Dify2OpenAI 运行流程详解
项目概述
Dify2OpenAI是一个网关服务，将Dify应用程序转换为OpenAI API兼容的接口，使开发者能够使用熟悉的OpenAI API格式来访问Dify的LLM、知识库、工具和工作流程能力。
整体架构
flowchart TB
Client([客户端])<-->|OpenAI格式请求/响应|Gateway
subgraph Gateway [Dify2OpenAI网关服务]
direction TB
style Gateway fill:#f5f5f5,stroke:#333,stroke-width:1px
Parser([配置解析器]):::component
ChatHandler([聊天处理器]):::component
CompletionHandler([补全处理器]):::component
WorkflowHandler([工作流处理器]):::component
ImageProcessor([图像处理器]):::component
Logger([日志系统]):::component
end
Gateway<-->|Dify格式请求/响应|DifyAPI([Dify API服务])
classDef component fill:#e1f5fe,stroke:#01579b,stroke-width:1px,color:#01579b
classDef external fill:#f0f4c3,stroke:#827717,stroke-width:1px,color:#827717
class Client,DifyAPI external
请求处理流程
sequenceDiagram
participant client as 客户端
participant gateway as Dify2OpenAI网关
participant parser as 配置解析器
participant selector as 处理器选择器
participant chat as 聊天处理器
participant completion as 补全处理器
participant workflow as 工作流处理器
participant image as 图像处理器
participant dify as Dify API
rect rgb(240, 248, 255)
note right of client: 请求处理阶段
client->>+gateway: 发送OpenAI格式请求
gateway->>+parser: 解析Authorization和model参数
parser-->>-gateway: 返回解析后的配置
gateway->>+selector: 根据BOT_TYPE选择处理器
end
rect rgb(255, 248, 240)
note right of selector: 分发到不同处理器
alt 聊天机器人(Chat)
selector->>+chat: 处理请求
chat->>+image: 处理图像内容(如有)
image-->>-chat: 返回处理后的图像
chat->>dify: 发送转换后的请求
deactivate chat
else 补全机器人(Completion)
selector->>+completion: 处理请求
completion->>+image: 处理图像内容(如有)
image-->>-completion: 返回处理后的图像
completion->>dify: 发送转换后的请求
deactivate completion
else 工作流机器人(Workflow)
selector->>+workflow: 处理请求
workflow->>+image: 处理图像内容(如有)
image-->>-workflow: 返回处理后的图像
workflow->>dify: 发送转换后的请求
deactivate workflow
end
deactivate selector
end
rect rgb(240, 255, 240)
note right of dify: 响应处理阶段
dify-->>gateway: 返回处理结果
gateway->>-client: 转换为OpenAI格式并返回
end
配置解析逻辑
系统支持三种不同的配置模式，用于灵活适应不同的集成场景：
flowchart TD
Start([开始解析]):::start --> CheckAuth{检查Authorization Header}
CheckAuth -->|格式无效| Error([返回错误]):::error
CheckAuth -->|包含多个参数| Method1[方式一:<br>全部在Authorization中]:::method
Method1 --> Extract1[提取DIFY_API_URL、<br>API_KEY、BOT_TYPE等]:::process
Extract1 --> Validate
CheckAuth -->|单一参数| CheckParam{检查model参数}
CheckParam -->|无效| Error
CheckParam -->|有效| CheckValue{检查Authorization值}
CheckValue -->|不含http| Method2[方式二:<br>Authorization是API_KEY]:::method
Method2 --> Extract2[从model中提取其他参数]:::process
Extract2 --> Validate
CheckValue -->|含http| Method3[方式三:<br>Authorization是DIFY_API_URL]:::method
Method3 --> Extract3[从model中提取其他参数]:::process
Extract3 --> Validate
Validate[验证必要参数]:::process --> Return([返回解析后的配置]):::end
classDef start fill:#d4edda,stroke:#28a745,stroke-width:2px,color:#28a745
classDef process fill:#e8f4f8,stroke:#4b8bb3,stroke-width:1px
classDef method fill:#fff3cd,stroke:#856404,stroke-width:1px,color:#856404
classDef error fill:#f8d7da,stroke:#dc3545,stroke-width:2px,color:#dc3545
classDef end fill:#d4edda,stroke:#28a745,stroke-width:2px,color:#28a745
三种不同的接入方式
方式一：所有配置在Authorization Header中
使用方式：
Authorization: Bearer DIFY_API_URL|API_KEY|BOT_TYPE|INPUT_VARIABLE|OUTPUT_VARIABLE
方式二：Authorization Header传递API_KEY
使用方式：
Authorization: Bearer API_KEY
"model": "dify|BOT_TYPE|DIFY_API_URL|INPUT_VARIABLE|OUTPUT_VARIABLE"
方式三：Authorization Header传递DIFY_API_URL
使用方式：
Authorization: Bearer DIFY_API_URL
"model": "dify|API_KEY|BOT_TYPE|INPUT_VARIABLE|OUTPUT_VARIABLE"
消息处理流程
flowchart TB
Start([接收请求]):::start --> Parse[解析消息结构]:::process
Parse --> ScanImages[扫描图像内容]:::process
ScanImages --> CheckType{图像类型检查}
CheckType -->|Base64数据| Upload[上传到Dify]:::upload
CheckType -->|远程URL| Direct[直接使用URL]:::url
Upload --> BuildReq[构建Dify请求]:::process
Direct --> BuildReq
Parse --> ExtractText[提取文本内容]:::process
ExtractText --> BuildReq
BuildReq --> CheckMode{是否流式模式}
CheckMode -->|是| StreamMode[设置流式响应]:::stream
CheckMode -->|否| BlockMode[设置阻塞响应]:::block
StreamMode --> SendReq[发送到Dify]:::send
BlockMode --> SendReq
