Dify介绍

Dify 是一款开源的大语言模型(LLM) 应用开发平台。它融合了后端即服务（Backend as Service）和 LLMOps 的理念，使开发者可以快速搭建生产级的生成式 AI 应用。即使你是非技术人员，也能参与到 AI 应用的定义和数据运营过程中。

问题1:docker_api_1 、docker_worker_1 服务和docker_sandbox_1 组件一直处于restarting

docker_api_1           /bin/bash /entrypoint.sh         Restarting
docker_db_1            docker-entrypoint.sh postg ...   Up (healthy)   5432/tcp
docker_nginx_1         sh -c cp /docker-entrypoin ...   Up             0.0.0.0:444->443/tcp,:::444->443/tcp, 0.0.0.0:7002->80/tcp,:::7002->80/tcp
docker_plugin_daemon_1 /bin/bash -c /app/entrypoi ...   Up          0.0.0.0:5003->5003/tcp    
docker_redis_1         docker-entrypoint.sh redis ...   Up (healthy)   6379/tcp
docker_sandbox_1       /main                            Restarting
docker_ssrf_proxy_1    sh -c cp /docker-entrypoin ...   Up             3128/tcp
docker_weaviate_1      /bin/weaviate --host 0.0.0 ...   Up
docker_web_1           /bin/sh ./entrypoint.sh          Up             3000/tcp
docker_worker_1        /bin/bash /entrypoint.sh         Restarting

刚开始按照 https://github.com/langgenius/dify/discussions/12312 将docker-compose.yaml 配置文件中LOG_FILE: ${LOG_FILE:-/app/logs/server.log} 调整为LOG_FILE: /app/logs/server.log 然后重启后，docker_api_1 暂时恢复正常，另外两个仍然异常。（为何是暂时，因为后面docker_api_1 又处于restarting状态）

这时查看服务日志

docker logs docker_sandbox_1
docker logs docker_worker_1

错误内容提示：

OpenBLAS blas_thread_init: pthread_create failed for thread 2 of 4: Operation not permitted
runtime/cgo: pthread_create failed :Operation not permitted

解决方法：添加特权运行配置

推测是容器权限不够，准备在docker-compose.yaml中添加特权运行配置内容。需要将docker_api_1 、docker_worker_1 docker_sandbox_1 都添加特权运行。

docker_api_1

  api:image: langgenius/dify-api:1.4.1restart: alwaysprivileged: true

docker_worker_1

  worker:image: langgenius/dify-api:1.4.1restart: alwaysprivileged: true

docker_sandbox_1

  sandbox:image: langgenius/dify-sandbox:0.2.12restart: alwaysprivileged: true

然后重新启动

docker compose down
docker compose up -d

重新查看服务都起来了，若是其他服务查看日志也是权限问题，按照此方法添加特权运行即可。

问题2：安装模型供应商提示【init environment for plugin langgenius/openai:0.0.26 failed too many times you should consider the package is corrupted or your network is unstable】

不论是在线安装还是离线安装都是安装失败。

查看docker_plugin_daemon_1 组件日志，

docker logs docker_plugin_daemon_1 

错误显示 operation not permitted，还是权限问题

解决方法: docker_plugin_daemon 添加特权配置内容

在docker-compose.yaml 对docker_plugin_daemon_1 添加特权运行配置

  plugin_daemon:image: langgenius/dify-plugin-daemon:0.1.1-localrestart: alwaysprivileged: true

然后重启

docker-compose down  
docker-compose up -d

这时重新安装还是失败，重新查看ocker_plugin_daemon_1 日志，提示

[ERROR]init environment failed: failed to install dependencies: exit status 2, output: error: Failed to unzip wheel: openai-1.64.0-py3-none-any.whlCaused by: an upstream reader returned an error: io error occurred: HTTP status client error (403 Forbidden) for url (https://pypi.tuna.tsinghua.edu.cn/packages/9a/e6d503e1ea474522b2a15a904bb/openai-1.64.0-py3-none-any.whl)Caused by: io error occurred: HTTP status client error (403 Forbidden) for url (https://pypi.tuna.tsinghua.edu.cn/packages/9a/1a/e62718f311daa26d208800976d7944e5ee64.0-py3-none-any.whl)Caused by: HTTP status client error (403 Forbidden) for url (https://pypi.tuna.tsinghua.edu.cn/packages/9a/1a/e62718f311daa26d208800976d7944e5ee6d503e1ea474522b2a15l)

之前修改过.env 配置的 PIP_MIRROR_URL 内容，看报错是这个配置导致的，调整pip源

解决方法：修改 .env配置文件中 PIP_MIRROR_URL值

PIP_MIRROR_URL 设置为：https://mirrors.aliyun.com/pypi/simple/

然后重新启动Dify

docker compose down
docker compose up -d

这时安装模型供应商插件安装成功