feat(cache): 实现 RoutingCache 和 StatsBuffer 优化数据库写入

- 新增 RoutingCache 组件，使用 sync.Map 缓存 Provider 和 Model
- 新增 StatsBuffer 组件，使用 sync.Map + atomic.Int64 缓冲统计数据
- 扩展 StatsRepository.BatchUpdate 支持批量增量更新
- 改造 RoutingService/StatsService/ProviderService/ModelService 集成缓存
- 更新 usage-statistics spec，新增 routing-cache 和 stats-buffer spec
- 新增单元测试覆盖缓存命中/失效/并发场景

backend/cmd/server/main.go

@@ -67,13 +67,25 @@ func main() {
 	modelRepo := repository.NewModelRepository(db)
 	statsRepo := repository.NewStatsRepository(db)
 
-	// 5. 初始化 service 层
-	providerService := service.NewProviderService(providerRepo, modelRepo)
-	modelService := service.NewModelService(modelRepo, providerRepo)
-	routingService := service.NewRoutingService(modelRepo, providerRepo)
-	statsService := service.NewStatsService(statsRepo)
+	// 5. 初始化缓存
+	routingCache := service.NewRoutingCache(modelRepo, providerRepo, zapLogger)
+	if err := routingCache.Preload(); err != nil {
+		zapLogger.Warn("缓存预热失败，将使用懒加载", zap.Error(err))
+	}
 
-	// 6. 创建 ConversionEngine
+	// 6. 初始化统计缓冲
+	statsBuffer := service.NewStatsBuffer(statsRepo, zapLogger,
+		service.WithFlushInterval(5*time.Second),
+		service.WithFlushThreshold(100))
+	statsBuffer.Start()
+
+	// 7. 初始化 service 层
+	providerService := service.NewProviderService(providerRepo, modelRepo, routingCache)
+	modelService := service.NewModelService(modelRepo, providerRepo, routingCache)
+	routingService := service.NewRoutingService(routingCache)
+	statsService := service.NewStatsService(statsRepo, statsBuffer)
+
+	// 8. 创建 ConversionEngine
 	registry := conversion.NewMemoryRegistry()
 	if err := registry.Register(openai.NewAdapter()); err != nil {
 		zapLogger.Fatal("注册 OpenAI 适配器失败", zap.String("error", err.Error()))
@@ -83,16 +95,16 @@ func main() {
 	}
 	engine := conversion.NewConversionEngine(registry, zapLogger)
 
-	// 7. 初始化 provider client
+	// 9. 初始化 provider client
 	providerClient := provider.NewClient()
 
-	// 8. 初始化 handler 层
+	// 10. 初始化 handler 层
 	proxyHandler := handler.NewProxyHandler(engine, providerClient, routingService, providerService, statsService)
 	providerHandler := handler.NewProviderHandler(providerService)
 	modelHandler := handler.NewModelHandler(modelService)
 	statsHandler := handler.NewStatsHandler(statsService)
 
-	// 9. 创建 Gin 引擎
+	// 11. 创建 Gin 引擎
 	gin.SetMode(gin.ReleaseMode)
 	r := gin.New()
 
@@ -103,7 +115,7 @@ func main() {
 
 	setupRoutes(r, proxyHandler, providerHandler, modelHandler, statsHandler)
 
-	// 10. 启动服务器
+	// 12. 启动服务器
 	srv := &http.Server{
 		Addr:         formatAddr(cfg.Server.Port),
 		Handler:      r,
@@ -131,6 +143,8 @@ func main() {
 		zapLogger.Fatal("服务器强制关闭", zap.String("error", err.Error()))
 	}
 
+	statsBuffer.Stop()
+
 	zapLogger.Info("服务器已关闭")
 }
 

backend/internal/repository/stats_repo.go

@@ -11,5 +11,6 @@ import (
 // StatsRepository 统计数据仓库接口
 type StatsRepository interface {
 	Record(providerID, modelName string) error
+	BatchUpdate(providerID, modelName string, date time.Time, delta int) error
 	Query(providerID, modelName string, startDate, endDate *time.Time) ([]domain.UsageStats, error)
 }

backend/internal/repository/stats_repo_impl.go

@@ -43,6 +43,28 @@ func (r *statsRepository) Record(providerID, modelName string) error {
 	})
 }
 
