Fix/runtime context thread id bug

[SARTHAK2511]

Gemini said
Markdown

Bug Fix Report: AttributeError in DeerFlow Agent

## Problem

The DeerFlow agent was unable to process any messages due to an AttributeError: 'NoneType' object has no attribute 'get'. This occurred because the code incorrectly assumed runtime.context would contain a dictionary with thread_id, but in LangGraph, runtime.context is None by default.

---

Root Cause

The original implementation incorrectly accessed thread_id using:

thread_id = runtime.context["thread_id"] # runtime.context is None

In LangGraph's architecture:

• Runtime.context defaults to None unless explicitly passed during graph invocation.
• The LangGraph API server does not pass any context when invoking graphs.
• The thread_id is actually stored in the RunnableConfig under config["configurable"]["thread_id"].

---

Solution

Replace all instances of runtime.context["thread_id"] access with the proper LangGraph pattern:

from langgraph.config import get_config

config = get_config()
thread_id = config.get("configurable", {}).get("thread_id")

Changes Made

Core Fixes (6 files)
backend/src/agents/middlewares/thread_data_middleware.py - Fixed thread directory initialization.

backend/src/agents/middlewares/memory_middleware.py - Fixed memory system thread access.

backend/src/agents/middlewares/uploads_middleware.py - Fixed file upload handling.

backend/src/sandbox/middleware.py - Fixed sandbox acquisition.

backend/src/sandbox/tools.py - Fixed sandbox tools thread access.

backend/src/tools/builtins/task_tool.py - Fixed subagent task delegation.

Additional Improvements (2 files)

backend/src/community/aio_sandbox/aio_sandbox_provider.py - Added cleanup for orphaned thread locks to prevent memory leaks.

backend/src/subagents/executor.py - Added thread pool cleanup with atexit handler and TTL-based task cleanup.

Testing & Impact
Testing Results
✅ Successfully built and ran with Docker Compose.

✅ Agent processes messages correctly (tested with simple math queries).

✅ Resolved AttributeError in middleware execution.

✅ Thread creation and sandbox acquisition are now working properly.

Impact
This was a critical bug that prevented the agent from functioning. This fix enables:

Proper message processing.

Thread-specific data directory creation.

Sandbox environment acquisition.

Full Memory system functionality.

Reliable file upload handling and subagent delegation.

Compatibility & Notes
Version: Works with LangGraph >= 1.0.6 (as specified in pyproject.toml).

Backward Compatibility: Maintains support for existing threads; no breaking changes to external APIs.

Note: This bug appeared to be a misunderstanding of the LangGraph Runtime API. This fix aligns the codebase with documented patterns for accessing configuration within middleware and tools.

[CLAassistant]

Thank you for your submission! We really appreciate it. Like many open source projects, we ask that you sign our Contributor License Agreement before we can accept your contribution.
_{You have signed the CLA already but the status is still pending? Let us recheck it.}

[zhiyunyao]

Hello, I am currently exploring the detailed implementation of deer-flow. To my knowledge, the frontend has explicitly set the thread_id field within runtime.context. Refer to hooks.ts.