eaiadmin
89ae31c998
- 将 wechat_official_account 重命名为 weixin_public_account，符合中文命名规范 - 新增 DOCX 文档生成技能、聊天历史、请求 ID 中间件 - 增强工作流、热点服务、文章服务等模块功能 - 前端同步重命名组件和 API - 新增架构文档 AR13/AR14、专员文档更新 - 补充测试用例(seed_specialists_test, db_migration_test) Co-Authored-AI: yes
481 lines
15 KiB
Go
481 lines
15 KiB
Go
package ai
|
||
|
||
import (
|
||
"bufio"
|
||
"bytes"
|
||
"encoding/json"
|
||
"fmt"
|
||
"io"
|
||
"net/http"
|
||
"strings"
|
||
"time"
|
||
|
||
"eai_agentplatform/backend/internal/config"
|
||
)
|
||
|
||
// ──────────────────────────────────────────────
|
||
// 基础类型
|
||
// ──────────────────────────────────────────────
|
||
|
||
// Message OpenAI 兼容消息
|
||
type Message struct {
|
||
Role string `json:"role"`
|
||
Content string `json:"content"`
|
||
}
|
||
|
||
// ChatResult 非流式调用结果
|
||
type ChatResult struct {
|
||
Content string
|
||
Model string
|
||
Provider string
|
||
FinishReason string
|
||
Usage struct {
|
||
PromptTokens int `json:"prompt_tokens"`
|
||
CompletionTokens int `json:"completion_tokens"`
|
||
TotalTokens int `json:"total_tokens"`
|
||
}
|
||
}
|
||
|
||
// ──────────────────────────────────────────────
|
||
// OpenAI 兼容 HTTP 客户端
|
||
// ──────────────────────────────────────────────
|
||
|
||
// Client 低层 HTTP 客户端（基于 RouteConfig）
|
||
type Client struct {
|
||
baseURL string
|
||
apiKey string
|
||
model string
|
||
maxTokens int
|
||
temperature float64
|
||
hc *http.Client
|
||
}
|
||
|
||
// defaultHTTPTimeout 默认 HTTP 整体超时；路由可用 timeout_seconds 单独覆盖
|
||
const defaultHTTPTimeout = 60 * time.Second
|
||
|
||
// NewClient 从 RouteConfig 创建客户端
|
||
func NewClient(aiRoute *config.RouteConfig) *Client {
|
||
timeout := defaultHTTPTimeout
|
||
if aiRoute.TimeoutSeconds > 0 {
|
||
timeout = time.Duration(aiRoute.TimeoutSeconds) * time.Second
|
||
}
|
||
return &Client{
|
||
baseURL: strings.TrimRight(aiRoute.BaseURL, "/"),
|
||
apiKey: aiRoute.APIKey,
|
||
model: aiRoute.Model,
|
||
maxTokens: aiRoute.MaxTokens,
|
||
temperature: aiRoute.Temperature,
|
||
hc: &http.Client{Timeout: timeout},
|
||
}
|
||
}
|
||
|
||
func (c *Client) url(path string) string {
|
||
return c.baseURL + path
|
||
}
|
||
|
||
func (c *Client) headers() map[string]string {
|
||
h := map[string]string{"Content-Type": "application/json"}
|
||
if c.apiKey != "" {
|
||
h["Authorization"] = "Bearer " + c.apiKey
|
||
}
|
||
return h
|
||
}
|
||
|
||
func (c *Client) post(path string, body any) (*http.Response, error) {
|
||
j, err := json.Marshal(body)
|
||
if err != nil {
|
||
return nil, err
|
||
}
|
||
req, err := http.NewRequest(http.MethodPost, c.url(path), bytes.NewReader(j))
|
||
if err != nil {
|
||
return nil, err
|
||
}
|
||
for k, v := range c.headers() {
|
||
req.Header.Set(k, v)
|
||
}
|
||
resp, err := c.hc.Do(req)
|
||
if err != nil {
|
||
return nil, err
|
||
}
|
||
return resp, nil
|
||
}
|
||
|
||
// ──────────────────────────────────────────────
|
||
// 非流式调用
|
||
// ──────────────────────────────────────────────
