langgenius/dify
1
0
Fork 0
mirror of https://github.com/langgenius/dify.git synced 2026-08-27 03:01:54 -04:00
Code Issues 775 Packages Projects Releases 10 Wiki Activity

Add a custom built-in tool to the workflow and run it, but the workflow gets stuck in the "running" mode and does not produce any results. #3743

New Issue
Closed
opened 2026-02-21 18:01:31 -05:00 by yindo · 0 comments
yindo commented 2026-02-21 18:01:31 -05:00
Owner
Copy Link
Copy Source

Originally created by @D-siheng on GitHub (May 27, 2024).

Self Checks

  • This is only for bug report, if you would like to ask a question, please head to Discussions.
  • I have searched for existing issues search for existing issues, including closed ones.
  • I confirm that I am using English to submit this report (我已阅读并同意 Language Policy).
  • Please do not modify this template :) and fill in all the required fields.

Dify version

0.6.7

Cloud or Self Hosted

Self Hosted (Docker)

Steps to reproduce

I wrote an built-in tool following the official guidelines and referencing the code of the Wiki search tool. The tool's function is to output document summaries using the document summary module of langchain.
The tool can be displayed normally in the list of built-in tools, and after testing, it can run normally in workflow and output document summaries.
I created a START-document summary-END workflow and published it. I ran the workflow in the explore interface, and all three nodes in the workflow displayed as completed, and the summary tool output the document summary content normally in the nodes. The end node also had normal input and output, but the workflow process was stuck in the running state and could not become the succeed state, which resulted in the summary result not being output on the screen.
After testing, the situation where the workflow process could not end only occurred when a docx document was to be summarized. Besides the workflow, I also tried the chatflow, which consisted of STRAT-document summary-ANSWER three nodes. Running the chatflow could normally end and output formatted document summary content.
My code is as follows:

# summarizer.yaml
identity:
  author: DLH
  name: summarizer
  label:
    en_US: Summarizer
    zh_Hans: 文档摘要
    pt_BR: Summarizer
  description:
    en_US: This tool is a documentation abstract of load_summarize_chain implementation using langchain
    zh_Hans: 该工具是使用langchain的load_summarize_chain实现的文档摘要
    pt_BR: This tool is a documentation abstract of load_summarize_chain implementation using langchain
  icon: icon.svg
credentials_for_provider:

# summarizer.py
from core.tools.errors import ToolProviderCredentialValidationError
from core.tools.provider.builtin_tool_provider import BuiltinToolProviderController

from core.tools.provider.builtin.summarizer.tools.langchain_summarize import SummarizeTool

class SummarizeProvider(BuiltinToolProviderController):
    def _validate_credentials(self, credentials: dict) -> None:
        try:
            SummarizeTool().fork_tool_runtime(
                meta={
                    "credentials": credentials,
                }
            ).invoke(
                user_id='',
                tool_parameters={
                    "File_Path": "file path",
                    # 1、添加按ChunkSize大小对Docx文档进行分段功能
                    "ChunkSize": 0
                },
            )
        except Exception as e:
            raise ToolProviderCredentialValidationError(str(e))

# langchain_summarize.yaml
identity:
  author: DLH
  name: summarizer
  label:
    en_US: Summarizer
    zh_Hans: 文档摘要
    pt_BR: Summarizer
  description:
    en_US: This tool is a documentation abstract of load_summarize_chain implementation using langchain
    zh_Hans: 该工具是使用langchain的load_summarize_chain实现的文档摘要
    pt_BR: This tool is a documentation abstract of load_summarize_chain implementation using langchain
  icon: icon.svg
credentials_for_provider:

# langchain_summarize.py
import os
import glob
import hashlib
import tiktoken
from langchain_community.llms import VLLMOpenAI
from langchain.prompts import PromptTemplate
# 1、添加按ChunkSize大小对Docx文档进行分段功能
from langchain.text_splitter import RecursiveCharacterTextSplitter
from langchain.text_splitter import CharacterTextSplitter
from langchain.chains.summarize import load_summarize_chain
from langchain.docstore.document import Document
from langchain_community.document_loaders import PyPDFLoader
from langchain_community.document_loaders import Docx2txtLoader
import logging
import traceback

from typing import Any, Optional, Union
from core.tools.entities.tool_entities import ToolInvokeMessage
from core.tools.tool.builtin_tool import BuiltinTool

class SummarizeTool(BuiltinTool):
    # 文档摘要功能函数
        def summarize_from_folder(self, folder, llm, ChunkSize):
            # 变量定义
            summaries = []

            logging.basicConfig(level=logging.DEBUG)

            logging.info('摘要函数 +——+ 开始执行')

