Files
openteam/server/internal/api/router.go
T
Sakurasan 8c12207478 perf: 目录缓存与流式正文快路径,请求热路径开销减半
按性能分析逐项落地,实测对比(16 并发、mock 上游零延迟、SQLite):

- 每请求 SQL 10 → 2:模型/渠道目录(模型行、绑定、渠道行)进程内缓存 10s,
  运行期状态(冷却、健康)仍每次实时判定;管理端写操作经中间件、
  健康状态变化经回调立即失效。新增 TestProxyRequestSQLBudget 钉住预算。
- 端到端 p50 1.3–1.6ms → 0.66–0.70ms,RPS 3.5–8.9k → 16–18k。
- 流式正文累积:字符串 += 改 strings.Builder(原为 O(n²)),
  逐行 JSON 提取改字节快路径,并加廉价的子串预过滤、拿到输出 usage 后停累积。
  500 块:949µs/7.9MB 分配 → 83µs/136KB。
- 请求体定点改写:模型映射从整包 unmarshal/marshal 改为顶层字段替换
  (100KB:585µs → 29µs),stream_options 注入同理;requestText 改为
  只在「上游没报用量」时按需提取,亲和键改用原始体前缀。
- 渠道满载背压:新增 Acquire 有界排队(OT_PROXY_CHANNEL_ACQUIRE_WAIT,
  默认 150ms);排队仍满时返回 503 + Retry-After 而不是 502。
  实测渠道上限 4、16 并发:无排队 81 次 503;排队 150ms 后 400/400 成功。
- 记账按批聚合:同批同用户一次扣款、流水批量插入、日表按 (user,model,日期)
  聚合,替代每行 5 条语句;幂等仍靠 ref_id 唯一索引兜底。
- gorm 开启 PrepareStmt(SQL 解析此前占 CPU 15%);tiktoken 编码器按模型复用。
- 管理端用量列表不再返回原始请求/响应体(可能几十 MB),
  改为 has_raw 标记 + GET /admin/usage/:id 按需拉取,前端弹窗打开时取。

验证:go vet、go test ./...、go test -race(proxy/channel/usage/tokenizer)、
vue-tsc、eslint、vite build 全绿;本地起 mock 上游实测列表/详情接口与
管理页弹窗按需加载。
2026-10-30 14:14:21 +08:00