|
||
|
||
// Generate 非流式对话，返回完整正文
|
||
func (c *Client) Generate(messages []Message) (string, error) {
|
||
resp, err := c.post("/chat/completions", map[string]any{
|
||
"model": c.model,
|
||
"messages": messages,
|
||
"stream": false,
|
||
"temperature": c.temperature,
|
||
"max_tokens": c.maxTokens,
|
||
})
|
||
if err != nil {
|
||
return "", fmt.Errorf("LLM 服务不可达: %w", err)
|
||
}
|
||
defer resp.Body.Close()
|
||
|
||
data, err := io.ReadAll(resp.Body)
|
||
if err != nil {
|
||
return "", fmt.Errorf("读取 LLM 响应失败: %w", err)
|
||
}
|
||
if resp.StatusCode != http.StatusOK {
|
||
return "", fmt.Errorf("LLM 返回 %d: %s", resp.StatusCode, truncate(string(data), 200))
|
||
}
|
||
|
||
var out struct {
|
||
Choices []struct {
|
||
Message struct {
|
||
Content string `json:"content"`
|
||
} `json:"message"`
|
||
} `json:"choices"`
|
||
}
|
||
if err := json.Unmarshal(data, &out); err != nil {
|
||
return "", fmt.Errorf("LLM 响应解析失败: %w", err)
|
||
}
|
||
if len(out.Choices) == 0 || strings.TrimSpace(out.Choices[0].Message.Content) == "" {
|
||
return "", fmt.Errorf("LLM 返回空正文")
|
||
}
|
||
return out.Choices[0].Message.Content, nil
|
||
}
|
||
|
||
// GenerateFull 非流式调用，返回完整 ChatResult（含 usage）
|
||
func (c *Client) GenerateFull(messages []Message) (*ChatResult, error) {
|
||
resp, err := c.post("/chat/completions", map[string]any{
|
||
"model": c.model,
|
||
"messages": messages,
|
||
"stream": false,
|
||
"temperature": c.temperature,
|
||
"max_tokens": c.maxTokens,
|
||
})
|
||
if err != nil {
|
||
return nil, fmt.Errorf("LLM 服务不可达: %w", err)
|
||
}
|
||
defer resp.Body.Close()
|
||
|
||
data, err := io.ReadAll(resp.Body)
|
||
if err != nil {
|
||
return nil, fmt.Errorf("读取 LLM 响应失败: %w", err)
|
||
}
|
||
if resp.StatusCode != http.StatusOK {
|
||
return nil, fmt.Errorf("LLM 返回 %d: %s", resp.StatusCode, truncate(string(data), 200))
|
||
}
|
||
|
||
var raw struct {
|
||
Model string `json:"model"`
|
||
Choices []struct {
|
||
Message struct {
|
||
Content string `json:"content"`
|
||
} `json:"message"`
|
||
FinishReason string `json:"finish_reason"`
|
||
} `json:"choices"`
|
||
Usage *struct {
|
||
PromptTokens int `json:"prompt_tokens"`
|
||
CompletionTokens int `json:"completion_tokens"`
|
||
TotalTokens int `json:"total_tokens"`
|
||
} `json:"usage"`
|
||
}
|
||
if err := json.Unmarshal(data, &raw); err != nil {
|
||
return nil, fmt.Errorf("LLM 响应解析失败: %w", err)
|
||
}
|
||
if len(raw.Choices) == 0 || strings.TrimSpace(raw.Choices[0].Message.Content) == "" {
|
||
return nil, fmt.Errorf("LLM 返回空正文")
|
||
}
|
||
|
||
result := &ChatResult{
|
||
Content: raw.Choices[0].Message.Content,
|
||
Model: raw.Model,
|
||
FinishReason: raw.Choices[0].FinishReason,
|
||
}
|
||
if raw.Usage != nil {
|
||
result.Usage.PromptTokens = raw.Usage.PromptTokens
|
||
result.Usage.CompletionTokens = raw.Usage.CompletionTokens
|
||
result.Usage.TotalTokens = raw.Usage.TotalTokens
|
||
}
|
||
