feat: add comprehensive debug logging for issue #477 hanging/freezing diagnosis (#662)

* feat: add comprehensive debug logging for issue #477 hanging/freezing diagnosis
- Add debug logging to src/server/app.py for event streaming and message chunk processing
- Track graph event flow with thread IDs for correlation
- Add detailed logging in interrupt event processing
- Add debug logging to src/agents/tool_interceptor.py for tool execution and interrupt handling
- Log interrupt decision flow and user feedback processing
- Add debug logging to src/graph/nodes.py for agent node execution
- Track step execution progress and agent coordination in research_team_node
- Add debug logging to src/agents/agents.py for agent creation and tool wrapping
- Update server.py to enable debug logging when --log-level debug is specified
- Add thread ID correlation throughout for better diagnostics
- Helps diagnose hanging/freezing issues during workflow execution

* Apply suggestions from code review

Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>

---------

Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>
This commit is contained in:
Willem Jiang
2025-10-27 08:21:30 +08:00
committed by GitHub
parent e9f0a02f1f
commit 83f1334db0
6 changed files with 171 additions and 27 deletions

View File

@@ -36,20 +36,42 @@ def create_agent(
Returns:
A configured agent graph
"""
logger.debug(
f"Creating agent '{agent_name}' of type '{agent_type}' "
f"with {len(tools)} tools and template '{prompt_template}'"
)
# Wrap tools with interrupt logic if specified
processed_tools = tools
if interrupt_before_tools:
logger.info(
f"Creating agent '{agent_name}' with tool-specific interrupts: {interrupt_before_tools}"
)
logger.debug(f"Wrapping {len(tools)} tools for agent '{agent_name}'")
processed_tools = wrap_tools_with_interceptor(tools, interrupt_before_tools)
logger.debug(f"Agent '{agent_name}' tool wrapping completed")
else:
logger.debug(f"Agent '{agent_name}' has no interrupt-before-tools configured")
return create_react_agent(
if agent_type not in AGENT_LLM_MAP:
logger.warning(
f"Agent type '{agent_type}' not found in AGENT_LLM_MAP. "
f"Falling back to default LLM type 'basic' for agent '{agent_name}'. "
"This may indicate a configuration issue."
)
llm_type = AGENT_LLM_MAP.get(agent_type, "basic")
logger.debug(f"Agent '{agent_name}' using LLM type: {llm_type}")
logger.debug(f"Creating ReAct agent '{agent_name}'")
agent = create_react_agent(
name=agent_name,
model=get_llm_by_type(AGENT_LLM_MAP[agent_type]),
model=get_llm_by_type(llm_type),
tools=processed_tools,
prompt=lambda state: apply_prompt_template(
prompt_template, state, locale=state.get("locale", "en-US")
),
pre_model_hook=pre_model_hook,
)
logger.info(f"Agent '{agent_name}' created successfully")
return agent

View File

@@ -84,47 +84,69 @@ class ToolInterceptor:
BaseTool: The wrapped tool with interrupt capability
"""
original_func = tool.func
logger.debug(f"Wrapping tool '{tool.name}' with interrupt capability")
def intercepted_func(*args: Any, **kwargs: Any) -> Any:
"""Execute the tool with interrupt check."""
tool_name = tool.name
logger.debug(f"[ToolInterceptor] Executing tool: {tool_name}")
# Format tool input for display
tool_input = args[0] if args else kwargs
tool_input_repr = ToolInterceptor._format_tool_input(tool_input)
logger.debug(f"[ToolInterceptor] Tool input: {tool_input_repr[:200]}")
if interceptor.should_interrupt(tool_name):
should_interrupt = interceptor.should_interrupt(tool_name)
logger.debug(f"[ToolInterceptor] should_interrupt={should_interrupt} for tool '{tool_name}'")
if should_interrupt:
logger.info(
f"Interrupting before tool '{tool_name}' with input: {tool_input_repr}"
f"[ToolInterceptor] Interrupting before tool '{tool_name}'"
)
logger.debug(
f"[ToolInterceptor] Interrupt message: About to execute tool '{tool_name}' with input: {tool_input_repr[:100]}..."
)
# Trigger interrupt and wait for user feedback
feedback = interrupt(
f"About to execute tool: '{tool_name}'\n\nInput:\n{tool_input_repr}\n\nApprove execution?"
)
logger.info(f"Interrupt feedback for '{tool_name}': {feedback}")
try:
feedback = interrupt(
f"About to execute tool: '{tool_name}'\n\nInput:\n{tool_input_repr}\n\nApprove execution?"
)
logger.debug(f"[ToolInterceptor] Interrupt returned with feedback: {f'{feedback[:100]}...' if feedback and len(feedback) > 100 else feedback if feedback else 'None'}")
except Exception as e:
logger.error(f"[ToolInterceptor] Error during interrupt: {str(e)}")
raise
logger.debug(f"[ToolInterceptor] Processing feedback approval for '{tool_name}'")
# Check if user approved
if not ToolInterceptor._parse_approval(feedback):
logger.warning(f"User rejected execution of tool '{tool_name}'")
is_approved = ToolInterceptor._parse_approval(feedback)
logger.info(f"[ToolInterceptor] Tool '{tool_name}' approval decision: {is_approved}")
if not is_approved:
logger.warning(f"[ToolInterceptor] User rejected execution of tool '{tool_name}'")
return {
"error": f"Tool execution rejected by user",
"tool": tool_name,
"status": "rejected",
}
logger.info(f"User approved execution of tool '{tool_name}'")
logger.info(f"[ToolInterceptor] User approved execution of tool '{tool_name}', proceeding")
# Execute the original tool
try:
logger.debug(f"[ToolInterceptor] Calling original function for tool '{tool_name}'")
result = original_func(*args, **kwargs)
logger.debug(f"Tool '{tool_name}' execution completed")
logger.info(f"[ToolInterceptor] Tool '{tool_name}' execution completed successfully")
logger.debug(f"[ToolInterceptor] Tool result length: {len(str(result))}")
return result
except Exception as e:
logger.error(f"Error executing tool '{tool_name}': {str(e)}")
logger.error(f"[ToolInterceptor] Error executing tool '{tool_name}': {str(e)}")
raise
# Replace the function and update the tool
# Use object.__setattr__ to bypass Pydantic validation
logger.debug(f"Attaching intercepted function to tool '{tool.name}'")
object.__setattr__(tool, "func", intercepted_func)
return tool