Files
ragflow/agent/tools/code_exec.py
eviaaaaa c59ae4c7c2 Fix: codeExec return types & error handling; Update Spark model mappings (#12896)
## What problem does this PR solve?

This PR addresses three specific issues to improve agent reliability and
model support:

1. **`codeExec` Output Limitation**: Previously, the `codeExec` tool was
strictly limited to returning `string` types. I updated the output
constraint to `object` to support structured data (Dicts, Lists, etc.)
required for complex downstream tasks.
2. **`codeExec` Error Handling**: Improved the execution logic so that
when runtime errors occur, the tool captures the exception and returns
the error message as the output instead of causing the process to abort
or fail silently.
3. **Spark Model Configuration**:
    - Added support for the `MAX-32k` model variant.
- Fixed the `Spark-Lite` mapping from `general` to `lite` to match the
latest API specifications.

## Type of change

- [x] Bug Fix (fixes execution logic and model mapping)
- [x] New Feature / Enhancement (adds model support and improves tool
flexibility)

## Key Changes

### `agent/tools/code_exec.py`
- Changed the output type definition from `string` to `object`.
- Refactored the execution flow to gracefully catch exceptions and
return error messages as part of the tool output.

### `rag/llm/chat_model.py`
- Added `"Spark-Max-32K": "max-32k"` to the model list.
- Updated `"Spark-Lite"` value from `"general"` to `"lite"`.

## Checklist
- [x] My code follows the style guidelines of this project.
- [x] I have performed a self-review of my own code.