return result, nil
|
||
}
|
||
|
||
// ──────────────────────────────────────────────
|
||
// 流式调用
|
||
// ──────────────────────────────────────────────
|
||
|
||
// GenerateStream SSE 流式调用，逐 chunk 回调
|
||
func (c *Client) GenerateStream(messages []Message, onChunk func(string)) error {
|
||
resp, err := c.post("/chat/completions", map[string]any{
|
||
"model": c.model,
|
||
"messages": messages,
|
||
"stream": true,
|
||
"temperature": c.temperature,
|
||
"max_tokens": c.maxTokens,
|
||
"stream_options": map[string]any{"include_usage": true},
|
||
})
|
||
if err != nil {
|
||
return fmt.Errorf("LLM 服务不可达: %w", err)
|
||
}
|
||
defer resp.Body.Close()
|
||
|
||
if resp.StatusCode != http.StatusOK {
|
||
data, _ := io.ReadAll(resp.Body)
|
||
return fmt.Errorf("LLM 返回 %d: %s", resp.StatusCode, truncate(string(data), 200))
|
||
}
|
||
|
||
// 流式空闲超时保护：若超过 idleTimeout 没有收到任何可消费 chunk，
|
||
// 判定为卡死并返回错误，避免前端无限等待（此前本地模型卡顿时会干等 120s）。
|
||
const idleTimeout = 45 * time.Second
|
||
|
||
scanner := bufio.NewScanner(resp.Body)
|
||
scanner.Buffer(make([]byte, 64*1024), 1024*1024)
|
||
scannerCh := make(chan string, 1)
|
||
// 驱动 scanner 的 goroutine，配合 select 实现空闲超时。
|
||
// 函数返回时 defer resp.Body.Close() 会让 scanner.Scan() 立即返回，
|
||
// goroutine 随之退出并通过 close 通知主循环（缓冲 channel 避免阻塞发送）。
|
||
go func() {
|
||
for scanner.Scan() {
|
||
scannerCh <- scanner.Text()
|
||
}
|
||
close(scannerCh)
|
||
}()
|
||
|
||
idleTimer := time.NewTimer(idleTimeout)
|
||
defer idleTimer.Stop()
|
||
for {
|
||
select {
|
||
case <-idleTimer.C:
|
||
return fmt.Errorf("LLM 流式响应空闲超时（%s 内无数据）", idleTimeout)
|
||
case line, ok := <-scannerCh:
|
||
if !ok {
|
||
return nil
|
||
}
|
||
if !strings.HasPrefix(line, "data:") {
|
||
continue
|
||
}
|
||
payload := strings.TrimSpace(strings.TrimPrefix(line, "data:"))
|
||
if payload == "[DONE]" {
|
||
return nil
|
||
}
|
||
var chunk struct {
|
||
Choices []struct {
|
||
Delta struct {
|
||
Content string `json:"content"`
|
||
} `json:"delta"`
|
||
} `json:"choices"`
|
||
}
|
||
if err := json.Unmarshal([]byte(payload), &chunk); err != nil {
|
||
continue
|
||
}
|
||
if len(chunk.Choices) > 0 && chunk.Choices[0].Delta.Content != "" {
|
||
// 收到有效内容，重置空闲计时
|
||
if !idleTimer.Stop() {
|
||
<-idleTimer.C
|
||
}
|
||
idleTimer.Reset(idleTimeout)
|
||
onChunk(chunk.Choices[0].Delta.Content)
|
||
}
|
||
}
|
||
}
|
||
}
|
||
|
||
// ──────────────────────────────────────────────
|
||
// Embedding
|
||
// ──────────────────────────────────────────────
|
||
|
||
// Embed 批量向量化
|
||
func (c *Client) Embed(inputs []string) ([][]float64, error) {
|
||
resp, err := c.post("/embeddings", map[string]any{
|
||
"model": c.model, "input": inputs,
|
||
})
|
||
if err != nil {
|
||
return nil, fmt.Errorf("embedding 服务不可达: %w", err)
|
||
}
|
||
defer resp.Body.Close()
|
||
|
||
data, err := io.ReadAll(resp.Body)
|
||
if err != nil {
|
||
return nil, fmt.Errorf("读取 embedding 响应失败: %w", err)
|
||
}
|
||