            # 加载大模型
            QWEN = VLLMOpenAI(
                max_tokens=10000, # 大模型的总tokens长度
                temperature=0.7, # 温度，越高生成的内容复杂度就越高
                openai_api_key="EMPTY", # openAI模型的APIKEY，因用的本地模型，因此设为EMPTY，如果使用联网模型则输入对应key值即可
                openai_api_base="http://xx.xx.x.xx:8080/v1", # 使用vllm推理引擎加载模型，这里填入引擎对外提供的API接口，因隐私安全问题使用xx代替
                model_name="/data/models/Qwen1.5-72B-Chat/" # 同上，填入对应模型的名称
            )

            # 文档读取
            for file in glob.glob(folder + "/*"):
                # 获取文件名
                file_name = os.path.basename(file)
                # pdf文档
                if file.endswith(".pdf"):
                    loader = PyPDFLoader(file)
                    # 返回一个文档数组存入docs中，每个元素为一个页面内容
                    docs = loader.load_and_split()
                # docx文档
                elif file.endswith(".docx"):
                    docx_loader = Docx2txtLoader(file)
                    # 1、添加按ChunkSize大小对Docx文档进行分段功能
                    if ChunkSize != 0: # 设置ChunkSize的值时，按ChunkSize大小对docx进行分段
                        text_splitter = RecursiveCharacterTextSplitter(chunk_size=ChunkSize, chunk_overlap=50)
                        data = docx_loader.load_and_split(text_splitter)
                    else: # 否则就按'/n/n'对docx进行分段
                        data = docx_loader.load()
                        text_splitter = CharacterTextSplitter(chunk_size=512, chunk_overlap=50)
                    docs = text_splitter.split_documents(data)
                else:
                    return self.create_text_message("Unsupported file type. Only support docx & pdf.")
                    # print(f"Unsupported file type: {file}.Only support docx & pdf.")
                    # continue
                # 文档摘要生成
                chain = load_summarize_chain(llm, chain_type="map_reduce", return_intermediate_steps=False)
                summary = chain.run(docs)
                # 创建一个包含文件名和摘要的字符串
                file_summary_str = f"Summary of {file_name}:\n\n{summary}"
                # print(file_summary_str)
                # print("\n")
                summaries.append(file_summary_str)

                logging.info('摘要函数 +——+ 结束')
            return "\n\n".join(summaries)
    
        def _invoke(self, 
                    user_id: str,  # 用户ID
                tool_parameters: dict[str, Any],  # 工具参数
            ) -> Union[ToolInvokeMessage, list[ToolInvokeMessage]]:
            """
                invoke tools
            """


            blobpath = "https://openaipublic.blob.core.windows.net/encodings/cl100k_base.tiktoken"
            cache_key = hashlib.sha1(blobpath.encode()).hexdigest()
            # 配置tiktoken的本地缓存目录
            tiktoken_cache_dir = "/app/api"
            os.environ["TIKTOKEN_CACHE_DIR"] = tiktoken_cache_dir
            assert os.path.exists(os.path.join(tiktoken_cache_dir, cache_key))

            folder = tool_parameters.get('File_Path', '')
            # 1、添加按ChunkSize大小对Docx文档进行分段功能
            chunksize = tool_parameters.get('ChunkSize', 0)
            if not folder:
                return self.create_text_message('Please input folder')
            logging.info('invoke函数 执行')
            result = self.summarize_from_folder(folder, QWEN, chunksize)

            # result = "\n\n".join(results)

            with open(filename, 'w') as f:
                traceback.print_stack(file=f)

            logging.info('invoke函数 结束')

            return self.create_text_message(self.summary(user_id=user_id,content=result))

✔️ Expected Behavior

The workflow should be able to complete normally and output JSON-formatted results.

❌ Actual Behavior

As long as all three nodes have ended and their status is checked, the workflow remains in the running state and does not output the final result.