Signed-off-by: evilhero <2278596667@qq.com>
2026-01-29 19:22:35 +08:00

382 lines
14 KiB
Python

#
# Copyright 2025 The InfiniFlow Authors. All Rights Reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
#
import ast
import base64
import json
import logging
import os
from abc import ABC
from typing import Optional
from pydantic import BaseModel, Field, field_validator
from strenum import StrEnum
from agent.tools.base import ToolBase, ToolMeta, ToolParamBase
from common import settings
from common.connection_utils import timeout
class Language(StrEnum):
PYTHON = "python"
NODEJS = "nodejs"
class CodeExecutionRequest(BaseModel):
code_b64: str = Field(..., description="Base64 encoded code string")
language: str = Field(default=Language.PYTHON.value, description="Programming language")
arguments: Optional[dict] = Field(default={}, description="Arguments")
@field_validator("code_b64")
@classmethod
def validate_base64(cls, v: str) -> str:
try:
base64.b64decode(v, validate=True)
return v
except Exception as e:
raise ValueError(f"Invalid base64 encoding: {str(e)}")
@field_validator("language", mode="before")
@classmethod
def normalize_language(cls, v) -> str:
if isinstance(v, str):
low = v.lower()
if low in ("python", "python3"):
return "python"
elif low in ("javascript", "nodejs"):
return "nodejs"
raise ValueError(f"Unsupported language: {v}")
class CodeExecParam(ToolParamBase):
"""
Define the code sandbox component parameters.
"""
def __init__(self):
self.meta: ToolMeta = {
"name": "execute_code",
"description": """
This tool has a sandbox that can execute code written in 'Python'/'Javascript'. It receives a piece of code and return a Json string.
Here's a code example for Python(`main` function MUST be included):
def main() -> dict:
\"\"\"
Generate Fibonacci numbers within 100.
\"\"\"
def fibonacci_recursive(n):
if n <= 1:
return n
else:
return fibonacci_recursive(n-1) + fibonacci_recursive(n-2)
return {
"result": fibonacci_recursive(100),
}
Here's a code example for Javascript(`main` function MUST be included and exported):
const axios = require('axios');
async function main(args) {
try {
const response = await axios.get('https://github.com/infiniflow/ragflow');
console.log('Body:', response.data);
} catch (error) {
console.error('Error:', error.message);
}
}
module.exports = { main };
""",
"parameters": {
"lang": {
"type": "string",
"description": "The programming language of this piece of code.",
"enum": ["python", "javascript"],
"required": True,
},
"script": {"type": "string", "description": "A piece of code in right format. There MUST be main function.", "required": True},
},
}
super().__init__()
self.lang = Language.PYTHON.value
self.script = 'def main(arg1: str, arg2: str) -> dict: return {"result": arg1 + arg2}'
self.arguments = {}
self.outputs = {"result": {"value": "", "type": "object"}}
def check(self):
self.check_valid_value(self.lang, "Support languages", ["python", "python3", "nodejs", "javascript"])
self.check_empty(self.script, "Script")
def get_input_form(self) -> dict[str, dict]:
res = {}
for k, v in self.arguments.items():
res[k] = {"type": "line", "name": k}
return res
class CodeExec(ToolBase, ABC):
component_name = "CodeExec"
@timeout(int(os.environ.get("COMPONENT_EXEC_TIMEOUT", 10 * 60)))
def _invoke(self, **kwargs):
if self.check_if_canceled("CodeExec processing"):
return
lang = kwargs.get("lang", self._param.lang)
script = kwargs.get("script", self._param.script)
arguments = {}
for k, v in self._param.arguments.items():
if kwargs.get(k):
arguments[k] = kwargs[k]
continue
arguments[k] = self._canvas.get_variable_value(v) if v else None
return self._execute_code(language=lang, code=script, arguments=arguments)
def _execute_code(self, language: str, code: str, arguments: dict):
import requests
if self.check_if_canceled("CodeExec execution"):
return self.output()
try:
# Try using the new sandbox provider system first
try:
from agent.sandbox.client import execute_code as sandbox_execute_code
if self.check_if_canceled("CodeExec execution"):
return
# Execute code using the provider system
result = sandbox_execute_code(
code=code,
language=language,
timeout=int(os.environ.get("COMPONENT_EXEC_TIMEOUT", 10 * 60)),
arguments=arguments
)
if self.check_if_canceled("CodeExec execution"):
return
# Process the result
if result.stderr:
self.set_output("_ERROR", result.stderr)
return
parsed_stdout = self._deserialize_stdout(result.stdout)
logging.info(f"[CodeExec]: Provider system -> {parsed_stdout}")
self._populate_outputs(parsed_stdout, result.stdout)
return
except (ImportError, RuntimeError) as provider_error:
# Provider system not available or not configured, fall back to HTTP
logging.info(f"[CodeExec]: Provider system not available, using HTTP fallback: {provider_error}")
# Fallback to direct HTTP request
code_b64 = self._encode_code(code)
code_req = CodeExecutionRequest(code_b64=code_b64, language=language, arguments=arguments).model_dump()
except Exception as e:
if self.check_if_canceled("CodeExec execution"):
return self.output()