if resp.StatusCode != http.StatusOK {
|
||
return nil, fmt.Errorf("embedding 返回 %d: %s", resp.StatusCode, truncate(string(data), 200))
|
||
}
|
||
|
||
var out struct {
|
||
Data []struct {
|
||
Embedding []float64 `json:"embedding"`
|
||
} `json:"data"`
|
||
}
|
||
if err := json.Unmarshal(data, &out); err != nil {
|
||
return nil, fmt.Errorf("embedding 响应解析失败: %w", err)
|
||
}
|
||
res := make([][]float64, len(out.Data))
|
||
for i, d := range out.Data {
|
||
res[i] = d.Embedding
|
||
}
|
||
return res, nil
|
||
}
|
||
|
||
// ──────────────────────────────────────────────
|
||
// Fallback 链调用
|
||
// ──────────────────────────────────────────────
|
||
|
||
// buildRouteChain 构建「主路由 + 回退链」。
|
||
//
|
||
// primary 为 nil 时必须显式报错而不是继续：链里存 nil 会在循环里 NewClient(nil)，
|
||
// 而取 primary.RouteID 更是直接空指针。历史上 batch_extract / contract_review
|
||
// 就是直接传了 nil，导致这两个接口必 500。调用方应先用 config.GetRoute 解析路由。
|
||
func buildRouteChain(primary *config.RouteConfig) ([]*config.RouteConfig, error) {
|
||
if primary == nil {
|
||
return nil, fmt.Errorf("主路由为 nil：调用方未解析 AI 路由，请先用 config.GetRoute 取路由")
|
||
}
|
||
aiRouteChain := []*config.RouteConfig{primary}
|
||
if fallbacks, err := config.GetFallbackRoutes(primary.RouteID); err == nil && len(fallbacks) > 0 {
|
||
aiRouteChain = append(aiRouteChain, fallbacks...)
|
||
}
|
||
return aiRouteChain, nil
|
||
}
|
||
|
||
// GenerateWithFallback 依次尝试主路由 + 回退链，首个成功返回
|
||
func GenerateWithFallback(primary *config.RouteConfig, messages []Message) (string, error) {
|
||
aiRouteChain, err := buildRouteChain(primary)
|
||
if err != nil {
|
||
return "", err
|
||
}
|
||
|
||
var lastErr error
|
||
for _, aiRoute := range aiRouteChain {
|
||
client := NewClient(aiRoute)
|
||
start := time.Now()
|
||
content, err := client.Generate(messages)
|
||
if err == nil {
|
||
config.ReportRouteHealth(config.RouteHealth{
|
||
AIRouteID: aiRoute.RouteID,
|
||
Category: aiRoute.Category,
|
||
Healthy: true,
|
||
Checked: true,
|
||
LatencyMs: time.Since(start).Milliseconds(),
|
||
LastCheckedAt: time.Now(),
|
||
})
|
||
return content, nil
|
||
}
|
||
config.ReportRouteHealth(config.RouteHealth{
|
||
AIRouteID: aiRoute.RouteID,
|
||
Category: aiRoute.Category,
|
||
Healthy: false,
|
||
Checked: true,
|
||
LatencyMs: time.Since(start).Milliseconds(),
|
||
LastCheckedAt: time.Now(),
|
||
LastError: err.Error(),
|
||
})
|
||
lastErr = fmt.Errorf("[%s] %w", aiRoute.RouteID, err)
|
||
}
|
||
return "", fmt.Errorf("所有路由均失败: %w", lastErr)
|
||
}
|
||
|
||
// GenerateFullWithFallback 非流式 + 回退链：返回结果与「实际命中」的路由（用于审计）
|
||
func GenerateFullWithFallback(primary *config.RouteConfig, messages []Message) (*ChatResult, *config.RouteConfig, error) {
|
||
aiRouteChain, err := buildRouteChain(primary)
|
||
if err != nil {
|
||
return nil, nil, err
|
||
}
|
||
|
||
var lastErr error
|
||
for _, aiRoute := range aiRouteChain {
|
||
client := NewClient(aiRoute)
|
||
start := time.Now()
|
||
res, err := client.GenerateFull(messages)
