fix: propagate error on tool failure (#2281)

Co-authored-by: Caren Thomas <caren@caren-mac.local>
This commit is contained in:
cthomas
2024-12-20 16:38:16 -08:00
committed by GitHub
parent 5bb4888cea
commit a5b1aac1fd
5 changed files with 64 additions and 5 deletions

View File

@@ -9,6 +9,7 @@ from typing import List, Optional, Tuple, Union
from letta.constants import (
BASE_TOOLS,
CLI_WARNING_PREFIX,
ERROR_MESSAGE_PREFIX,
FIRST_MESSAGE_ATTEMPTS,
FUNC_FAILED_HEARTBEAT_MESSAGE,
LLM_MAX_TOKENS,
@@ -498,7 +499,7 @@ class Agent(BaseAgent):
function_args.pop("self", None)
# error_msg = f"Error calling function {function_name} with args {function_args}: {str(e)}"
# Less detailed - don't provide full args, idea is that it should be in recent context so no need (just adds noise)
error_msg = f"Error calling function {function_name}: {str(e)}"
error_msg = get_friendly_error_msg(function_name=function_name, exception_name=type(e).__name__, exception_message=str(e))
error_msg_user = f"{error_msg}\n{traceback.format_exc()}"
printd(error_msg_user)
function_response = package_function_response(False, error_msg)
@@ -521,8 +522,29 @@ class Agent(BaseAgent):
self.interface.function_message(f"Error: {error_msg}", msg_obj=messages[-1])
return messages, False, True # force a heartbeat to allow agent to handle error
# Step 4: check if function response is an error
if function_response_string.startswith(ERROR_MESSAGE_PREFIX):
function_response = package_function_response(False, function_response_string)
# TODO: truncate error message somehow
messages.append(
Message.dict_to_message(
agent_id=self.agent_state.id,
user_id=self.agent_state.created_by_id,
model=self.model,
openai_message_dict={
"role": "tool",
"name": function_name,
"content": function_response,
"tool_call_id": tool_call_id,
},
)
) # extend conversation with function response
self.interface.function_message(f"Ran {function_name}({function_args})", msg_obj=messages[-1])
self.interface.function_message(f"Error: {function_response_string}", msg_obj=messages[-1])
return messages, False, True # force a heartbeat to allow agent to handle error
# If no failures happened along the way: ...
# Step 4: send the info on the function call and function response to GPT
# Step 5: send the info on the function call and function response to GPT
messages.append(
Message.dict_to_message(
agent_id=self.agent_state.id,

View File

@@ -69,6 +69,8 @@ INITIAL_BOOT_MESSAGE_SEND_MESSAGE_FIRST_MSG = STARTUP_QUOTES[2]
CLI_WARNING_PREFIX = "Warning: "
ERROR_MESSAGE_PREFIX = "Error"
NON_USER_MSG_PREFIX = "[This is an automated system message hidden from the user] "
# Constants to do with summarization / conversation length window

View File

@@ -238,7 +238,7 @@ class QueuingInterface(AgentInterface):
new_message = {"function_return": msg, "status": "success"}
elif msg.startswith("Error: "):
msg = msg.replace("Error: ", "")
msg = msg.replace("Error: ", "", 1)
new_message = {"function_return": msg, "status": "error"}
else:
@@ -951,7 +951,7 @@ class StreamingServerInterface(AgentChunkStreamingInterface):
)
elif msg.startswith("Error: "):
msg = msg.replace("Error: ", "")
msg = msg.replace("Error: ", "", 1)
# new_message = {"function_return": msg, "status": "error"}
assert msg_obj.tool_call_id is not None
new_message = ToolReturnMessage(

View File

@@ -28,6 +28,7 @@ from letta.constants import (
CLI_WARNING_PREFIX,
CORE_MEMORY_HUMAN_CHAR_LIMIT,
CORE_MEMORY_PERSONA_CHAR_LIMIT,
ERROR_MESSAGE_PREFIX,
LETTA_DIR,
MAX_FILENAME_LENGTH,
TOOL_CALL_ID_MAX_LEN,
@@ -1122,7 +1123,7 @@ def sanitize_filename(filename: str) -> str:
def get_friendly_error_msg(function_name: str, exception_name: str, exception_message: str):
from letta.constants import MAX_ERROR_MESSAGE_CHAR_LIMIT
error_msg = f"Error executing function {function_name}: {exception_name}: {exception_message}"
error_msg = f"{ERROR_MESSAGE_PREFIX} executing function {function_name}: {exception_name}: {exception_message}"
if len(error_msg) > MAX_ERROR_MESSAGE_CHAR_LIMIT:
error_msg = error_msg[:MAX_ERROR_MESSAGE_CHAR_LIMIT]
return error_msg

View File

@@ -1,4 +1,5 @@
import asyncio
import json
import os
import threading
import time
@@ -382,6 +383,39 @@ def test_function_return_limit(client: Union[LocalClient, RESTClient]):
client.delete_agent(agent_id=agent.id)
def test_function_always_error(client: Union[LocalClient, RESTClient]):
"""Test to see if function that errors works correctly"""
def always_error():
"""
Always throw an error.
"""
return 5/0
tool = client.create_or_update_tool(func=always_error)
agent = client.create_agent(tool_ids=[tool.id])
# get function response
response = client.send_message(agent_id=agent.id, message="call the always_error function", role="user")
print(response.messages)
response_message = None
for message in response.messages:
if isinstance(message, FunctionReturn):
response_message = message
break
assert response_message, "FunctionReturn message not found in response"
assert response_message.status == "error"
if isinstance(client, RESTClient):
assert response_message.function_return == "Error executing function always_error: ZeroDivisionError: division by zero"
else:
response_json = json.loads(response_message.function_return)
assert response_json['status'] == "Failed"
assert response_json['message'] == "Error executing function always_error: ZeroDivisionError: division by zero"
client.delete_agent(agent_id=agent.id)
@pytest.mark.asyncio
async def test_send_message_parallel(client: Union[LocalClient, RESTClient], agent: AgentState, request):
"""