+        """大小写变体（如 'Sandbox NOT FOUND'）也应触发 ResourceNotExistError。"""
+        mock_data_api = MagicMock()
+        mock_data_api.delete_sandbox.return_value = {
+            "code": "FAILED",
+            "message": "Sandbox NOT FOUND",
+        }
+        mock_data_api_class.return_value = mock_data_api
+
+        client = SandboxClient()
+        with pytest.raises(ResourceNotExistError):
+            client.delete_sandbox("sandbox-123")
+
+    @patch("agentrun.sandbox.client.SandboxControlAPI")
+    @patch("agentrun.sandbox.client.SandboxDataAPI")
+    def test_delete_sandbox_other_failure_message_raises_client_error(
+        self, mock_data_api_class, mock_control_api_class
+    ):
+        """无关 not-found 的失败消息（如 'sandbox is busy'）应仍抛 ClientError。"""
+        mock_data_api = MagicMock()
+        mock_data_api.delete_sandbox.return_value = {
+            "code": "FAILED",
+            "message": "sandbox is busy",
+        }
+        mock_data_api_class.return_value = mock_data_api
+
+        client = SandboxClient()
+        with pytest.raises(ClientError, match="Failed to stop sandbox"):
+            client.delete_sandbox("sandbox-123")
+
+    @patch("agentrun.sandbox.client.SandboxControlAPI")
+    @patch("agentrun.sandbox.client.SandboxDataAPI")
+    def test_delete_sandbox_empty_message_raises_client_error(
+        self, mock_data_api_class, mock_control_api_class
+    ):
+        """message 为空时不应误触 not-found 逻辑，应抛 ClientError。"""
+        mock_data_api = MagicMock()
+        mock_data_api.delete_sandbox.return_value = {
+            "code": "FAILED",
+            "message": "",
+        }
+        mock_data_api_class.return_value = mock_data_api
+
+        client = SandboxClient()
+        with pytest.raises(ClientError, match="Failed to stop sandbox"):
+            client.delete_sandbox("sandbox-123")
+
+    @patch("agentrun.sandbox.client.SandboxControlAPI")
+    @patch("agentrun.sandbox.client.SandboxDataAPI")
+    @pytest.mark.asyncio
+    async def test_delete_sandbox_async_not_found_in_response_raises_resource_not_exist(
+        self, mock_data_api_class, mock_control_api_class
+    ):
+        """数据面业务层返回 not-found 时（async），与 HTTP 404 路径统一抛 ResourceNotExistError。"""
+        mock_data_api = MagicMock()
+        mock_data_api.delete_sandbox_async = AsyncMock(
+            return_value={
+                "code": "FAILED",
+                "message": "sandbox not found",
+            }
+        )
+        mock_data_api_class.return_value = mock_data_api
+
+        client = SandboxClient()
+        with pytest.raises(ResourceNotExistError):
+            await client.delete_sandbox_async("sandbox-123")
+
+    @patch("agentrun.sandbox.client.SandboxControlAPI")
+    @patch("agentrun.sandbox.client.SandboxDataAPI")
+    @pytest.mark.asyncio
+    async def test_delete_sandbox_async_other_failure_raises_client_error(
+        self, mock_data_api_class, mock_control_api_class
+    ):
+        """无关 not-found 的失败消息（async）应仍抛 ClientError。"""
+        mock_data_api = MagicMock()
+        mock_data_api.delete_sandbox_async = AsyncMock(
+            return_value={
+                "code": "FAILED",
+                "message": "sandbox is busy",
+            }
+        )
+        mock_data_api_class.return_value = mock_data_api
+
+        client = SandboxClient()
+        with pytest.raises(ClientError, match="Failed to stop sandbox"):
+            await client.delete_sandbox_async("sandbox-123")
+
     @patch("agentrun.sandbox.client.SandboxControlAPI")
     @patch("agentrun.sandbox.client.SandboxDataAPI")
     @pytest.mark.asyncio