Files
AIRouter/internal/server/server.go
T
freedakgmail 63eec59773
CI / lint (push) Has been cancelled
CI / test (push) Has been cancelled
CI / build (push) Has been cancelled
CI / security-scan (push) Has been cancelled
feat: 添加vLLM适配器、修复SSE流式输出和中间件问题
- 新增 vLLM 适配器 (internal/adapter/vllm.go),支持 OpenAI 兼容 API
- 修复 responseWriter 未实现 http.Flusher 导致 SSE 流式输出 500 错误
- 调整中间件顺序:BodyLimit 移至 Auth 之前,提前拒绝超大请求
- BodyLimit 增强:检查 Content-Length header
- 导出 Server.Authenticator() 方法供测试使用
- 配置更新:使用本地 deepseek-r1:1.5b 模型,新增 vllm-chat 逻辑模型
- 修复 handlers.go 中未使用的 target 参数 lint 警告
2026-08-03 07:52:50 +08:00

195 lines
5.1 KiB
Go

package server
import (
"context"
"fmt"
"net/http"
"os"
"path/filepath"
"strings"
"time"
"github.com/edgeai/gateway/internal/adapter"
"github.com/edgeai/gateway/internal/auth"
"github.com/edgeai/gateway/internal/config"
"github.com/edgeai/gateway/internal/handler"
"github.com/edgeai/gateway/internal/middleware"
"github.com/edgeai/gateway/internal/observability"
"github.com/edgeai/gateway/internal/router"
"github.com/edgeai/gateway/internal/scheduler"
"github.com/edgeai/gateway/internal/session"
)
// Server is the main HTTP server for the AI gateway.
type Server struct {
cfg *config.Config
logger *observability.Logger
metrics *observability.Metrics
HTTPSrv *http.Server
auth *auth.Authenticator
registry *adapter.Registry
modelMap *router.LogicalModelMapping
scheduler *scheduler.Scheduler
sessions *session.Store
}
// New creates a new Server instance with all components wired.
func New(cfg *config.Config, logger *observability.Logger) (*Server, error) {
// Ensure data directory exists
dbPath := extractDBPath(cfg.Storage.SessionDB)
if dbPath != "" {
os.MkdirAll(filepath.Dir(dbPath), 0755)
}
// Initialize auth
authPath := filepath.Join(filepath.Dir(dbPath), "auth.db")
authenticator, err := auth.NewAuthenticator(authPath, logger)
if err != nil {
return nil, fmt.Errorf("init auth: %w", err)
}
// Initialize session store
sessionStore, err := session.NewStore(dbPath)
if err != nil {
return nil, fmt.Errorf("init session store: %w", err)
}
// Initialize adapter registry
registry := adapter.NewRegistry()
// Initialize logical model mapping
modelMap := router.NewLogicalModelMapping(cfg)
// Register adapters for each unique endpoint
registered := make(map[string]bool)
for _, mc := range cfg.Models {
key := mc.Provider + "|" + mc.Endpoint
if !registered[key] {
switch mc.Provider {
case "ollama":
registry.Register(mc.Provider, adapter.NewOllamaAdapter(mc.Endpoint))
case "vllm":
registry.Register(mc.Provider, adapter.NewVLLMAdapter(mc.Endpoint))
}
registered[key] = true
}
}
// Initialize scheduler
sched := scheduler.NewScheduler(&cfg.Scheduler, logger)
// Initialize metrics
metrics := observability.NewMetrics()
s := &Server{
cfg: cfg,
logger: logger,
metrics: metrics,
auth: authenticator,
registry: registry,
modelMap: modelMap,
scheduler: sched,
sessions: sessionStore,
}
mux := http.NewServeMux()
s.registerRoutes(mux)
// Apply middleware chain (order: Recovery → Logging → RequestID → BodyLimit → Auth → handler)
h := middleware.RequestID(mux)
h = middleware.BodyLimit(cfg.Server.MaxRequestBodyMB)(h)
h = s.auth.Middleware(h)
h = middleware.Logging(logger)(h)
h = middleware.Recovery(logger)(h)
s.HTTPSrv = &http.Server{
Addr: fmt.Sprintf("%s:%d", cfg.Server.Host, cfg.Server.Port),
Handler: h,
ReadTimeout: 30 * time.Second,
WriteTimeout: 0, // no write timeout for SSE
IdleTimeout: 120 * time.Second,
}
return s, nil
}
func (s *Server) registerRoutes(mux *http.ServeMux) {
// Health and readiness
mux.HandleFunc("/health", s.handleHealth)
mux.HandleFunc("/ready", s.handleReady)
// Metrics
mux.HandleFunc(s.cfg.Observability.MetricsPath, s.metrics.Handler())
// OpenAI-compatible API
mux.HandleFunc("/v1/chat/completions", s.handleChatCompletions)
mux.HandleFunc("/v1/models", s.handleModels)
// Session management
mux.HandleFunc("/v1/sessions", s.handleSessions)
mux.HandleFunc("/v1/sessions/", s.handleSessionByID)
}
// Authenticator returns the authenticator instance (for testing/management).
func (s *Server) Authenticator() *auth.Authenticator {
return s.auth
}
// Start begins listening for HTTP requests.
func (s *Server) Start() error {
s.logger.Info("http server starting", observability.F().
Event("server_start").
Set("addr", s.HTTPSrv.Addr))
return s.HTTPSrv.ListenAndServe()
}
// Shutdown gracefully shuts down the server.
func (s *Server) Shutdown() error {
ctx, cancel := context.WithTimeout(context.Background(), 10*time.Second)
defer cancel()
s.scheduler.Stop()
if s.sessions != nil {
s.sessions.Close()
}
if s.auth != nil {
s.auth.Close()
}
return s.HTTPSrv.Shutdown(ctx)
}
func (s *Server) handleHealth(w http.ResponseWriter, r *http.Request) {
handler.WriteJSON(w, http.StatusOK, map[string]string{"status": "ok"})
}
func (s *Server) handleReady(w http.ResponseWriter, r *http.Request) {
w.Header().Set("Content-Type", "application/json")
ready := true
reasons := []string{}
for _, name := range s.registry.Names() {
a, _ := s.registry.Get(name)
if err := a.HealthCheck(r.Context()); err != nil {
ready = false
reasons = append(reasons, fmt.Sprintf("%s: %v", name, err))
}
}
if ready {
w.WriteHeader(http.StatusOK)
w.Write([]byte(`{"status":"ready"}`))
} else {
w.WriteHeader(http.StatusServiceUnavailable)
fmt.Fprintf(w, `{"status":"not_ready","reasons":["%s"]}`, strings.Join(reasons, `","`))
}
}
func extractDBPath(connStr string) string {
if strings.HasPrefix(connStr, "sqlite://") {
return strings.TrimPrefix(connStr, "sqlite://")
}
return connStr
}