Originally created by @D-siheng on GitHub (May 27, 2024). ### Self Checks - [X] This is only for bug report, if you would like to ask a question, please head to [Discussions](https://github.com/langgenius/dify/discussions/categories/general). - [X] I have searched for existing issues [search for existing issues](https://github.com/langgenius/dify/issues), including closed ones. - [X] I confirm that I am using English to submit this report (我已阅读并同意 [Language Policy](https://github.com/langgenius/dify/issues/1542)). - [X] Please do not modify this template :) and fill in all the required fields. ### Dify version 0.6.7 ### Cloud or Self Hosted Self Hosted (Docker) ### Steps to reproduce I wrote an built-in tool following the official guidelines and referencing the code of the Wiki search tool. The tool's function is to output document summaries using the document summary module of langchain. The tool can be displayed normally in the list of built-in tools, and after testing, it can run normally in workflow and output document summaries. I created a START-document summary-END workflow and published it. I ran the workflow in the `explore` interface, and all three nodes in the workflow displayed as completed, and the summary tool output the document summary content normally in the nodes. The end node also had normal input and output, but the workflow process was stuck in the running state and could not become the succeed state, which resulted in the summary result not being output on the screen. After testing, the situation where the workflow process could not end only occurred when a docx document was to be summarized. Besides the workflow, I also tried the chatflow, which consisted of STRAT-document summary-ANSWER three nodes. Running the chatflow could normally end and output formatted document summary content. My code is as follows: ```yaml # summarizer.yaml identity: author: DLH name: summarizer label: en_US: Summarizer zh_Hans: 文档摘要 pt_BR: Summarizer description: en_US: This tool is a documentation abstract of load_summarize_chain implementation using langchain zh_Hans: 该工具是使用langchain的load_summarize_chain实现的文档摘要 pt_BR: This tool is a documentation abstract of load_summarize_chain implementation using langchain icon: icon.svg credentials_for_provider: ``` ```python # summarizer.py from core.tools.errors import ToolProviderCredentialValidationError from core.tools.provider.builtin_tool_provider import BuiltinToolProviderController from core.tools.provider.builtin.summarizer.tools.langchain_summarize import SummarizeTool class SummarizeProvider(BuiltinToolProviderController): def _validate_credentials(self, credentials: dict) -> None: try: SummarizeTool().fork_tool_runtime( meta={ "credentials": credentials, } ).invoke( user_id='', tool_parameters={ "File_Path": "file path", # 1、添加按ChunkSize大小对Docx文档进行分段功能 "ChunkSize": 0 }, ) except Exception as e: raise ToolProviderCredentialValidationError(str(e)) ``` ```yaml # langchain_summarize.yaml identity: author: DLH name: summarizer label: en_US: Summarizer zh_Hans: 文档摘要 pt_BR: Summarizer description: en_US: This tool is a documentation abstract of load_summarize_chain implementation using langchain zh_Hans: 该工具是使用langchain的load_summarize_chain实现的文档摘要 pt_BR: This tool is a documentation abstract of load_summarize_chain implementation using langchain icon: icon.svg credentials_for_provider: ``` ```python # langchain_summarize.py import os import glob import hashlib import tiktoken from langchain_community.llms import VLLMOpenAI from langchain.prompts import PromptTemplate # 1、添加按ChunkSize大小对Docx文档进行分段功能 from langchain.text_splitter import RecursiveCharacterTextSplitter from langchain.text_splitter import CharacterTextSplitter from langchain.chains.summarize import load_summarize_chain from langchain.docstore.document import Document from langchain_community.document_loaders import PyPDFLoader from langchain_community.document_loaders import Docx2txtLoader import logging import traceback from typing import Any, Optional, Union from core.tools.entities.tool_entities import