SendReq --> ReceiveResp[接收Dify响应]:::receive
ReceiveResp --> Convert[转换为OpenAI格式]:::process
Convert --> Return([返回给客户端]):::end
classDef start fill:#d4edda,stroke:#28a745,stroke-width:2px,color:#28a745
classDef process fill:#e8f4f8,stroke:#4b8bb3,stroke-width:1px
classDef upload fill:#e6c3e6,stroke:#8e44ad,stroke-width:1px,color:#8e44ad
classDef url fill:#c3e6cb,stroke:#155724,stroke-width:1px,color:#155724
classDef stream fill:#b8daff,stroke:#004085,stroke-width:1px,color:#004085
classDef block fill:#ffeeba,stroke:#856404,stroke-width:1px,color:#856404
classDef send fill:#d6d8d9,stroke:#1b1e21,stroke-width:1px,color:#1b1e21
classDef receive fill:#f5c6cb,stroke:#721c24,stroke-width:1px,color:#721c24
classDef end fill:#d4edda,stroke:#28a745,stroke-width:2px,color:#28a745
图像处理逻辑
flowchart TB
Start([检测到图像]):::start --> CheckType{图像类型?}
subgraph Base64处理 [Base64图像处理流程]
direction TB
style Base64处理 fill:#f0f0ff,stroke:#8080ff,stroke-width:1px
ParseData[解析base64数据]:::process --> ExtractMime[提取MIME类型]:::process
ExtractMime --> CreateBuffer[创建文件Buffer]:::process
CreateBuffer --> CreateForm[创建FormData]:::process
CreateForm --> Upload[上传到Dify]:::upload
Upload --> GetFileId[获取文件ID]:::process
end
subgraph URL处理 [URL图像处理流程]
direction TB
style URL处理 fill:#f0fff0,stroke:#80ff80,stroke-width:1px
ExtractExt[提取文件扩展名]:::process --> DetermineType[确定文件类型]:::process
end
CheckType -->|Base64| Base64处理
CheckType -->|URL| URL处理
GetFileId --> AddToReq([添加到请求中]):::end
DetermineType --> AddToReq
classDef start fill:#d4edda,stroke:#28a745,stroke-width:2px,color:#28a745
classDef process fill:#e8f4f8,stroke:#4b8bb3,stroke-width:1px
classDef upload fill:#e6c3e6,stroke:#8e44ad,stroke-width:1px,color:#8e44ad
classDef end fill:#d4edda,stroke:#28a745,stroke-width:2px,color:#28a745
响应处理流程
sequenceDiagram
participant handler as 处理器
participant dify as Dify API
participant client as 客户端
rect rgb(240, 248, 255)
handler->>dify: 发送请求
end
alt 流式模式 (stream=true)
rect rgb(240, 255, 240)
dify-->>handler: 返回数据流
handler->>handler: 创建PassThrough流
loop 每个数据块
handler->>handler: 转换为OpenAI格式
handler->>client: 发送数据块
end
handler->>client: 发送结束标记 (data: [DONE])
end
else 阻塞模式 (stream=false)
rect rgb(255, 248, 240)
dify-->>handler: 返回完整响应
handler->>handler: 转换为OpenAI格式
handler->>client: 发送完整响应
end
end
日志系统
flowchart LR
ReqLog([请求日志]):::logtype --> Logger
RespLog([响应日志]):::logtype --> Logger
ApiLog([API调用日志]):::logtype --> Logger
ErrorLog([错误日志]):::logtype --> Logger
subgraph LogSystem [日志处理系统]
direction TB
style LogSystem fill:#f8f9fa,stroke:#6c757d,stroke-width:1px
Logger([日志系统]):::component --> Sanitize[敏感信息处理]:::process
Sanitize --> Output[日志输出]:::process
end
Output --> Console([控制台输出]):::output
Output --> Files([文件存储]):::output
Files --> Rotation[日志轮转]:::process
Rotation --> Cleanup[过期日志清理]:::process
classDef logtype fill:#e2f0d9,stroke:#548235,stroke-width:1px,color:#548235
classDef component fill:#deebf7,stroke:#4472c4,stroke-width:1px,color:#4472c4
classDef process fill:#e8f4f8,stroke:#4b8bb3,stroke-width:1px
classDef output fill:#fff2cc,stroke:#ed7d31,stroke-width:1px,color:#ed7d31
部署架构
flowchart TB
classDef deploy fill:#ede7f6,stroke:#673ab7,stroke-width:1px,color:#673ab7
classDef server fill:#e3f2fd,stroke:#2196f3,stroke-width:1px,color:#2196f3
classDef component fill:#e8f5e9,stroke:#4caf50,stroke-width:1px,color:#4caf50
classDef external fill:#fff3e0,stroke:#ff9800,stroke-width:1px,color:#ff9800
Local([本地部署]):::deploy --> PM2([PM2进程管理]):::server
PM2 --> Server([Express服务器]):::server
Cloud([云部署]):::deploy --> Vercel([Vercel平台]):::server
Vercel --> ServerlessFunc([无服务器函数]):::server
subgraph 处理层
direction TB
style 处理层 fill:#f5f5f5,stroke:#333,stroke-width:1px
Server --> Handler([请求处理器]):::component
ServerlessFunc --> Handler
end
Handler --> DifyAPI([Dify API]):::external
总结
Dify2OpenAI网关服务通过巧妙的转换机制，使得开发者能够用统一的OpenAI API格式来调用Dify的各种能力，极大地简化了开发流程和集成难度。系统支持多种配置方式和机器人类型，能够灵活适应不同的应用场景需求。