|
||
if err == nil {
|
||
config.ReportRouteHealth(config.RouteHealth{
|
||
AIRouteID: aiRoute.RouteID,
|
||
Category: aiRoute.Category,
|
||
Healthy: true,
|
||
Checked: true,
|
||
LatencyMs: time.Since(start).Milliseconds(),
|
||
LastCheckedAt: time.Now(),
|
||
})
|
||
res.Provider = aiRoute.Provider
|
||
return res, aiRoute, nil
|
||
}
|
||
config.ReportRouteHealth(config.RouteHealth{
|
||
AIRouteID: aiRoute.RouteID,
|
||
Category: aiRoute.Category,
|
||
Healthy: false,
|
||
Checked: true,
|
||
LatencyMs: time.Since(start).Milliseconds(),
|
||
LastCheckedAt: time.Now(),
|
||
LastError: err.Error(),
|
||
})
|
||
lastErr = fmt.Errorf("[%s] %w", aiRoute.RouteID, err)
|
||
}
|
||
return nil, nil, fmt.Errorf("所有路由均失败: %w", lastErr)
|
||
}
|
||
|
||
// GenerateStreamWithFallback 流式 + 回退链：返回实际命中的路由
|
||
func GenerateStreamWithFallback(primary *config.RouteConfig, messages []Message, onChunk func(string)) (*config.RouteConfig, error) {
|
||
aiRouteChain, err := buildRouteChain(primary)
|
||
if err != nil {
|
||
return nil, err
|
||
}
|
||
|
||
var lastErr error
|
||
for _, aiRoute := range aiRouteChain {
|
||
if requiresAPIKey(aiRoute) && strings.TrimSpace(aiRoute.APIKey) == "" {
|
||
config.ReportRouteHealth(config.RouteHealth{
|
||
AIRouteID: aiRoute.RouteID,
|
||
Category: aiRoute.Category,
|
||
Healthy: false,
|
||
Checked: true,
|
||
LastCheckedAt: time.Now(),
|
||
LastError: "未配置 API Key",
|
||
})
|
||
lastErr = fmt.Errorf("[%s] 未配置 API Key", aiRoute.RouteID)
|
||
continue
|
||
}
|
||
client := NewClient(aiRoute)
|
||
start := time.Now()
|
||
if err := client.GenerateStream(messages, onChunk); err == nil {
|
||
config.ReportRouteHealth(config.RouteHealth{
|
||
AIRouteID: aiRoute.RouteID,
|
||
Category: aiRoute.Category,
|
||
Healthy: true,
|
||
Checked: true,
|
||
LatencyMs: time.Since(start).Milliseconds(),
|
||
LastCheckedAt: time.Now(),
|
||
})
|
||
return aiRoute, nil
|
||
} else {
|
||
config.ReportRouteHealth(config.RouteHealth{
|
||
AIRouteID: aiRoute.RouteID,
|
||
Category: aiRoute.Category,
|
||
Healthy: false,
|
||
Checked: true,
|
||
LatencyMs: time.Since(start).Milliseconds(),
|
||
LastCheckedAt: time.Now(),
|
||
LastError: err.Error(),
|
||
})
|
||
lastErr = fmt.Errorf("[%s] %w", aiRoute.RouteID, err)
|
||
}
|
||
}
|
||
return nil, fmt.Errorf("所有路由均失败: %w", lastErr)
|
||
}
|
||
|
||
func requiresAPIKey(aiRoute *config.RouteConfig) bool {
|
||
if aiRoute == nil {
|
||
return false
|
||
}
|
||
baseURL := strings.ToLower(strings.TrimSpace(aiRoute.BaseURL))
|
||
if strings.Contains(baseURL, "openrouter.ai") || strings.Contains(baseURL, "openai.com") {
|
||
return true
|
||
}
|
||
provider := strings.ToLower(strings.TrimSpace(aiRoute.Provider))
|
||
return provider == "openrouter" || provider == "openai"
|
||
}
|
||
|
||
func truncate(s string, n int) string {
|
||
r := []rune(s)
|
||
if len(r) <= n {
|
||
return s
|
||
}
|
||
return string(r[:n]) + "..."
|
||
}
|