ToolInvokeMessage from core.tools.tool.builtin_tool import BuiltinTool class SummarizeTool(BuiltinTool): # 文档摘要功能函数 def summarize_from_folder(self, folder, llm, ChunkSize): # 变量定义 summaries = [] logging.basicConfig(level=logging.DEBUG) logging.info('摘要函数 +——+ 开始执行') # 加载大模型 QWEN = VLLMOpenAI( max_tokens=10000, # 大模型的总tokens长度 temperature=0.7, # 温度，越高生成的内容复杂度就越高 openai_api_key="EMPTY", # openAI模型的APIKEY，因用的本地模型，因此设为EMPTY，如果使用联网模型则输入对应key值即可 openai_api_base="http://xx.xx.x.xx:8080/v1", # 使用vllm推理引擎加载模型，这里填入引擎对外提供的API接口，因隐私安全问题使用xx代替 model_name="/data/models/Qwen1.5-72B-Chat/" # 同上，填入对应模型的名称 ) # 文档读取 for file in glob.glob(folder + "/*"): # 获取文件名 file_name = os.path.basename(file) # pdf文档 if file.endswith(".pdf"): loader = PyPDFLoader(file) # 返回一个文档数组存入docs中，每个元素为一个页面内容 docs = loader.load_and_split() # docx文档 elif file.endswith(".docx"): docx_loader = Docx2txtLoader(file) # 1、添加按ChunkSize大小对Docx文档进行分段功能 if ChunkSize != 0: # 设置ChunkSize的值时，按ChunkSize大小对docx进行分段 text_splitter = RecursiveCharacterTextSplitter(chunk_size=ChunkSize, chunk_overlap=50) data = docx_loader.load_and_split(text_splitter) else: # 否则就按'/n/n'对docx进行分段 data = docx_loader.load() text_splitter = CharacterTextSplitter(chunk_size=512, chunk_overlap=50) docs = text_splitter.split_documents(data) else: return self.create_text_message("Unsupported file type. Only support docx & pdf.") # print(f"Unsupported file type: {file}.Only support docx & pdf.") # continue # 文档摘要生成 chain = load_summarize_chain(llm, chain_type="map_reduce", return_intermediate_steps=False) summary = chain.run(docs) # 创建一个包含文件名和摘要的字符串 file_summary_str = f"Summary of {file_name}:\n\n{summary}" # print(file_summary_str) # print("\n") summaries.append(file_summary_str) logging.info('摘要函数 +——+ 结束') return "\n\n".join(summaries) def _invoke(self, user_id: str, # 用户ID tool_parameters: dict[str, Any], # 工具参数 ) -> Union[ToolInvokeMessage, list[ToolInvokeMessage]]: """ invoke tools """ blobpath = "https://openaipublic.blob.core.windows.net/encodings/cl100k_base.tiktoken" cache_key = hashlib.sha1(blobpath.encode()).hexdigest() # 配置tiktoken的本地缓存目录 tiktoken_cache_dir = "/app/api" os.environ["TIKTOKEN_CACHE_DIR"] = tiktoken_cache_dir assert os.path.exists(os.path.join(tiktoken_cache_dir, cache_key)) folder = tool_parameters.get('File_Path', '') # 1、添加按ChunkSize大小对Docx文档进行分段功能 chunksize = tool_parameters.get('ChunkSize', 0) if not folder: return self.create_text_message('Please input folder') logging.info('invoke函数 执行') result = self.summarize_from_folder(folder, QWEN, chunksize) # result = "\n\n".join(results) with open(filename, 'w') as f: traceback.print_stack(file=f) logging.info('invoke函数 结束') return self.create_text_message(self.summary(user_id=user_id,content=result)) ``` ### ✔️ Expected Behavior The workflow should be able to complete normally and output JSON-formatted results. ### ❌ Actual Behavior As long as all three nodes have ended and their status is checked, the workflow remains in the running state and does not output the final result.
yindo added the 🐞 bug label 2026-02-21 18:01:31 -05:00
yindo closed this issue 2026-02-21 18:01:31 -05:00
Sign in to join this conversation.
No Branch/Tag Specified
Branches Tags
Labels
Clear labels
Knowledge Pipeline
Proposal
WIP
cloud
dependencies
feat:webapp
good first issue
outdated
performance
pull-request
Mirrored from GitHub Pull Request
 python:uv
refactor
size:L
stale
status: accepting prs
status: blocked
status: in discussion
stt & tts
vulnerablity
web
wont-fix
🌊 feat:workflow
🌚 invalid
🐞 bug
👻 feat:rag
💪 enhancement
📚 documentation
🔨 feat:tools
🙋‍♂️ question
🙏 help wanted
🤔 cant-reproduce
🤖 feat:agent
No Label 🐞 bug
Milestone
No items
No Milestone
Projects
Clear projects
No project
Assignees
Clear assignees
yindo
No Assignees
1 Participants
Notifications
Due Date
No due date set.
Dependencies

No dependencies set.

Reference: langgenius/dify#3743
Delete Branch "%!s()"

Deleting a branch is permanent. Although the deleted branch may continue to exist for a short time before it actually gets removed, it CANNOT be undone in most cases. Continue?