self.set_output("_ERROR", "construct code request error: " + str(e))
return self.output()
try:
if self.check_if_canceled("CodeExec execution"):
self.set_output("_ERROR", "Task has been canceled")
return self.output()
resp = requests.post(url=f"http://{settings.SANDBOX_HOST}:9385/run", json=code_req, timeout=int(os.environ.get("COMPONENT_EXEC_TIMEOUT", 10 * 60)))
logging.info(f"http://{settings.SANDBOX_HOST}:9385/run, code_req: {code_req}, resp.status_code {resp.status_code}:")
if self.check_if_canceled("CodeExec execution"):
return "Task has been canceled"
if resp.status_code != 200:
resp.raise_for_status()
body = resp.json()
if body:
stderr = body.get("stderr")
if stderr:
self.set_output("_ERROR", stderr)
return self.output()
raw_stdout = body.get("stdout", "")
parsed_stdout = self._deserialize_stdout(raw_stdout)
logging.info(f"[CodeExec]: http://{settings.SANDBOX_HOST}:9385/run -> {parsed_stdout}")
self._populate_outputs(parsed_stdout, raw_stdout)
else:
self.set_output("_ERROR", "There is no response from sandbox")
return self.output()
except Exception as e:
if self.check_if_canceled("CodeExec execution"):
return self.output()
self.set_output("_ERROR", "Exception executing code: " + str(e))
return self.output()
def _encode_code(self, code: str) -> str:
return base64.b64encode(code.encode("utf-8")).decode("utf-8")
def thoughts(self) -> str:
return "Running a short script to process data."
def _deserialize_stdout(self, stdout: str):
text = str(stdout).strip()
if not text:
return ""
for loader in (json.loads, ast.literal_eval):
try:
return loader(text)
except Exception:
continue
return text
def _coerce_output_value(self, value, expected_type: Optional[str]):
if expected_type is None:
return value
etype = expected_type.strip().lower()
inner_type = None
if etype.startswith("array<") and etype.endswith(">"):
inner_type = etype[6:-1].strip()
etype = "array"
try:
if etype == "string":
return "" if value is None else str(value)
if etype == "number":
if value is None or value == "":
return None
if isinstance(value, (int, float)):
return value
if isinstance(value, str):
try:
return float(value)
except Exception:
return value
return float(value)
if etype == "boolean":
if isinstance(value, bool):
return value
if isinstance(value, str):
lv = value.lower()
if lv in ("true", "1", "yes", "y", "on"):
return True
if lv in ("false", "0", "no", "n", "off"):
return False
return bool(value)
if etype == "array":
candidate = value
if isinstance(candidate, str):
parsed = self._deserialize_stdout(candidate)
candidate = parsed
if isinstance(candidate, tuple):
candidate = list(candidate)
if not isinstance(candidate, list):
candidate = [] if candidate is None else [candidate]
if inner_type == "string":
return ["" if v is None else str(v) for v in candidate]
if inner_type == "number":
coerced = []
for v in candidate:
try:
if v is None or v == "":
coerced.append(None)
elif isinstance(v, (int, float)):
coerced.append(v)
else:
coerced.append(float(v))
except Exception:
coerced.append(v)
return coerced
return candidate
if etype == "object":
if isinstance(value, dict):
return value
if isinstance(value, str):
parsed = self._deserialize_stdout(value)
if isinstance(parsed, dict):
return parsed
return value
except Exception:
return value
return value
def _populate_outputs(self, parsed_stdout, raw_stdout: str):
outputs_items = list(self._param.outputs.items())
logging.info(f"[CodeExec]: outputs schema keys: {[k for k, _ in outputs_items]}")
if not outputs_items:
return
if isinstance(parsed_stdout, dict):
for key, meta in outputs_items:
if key.startswith("_"):
continue
val = self._get_by_path(parsed_stdout, key)
if val is None and len(outputs_items) == 1:
val = parsed_stdout
coerced = self._coerce_output_value(val, meta.get("type"))
logging.info(f"[CodeExec]: populate dict key='{key}' raw='{val}' coerced='{coerced}'")
self.set_output(key, coerced)
return
if isinstance(parsed_stdout, (list, tuple)):
for idx, (key, meta) in enumerate(outputs_items):
if key.startswith("_"):
continue
val = parsed_stdout[idx] if idx < len(parsed_stdout) else None
coerced = self._coerce_output_value(val, meta.get("type"))
logging.info(f"[CodeExec]: populate list key='{key}' raw='{val}' coerced='{coerced}'")
self.set_output(key, coerced)
return
default_val = parsed_stdout if parsed_stdout is not None else raw_stdout
for idx, (key, meta) in enumerate(outputs_items):
if key.startswith("_"):
continue
val = default_val if idx == 0 else None
coerced = self._coerce_output_value(val, meta.get("type"))
logging.info(f"[CodeExec]: populate scalar key='{key}' raw='{val}' coerced='{coerced}'")
self.set_output(key, coerced)
def _get_by_path(self, data, path: str):
if not path:
return None
cur = data
for part in path.split("."):
part = part.strip()
if not part:
return None
if isinstance(cur, dict):
cur = cur.get(part)
elif isinstance(cur, list):
try:
idx = int(part)
cur = cur[idx]
except Exception:
return None
else:
return None
if cur is None:
return None
logging.info(f"[CodeExec]: resolve path '{path}' -> {cur}")
return cur