+func (r *statsRepository) BatchUpdate(providerID, modelName string, date time.Time, delta int) error {
+	return r.db.Transaction(func(tx *gorm.DB) error {
+		var stats config.UsageStats
+		err := tx.Where("provider_id = ? AND model_name = ? AND date = ?",
+			providerID, modelName, date).First(&stats).Error
+
+		if errors.Is(err, gorm.ErrRecordNotFound) {
+			return tx.Create(&config.UsageStats{
+				ProviderID:   providerID,
+				ModelName:    modelName,
+				RequestCount: delta,
+				Date:         date,
+			}).Error
+		} else if err != nil {
+			return err
+		}
+
+		return tx.Model(&stats).
+			Update("request_count", gorm.Expr("request_count + ?", delta)).Error
+	})
+}
+
 func (r *statsRepository) Query(providerID, modelName string, startDate, endDate *time.Time) ([]domain.UsageStats, error) {
 	var stats []config.UsageStats
 	query := r.db.Model(&config.UsageStats{})

backend/internal/service/model_service_impl.go

@@ -11,27 +11,30 @@ import (
 type modelService struct {
 	modelRepo    repository.ModelRepository
 	providerRepo repository.ProviderRepository
+	cache        *RoutingCache
 }
 
-func NewModelService(modelRepo repository.ModelRepository, providerRepo repository.ProviderRepository) ModelService {
-	return &modelService{modelRepo: modelRepo, providerRepo: providerRepo}
+func NewModelService(modelRepo repository.ModelRepository, providerRepo repository.ProviderRepository, cache *RoutingCache) ModelService {
+	return &modelService{modelRepo: modelRepo, providerRepo: providerRepo, cache: cache}
 }
 
 func (s *modelService) Create(model *domain.Model) error {
-	// 校验供应商存在
 	if _, err := s.providerRepo.GetByID(model.ProviderID); err != nil {
 		return appErrors.ErrProviderNotFound
 	}
 
-	// 联合唯一校验：同一供应商下 model_name 不重复
 	if err := s.checkDuplicateModelName(model.ProviderID, model.ModelName, ""); err != nil {
 		return err
 	}
 
-	// 自动生成 UUID 作为 id
 	model.ID = uuid.New().String()
 	model.Enabled = true
-	return s.modelRepo.Create(model)
+	err := s.modelRepo.Create(model)
+	if err != nil {
+		return err
+	}
+	s.cache.SetModel(model)
+	return nil
 }
 
 func (s *modelService) Get(id string) (*domain.Model, error) {
@@ -47,20 +50,17 @@ func (s *modelService) ListEnabled() ([]domain.Model, error) {
 }
 
 func (s *modelService) Update(id string, updates map[string]interface{}) error {
-	// 获取当前模型
 	current, err := s.modelRepo.GetByID(id)
 	if err != nil {
 		return appErrors.ErrModelNotFound
 	}
 
-	// 如果更新 provider_id，校验新供应商存在
 	if providerID, ok := updates["provider_id"].(string); ok {
 		if _, err := s.providerRepo.GetByID(providerID); err != nil {
 			return appErrors.ErrProviderNotFound
 		}
 	}
 
-	// 确定更新后的 provider_id 和 model_name
 	newProviderID := current.ProviderID
 	if v, ok := updates["provider_id"].(string); ok {
 		newProviderID = v
@@ -70,18 +70,37 @@ func (s *modelService) Update(id string, updates map[string]interface{}) error {
 		newModelName = v
 	}
 
-	// 如果 provider_id 或 model_name 发生变化，校验联合唯一
 	if newProviderID != current.ProviderID || newModelName != current.ModelName {
 		if err := s.checkDuplicateModelName(newProviderID, newModelName, id); err != nil {
 			return err
 		}
 	}
 
-	return s.modelRepo.Update(id, updates)
+	err = s.modelRepo.Update(id, updates)
+	if err != nil {
+		return err
+	}
+
+	s.cache.InvalidateModel(current.ProviderID, current.ModelName)
+	if newProviderID != current.ProviderID || newModelName != current.ModelName {
+		s.cache.InvalidateModel(newProviderID, newModelName)
+	}
+	return nil
 }
 
 func (s *modelService) Delete(id string) error {
-	return s.modelRepo.Delete(id)
+	model, err := s.modelRepo.GetByID(id)
+	if err != nil {
+		return appErrors.ErrModelNotFound
+	}
+
+	err = s.modelRepo.Delete(id)
+	if err != nil {
+		return err
+	}
+
+	s.cache.InvalidateModel(model.ProviderID, model.ModelName)
+	return nil
 }
 
