Write the Code. Change the World.

7月 11

AI 对话项目,会用到流式返回。一般客户端不会直接调用 AI 对话,而是调用自己的服务,自己的服务再调用 AI. 服务端可以通过 websocket 的方式来达到这个效果,也可以通过 SSE 的方式来达到这个效果。而 AI 也可能会提供这两种方式来提供服务。这样一来,项目就可能有四种组合方式来进行 AI 对话。如下:

客户端 服务端 AI
APP、浏览器 websocket websocket
APP、浏览器 websocket SSE
APP、浏览器 SSE websocket
APP、浏览器 SSE SSE

https://developer.mozilla.org/zh-CN/docs/Web/API/Server-sent_events/Using_server-sent_events

找 AI

websocket 阵营

https://www.xfyun.cn/doc/spark/Web.html#_1-%E6%8E%A5%E5%8F%A3%E8%AF%B4%E6%98%8E

SSE 阵营

https://agents.baidu.com/docs/develop/out-deployment/conversation/

尝试实现四种组合

从最简单的开始。服务端通过 SSE 的方式, AI 也是 SSE 的方式。

先封装一个 curl 工具包,创建 internal/pkg/curl/curl.go,内容如下。

package curl

import (
    "bytes"
    "context"
    "fmt"
    "io"
    "net/http"

    "github.com/gogf/gf/v2/frame/g"
    "github.com/gogf/gf/v2/net/gclient"
)

type HttpClient struct {
    client *gclient.Client
}

func NewHttpClient() *HttpClient {
    return &HttpClient{
        client: g.Client(),
    }
}

func (h *HttpClient) Get(ctx context.Context, url string, headers map[string]string) (string, error) {
    res, err := h.client.Get(ctx, url, headers)
    if err != nil {
        return "", err
    }
    defer res.Close()

    return res.ReadAllString(), nil
}

func (h *HttpClient) Post(ctx context.Context, url string, data interface{}, headers map[string]string) (string, error) {
    for k, v := range headers {
        h.client.SetHeader(k, v)
    }

    res, err := h.client.Post(ctx, url, data, headers)
    if err != nil {
        return "", err
    }
    defer res.Close()
    return res.ReadAllString(), nil
}

func (h *HttpClient) PostStream(ctx context.Context, url string, data interface{}, headers map[string]string) (io.ReadCloser, error) {
    // 将 data 转换为 io.Reader
    var body io.Reader
    switch v := data.(type) {
    case string:
        body = bytes.NewReader([]byte(v))
    case []byte:
        body = bytes.NewReader(v)
    default:
        return nil, fmt.Errorf("unsupported data type: %T", v)
    }

    // 创建 HTTP 请求
    req, err := http.NewRequestWithContext(ctx, "POST", url, body)
    if err != nil {
        return nil, err
    }

    // 设置请求头
    for k, v := range headers {
        req.Header.Set(k, v)
    }

    // 发起请求
    resp, err := http.DefaultClient.Do(req)
    if err != nil {
        return nil, err
    }

    // 检查响应状态码
    if resp.StatusCode != http.StatusOK {
        return nil, fmt.Errorf("request failed with status code: %d", resp.StatusCode)
    }

    return resp.Body, nil
}

现在来完善 chat 接口。

7月 11

AI 聊天,会用到流式返回的接口。用 goframe 框架尝试尝试。

开始

初始化项目

gf init stream.demo -u

cd stream.demo

gf run main.go

默认路由信息是这样的。

build: main.go
go build -o .\main.exe  main.go
.\main.exe
build running pid: 44364
2025-07-11T13:44:19.160+08:00 [INFO] pid[44364]: http server started listening on [:8000]
2025-07-11T13:44:19.160+08:00 [INFO] {ac79854f201c51188f16b244dba07b30} swagger ui is serving at address: http://127.0.0.1:8000/swagger/
2025-07-11T13:44:19.161+08:00 [INFO] {ac79854f201c51188f16b244dba07b30} openapi specification is serving at address: http://127.0.0.1:8000/api.json

  ADDRESS | METHOD |   ROUTE    |                           HANDLER                           |           MIDDLEWARE
----------|--------|------------|-------------------------------------------------------------|----------------------------------
  :8000   | ALL    | /api.json  | github.com/gogf/gf/v2/net/ghttp.(*Server).openapiSpec       |
----------|--------|------------|-------------------------------------------------------------|----------------------------------
  :8000   | GET    | /hello     | stream.demo/internal/controller/hello.(*ControllerV1).Hello | ghttp.MiddlewareHandlerResponse
----------|--------|------------|-------------------------------------------------------------|----------------------------------
  :8000   | ALL    | /swagger/* | github.com/gogf/gf/v2/net/ghttp.(*Server).swaggerUI         | HOOK_BEFORE_SERVE
----------|--------|------------|-------------------------------------------------------------|----------------------------------

提交版本

git init -b main
git add .
git commit -m 'initialize'