154 lines
5.5 KiB
Go
Raw Blame History

This file contains ambiguous Unicode characters
This file contains Unicode characters that might be confused with other characters. If you think that this is intentional, you can safely ignore this warning. Use the Escape button to reveal them.
package api
import (
"net/http"
"os"
"strings"
"github.com/gin-gonic/gin"
"github.com/openteam/server/internal/api/middleware"
"github.com/openteam/server/internal/app"
"github.com/openteam/server/internal/proxy"
)
// NewRouter 装配所有路由:
// - /v1/* 代理端点(Bearer API Key,OpenAI 格式错误体)
// - /api/v1/* 管理 API(会话 JWT)
func NewRouter(a *app.App, gw *proxy.Gateway) *gin.Engine {
if a.Cfg.Env == "production" {
gin.SetMode(gin.ReleaseMode)
}
r := gin.New()
// 只有显式配置了反代列表才信任 X-Forwarded-For;否则 ClientIP 退回 RemoteAddr,
// 避免伪造头把认证限速绕开。
if len(a.Cfg.TrustedProxies) > 0 {
if err := r.SetTrustedProxies(a.Cfg.TrustedProxies); err != nil {
panic("invalid trusted proxies: " + err.Error())
}
}
r.Use(gin.Logger(), gin.Recovery(), middleware.CORS(a.Cfg.CORSAllowOrigins), middleware.SecurityHeaders())
h := NewHandler(a, gw, a.Passkeys)
// --- 代理端点(对外)---
proxyGroup := r.Group("/v1", middleware.BodyLimit(int64(a.Cfg.Proxy.MaxBodyMB)*1024*1024))
{
proxyGroup.Any("/chat/completions", gw.Auth, gw.Handle)
proxyGroup.Any("/responses", gw.Auth, gw.Handle)
proxyGroup.Any("/messages", gw.Auth, gw.Handle)
proxyGroup.Any("/models", gw.Auth, gw.Handle)
}
// 静态资源(前端构建产物,存在时托管)
const dist = "web/dist"
if _, err := os.Stat(dist); err == nil {
// /assets 文件名含内容 hash,不可变:长缓存 + immutable,避免每次回源
assets := r.Group("/assets")
assets.Use(func(c *gin.Context) {
c.Header("Cache-Control", "public, max-age=31536000, immutable")
c.Next()
})
assets.Static("", dist+"/assets")
r.StaticFile("/favicon.svg", dist+"/favicon.svg")
}
// 未匹配路由:/v1/* 走代理鉴权;其余回退 SPA 或 404
r.NoRoute(func(c *gin.Context) {
if strings.HasPrefix(c.Request.URL.Path, "/v1") {
gw.Auth(c)
if !c.IsAborted() {
gw.Handle(c)
}
return
}
if _, err := os.Stat(dist); err == nil {
// SPA 入口:必须每次回源校验,否则浏览器启发式缓存会保留引用已删除 chunk 的旧 HTML → 黑屏
c.Header("Cache-Control", "no-cache")
c.File(dist + "/index.html")
return
}
c.JSON(http.StatusNotFound, gin.H{"error": "not found"})
})
// --- 管理 API ---
// 请求体上限 1MB:全部接口都是小 JSON(配置/密钥/用户),未认证面(login/register)
// 不给大 body 的内存放大空间。
api := r.Group("/api/v1", middleware.BodyLimit(1<<20))
{
auth := api.Group("/auth")
{
auth.POST("/register", h.Register)
auth.POST("/login", h.Login)
auth.POST("/refresh", h.Refresh)
auth.POST("/logout", h.Logout)
auth.POST("/password", middleware.SessionAuth(a), h.ChangePassword)
auth.GET("/me", middleware.SessionAuth(a), h.Me)
}
webauthn := api.Group("/webauthn")
{
webauthn.POST("/register/begin", middleware.SessionAuth(a), h.PasskeyRegisterBegin)
webauthn.POST("/register/complete", middleware.SessionAuth(a), h.PasskeyRegisterComplete)
webauthn.POST("/login/begin", h.PasskeyLoginBegin)
webauthn.POST("/login/complete", h.PasskeyLoginComplete)
webauthn.GET("/passkeys", middleware.SessionAuth(a), h.PasskeyList)
webauthn.DELETE("/passkeys/:id", middleware.SessionAuth(a), h.PasskeyDelete)
}
user := api.Group("", middleware.SessionAuth(a))
{
user.GET("/user/profile", h.UserProfile)
user.GET("/user/balance", h.UserBalance)
user.GET("/user/models", h.UserModels)
user.GET("/usage/summary", h.UsageSummary)
user.GET("/usage/stats", h.UsageStats)
user.GET("/usage/logs", h.UsageLogs)
user.POST("/keys", h.CreateKey)
user.GET("/keys", h.ListKeys)
user.PATCH("/keys/:id", h.PatchKey)
user.DELETE("/keys/:id", h.DeleteKey)
}
admin := api.Group("/admin", middleware.SessionAuth(a), middleware.AdminOnly, h.InvalidateCatalogOnWrite)
{
// 用户
admin.GET("/users", h.AdminUsers)
admin.PATCH("/users/:id", h.AdminPatchUser)
admin.POST("/users/:id/balance", h.AdminAdjustBalance)
// 渠道
admin.GET("/channels", h.AdminChannels)
admin.GET("/channels/stats", h.AdminChannelStats)
admin.POST("/channels", h.AdminCreateChannel)
admin.PUT("/channels/:id", h.AdminUpdateChannel)
admin.DELETE("/channels/:id", h.AdminDeleteChannel)
admin.POST("/channels/:id/test", h.AdminTestChannel)
admin.POST("/channels/:id/recover", h.AdminRecoverChannel)
admin.GET("/channels/:id/models/remote", h.AdminChannelRemoteModels)
admin.GET("/channels/:id/models", h.AdminChannelModels)
admin.POST("/channels/:id/models", h.AdminChannelAddModel)
admin.PATCH("/channels/:id/models/:bid", h.AdminChannelUpdateModel)
admin.DELETE("/channels/:id/models/:bid", h.AdminChannelDeleteModel)
// 模型与定价
admin.GET("/models", h.AdminModels)
admin.DELETE("/models/unused", h.AdminDeleteUnusedModels)
admin.POST("/models", h.AdminCreateModel)
admin.PUT("/models/:id", h.AdminUpdateModel)
admin.DELETE("/models/:id", h.AdminDeleteModel)
admin.POST("/models/:id/bindings", h.AdminCreateModelBinding)
admin.DELETE("/models/:id/bindings/:bid", h.AdminDeleteModelBinding)
// 统计与用量
admin.GET("/stats/overview", h.AdminStatsOverview)
admin.GET("/usage", h.AdminUsage)
admin.GET("/usage/:id", h.AdminUsageDetail)
// 配置
admin.GET("/config", h.AdminConfig)
admin.PUT("/config", h.AdminPutConfig)
// 充值审核(M6 预留)
}
}
r.GET("/healthz", func(c *gin.Context) {
c.JSON(http.StatusOK, gin.H{"status": "ok"})
})
return r
}