定义接口,创建控制器,定义路由

新建 api/chat/v1/chat.go 文件,定义以下请求和返回

package v1

import "github.com/gogf/gf/v2/frame/g"

type ChatReq struct {
    g.Meta  `path:"/chat" method:"post" tags:"聊天"  summary:"聊天请求"`
    Uid     uint64 `v:"required#用户ID不能为空" json:"uid" dc:"用户ID"`
    Message string `v:"required#消息内容不能为空" json:"message" dc:"消息内容"`
}

type ChatRes struct {
    IsEnd bool `json:"is_end" dc:"是否结束"`
}

执行 gf gen ctrl 生成控制器。修改 internal/cmd/cmd.go,增加路由配置。

import (
    ……
    "stream.demo/internal/controller/hello"
)

……
s.Group("/api", func(group *ghttp.RouterGroup) {
    group.Bind(
        chat.NewV1(),
    )
})

再运行 gf run main.go,把项目跑起来看看。发现 chat 路由已经好了。

build: main.go
go build -o .\main.exe  main.go
.\main.exe
build running pid: 44148
2025-07-11T14:13:29.676+08:00 [INFO] pid[44148]: http server started listening on [:8000]
2025-07-11T14:13:29.676+08:00 [INFO] {5065f8ccb71d5118823d646fdde40224} swagger ui is serving at address: http://127.0.0.1:8000/swagger/
2025-07-11T14:13:29.677+08:00 [INFO] {5065f8ccb71d5118823d646fdde40224} openapi specification is serving at address: http://127.0.0.1:8000/api.json

  ADDRESS | METHOD |   ROUTE    |                           HANDLER                           |           MIDDLEWARE
----------|--------|------------|-------------------------------------------------------------|----------------------------------
  :8000   | ALL    | /api.json  | github.com/gogf/gf/v2/net/ghttp.(*Server).openapiSpec       |
----------|--------|------------|-------------------------------------------------------------|----------------------------------
  :8000   | POST   | /api/chat      | stream.demo/internal/controller/chat.(*ControllerV1).Chat   | ghttp.MiddlewareHandlerResponse
----------|--------|------------|-------------------------------------------------------------|----------------------------------
  :8000   | GET    | /hello     | stream.demo/internal/controller/hello.(*ControllerV1).Hello | ghttp.MiddlewareHandlerResponse
----------|--------|------------|-------------------------------------------------------------|----------------------------------
  :8000   | ALL    | /swagger/* | github.com/gogf/gf/v2/net/ghttp.(*Server).swaggerUI         | HOOK_BEFORE_SERVE
----------|--------|------------|-------------------------------------------------------------|----------------------------------

开始流式 demo

编辑 internal/controller/chat/chat_v1_chat.go

package chat

import (
    "context"
    "fmt"
    "time"

    "github.com/gogf/gf/v2/net/ghttp"

    v1 "stream.demo/api/chat/v1"
)

func (c *ControllerV1) Chat(ctx context.Context, req *v1.ChatReq) (res *v1.ChatRes, err error) {
    r := ghttp.RequestFromCtx(ctx)
    r.Response.Header().Set("Content-Type", "text/event-stream")
    r.Response.Header().Set("Cache-Control", "no-cache")
    r.Response.Header().Set("Connection", "keep-alive")
    r.Response.Header().Set("Content-Encoding", "identity")

    for i := range 10 {
        data := fmt.Sprintf(`{"uid":%d, "message":"%s","index":%d}`, req.Uid, req.Message, i)
        r.Response.Writefln("data: %s\n\n", data)
        r.Response.Flush()
        time.Sleep(time.Millisecond * 1000)
    }

    fmt.Println("发送结束")
    res = &v1.ChatRes{}
    res.IsEnd = true
    return
}

重新运行项目。

测试

可以使用 curl 命令,可以使用 apifox 这种工具,也可以使用浏览器。 这里使用浏览器加 js 的方式来搞。

使用 chrome 打开 http://127.0.0.1:8000 ,然后打开控制台。输入以下测试代码,回车。

fetch("http://127.0.0.1:8000/api/chat", {
  method: "POST",
  headers: {
    "Content-Type": "application/json",
  },
  body: JSON.stringify({
    message: "舞低杨柳楼心月,歌尽桃花扇底风",
    uid: 1,
  }),
})
  .then(readStream)
  .catch(console.error);

async function readStream(res) {
  const reader = res.body.getReader();
  const decoder = new TextDecoder("utf-8");
  let buffer = "";

  while (true) {
    const { done, value } = await reader.read();
    if (done) break;
    buffer += decoder.decode(value, { stream: true });

    let lines = buffer.split("\n\n");
    buffer = lines.pop();

    for (const chunk of lines) {
      const dataLine = chunk.trim();
      if (dataLine.startsWith("data: ")) {
        const data = dataLine.slice(6);
        console.log(getNowTime(), "消息:", data);
      }
    }
  }
}

function getNowTime() {
  const date = new Date();
  return date.toLocaleTimeString("zh-CN", {
    hour12: false,
    timeZone: "Asia/Shanghai",
  });
}

截图如下:

提交版本控制。

git add .
git commit -m '增加 chat 接口,并测试流式响应'

nginx 处理

如果走 nginx 代理,需要处理关闭相应缓冲

location /api/ {
    proxy_pass http://你的后端服务地址;
    proxy_set_header Host $host;
    proxy_set_header X-Real-IP $remote_addr;
    proxy_set_header X-Forwarded-For $proxy_add_x_forwarded_for;
    proxy_buffering off;
    proxy_cache off;
    proxy_request_buffering off;
    chunked_transfer_encoding on;
}