 // checkDuplicateModelName 校验同一供应商下 model_name 是否重复

backend/internal/service/provider_service_impl.go

@@ -13,23 +13,27 @@ import (
 type providerService struct {
 	providerRepo repository.ProviderRepository
 	modelRepo    repository.ModelRepository
+	cache        *RoutingCache
 }
 
-func NewProviderService(providerRepo repository.ProviderRepository, modelRepo repository.ModelRepository) ProviderService {
-	return &providerService{providerRepo: providerRepo, modelRepo: modelRepo}
+func NewProviderService(providerRepo repository.ProviderRepository, modelRepo repository.ModelRepository, cache *RoutingCache) ProviderService {
+	return &providerService{providerRepo: providerRepo, modelRepo: modelRepo, cache: cache}
 }
 
 func (s *providerService) Create(provider *domain.Provider) error {
-	// 校验 provider_id 字符集
 	if err := modelid.ValidateProviderID(provider.ID); err != nil {
 		return appErrors.ErrInvalidProviderID
 	}
 	provider.Enabled = true
 	err := s.providerRepo.Create(provider)
-	if err != nil && isUniqueConstraintError(err) {
-		return appErrors.ErrConflict
+	if err != nil {
+		if isUniqueConstraintError(err) {
+			return appErrors.ErrConflict
+		}
+		return err
 	}
-	return err
+	s.cache.SetProvider(provider)
+	return nil
 }
 
 func (s *providerService) Get(id string) (*domain.Provider, error) {
@@ -44,11 +48,21 @@ func (s *providerService) Update(id string, updates map[string]interface{}) erro
 	if _, ok := updates["id"]; ok {
 		return appErrors.ErrImmutableField
 	}
-	return s.providerRepo.Update(id, updates)
+	err := s.providerRepo.Update(id, updates)
+	if err != nil {
+		return err
+	}
+	s.cache.InvalidateProvider(id)
+	return nil
 }
 
 func (s *providerService) Delete(id string) error {
-	return s.providerRepo.Delete(id)
+	err := s.providerRepo.Delete(id)
+	if err != nil {
+		return err
+	}
+	s.cache.InvalidateProvider(id)
+	return nil
 }
 
 // ListEnabledModels 返回所有启用的模型（用于 Models 接口本地聚合）

backend/internal/service/routing_cache.go

@@ -0,0 +1,134 @@
+package service
+
+import (
+	"strings"
+	"sync"
+
+	"go.uber.org/zap"
+
+	"nex/backend/internal/domain"
+	"nex/backend/internal/repository"
+)
+
+type RoutingCache struct {
+	providers sync.Map
+	models    sync.Map
+
+	modelRepo    repository.ModelRepository
+	providerRepo repository.ProviderRepository
+	logger       *zap.Logger
+}
+
+func NewRoutingCache(
+	modelRepo repository.ModelRepository,
+	providerRepo repository.ProviderRepository,
+	logger *zap.Logger,
+) *RoutingCache {
+	return &RoutingCache{
+		modelRepo:    modelRepo,
+		providerRepo: providerRepo,
+		logger:       logger,
+	}
+}
+
+func (c *RoutingCache) GetProvider(id string) (*domain.Provider, error) {
+	if v, ok := c.providers.Load(id); ok {
+		return v.(*domain.Provider), nil
+	}
+
+	provider, err := c.providerRepo.GetByID(id)
+	if err != nil {
+		return nil, err
+	}
+
+	if v, ok := c.providers.Load(id); ok {
+		return v.(*domain.Provider), nil
+	}
+
+	c.providers.Store(id, provider)
+	return provider, nil
+}
+
+func (c *RoutingCache) GetModel(providerID, modelName string) (*domain.Model, error) {
+	key := providerID + "/" + modelName
+
+	if v, ok := c.models.Load(key); ok {
+		return v.(*domain.Model), nil
+	}
+
+	model, err := c.modelRepo.FindByProviderAndModelName(providerID, modelName)
+	if err != nil {
+		return nil, err
+	}
+
+	if v, ok := c.models.Load(key); ok {
+		return v.(*domain.Model), nil
+	}
+
+	c.models.Store(key, model)
+	return model, nil
+}
+
+func (c *RoutingCache) SetProvider(provider *domain.Provider) {
+	c.providers.Store(provider.ID, provider)
+}
+
+func (c *RoutingCache) SetModel(model *domain.Model) {
+	key := model.ProviderID + "/" + model.ModelName
+	c.models.Store(key, model)
+}
+
+func (c *RoutingCache) InvalidateProvider(id string) {
+	c.providers.Delete(id)
+	c.invalidateModelsByProvider(id)
+	c.logger.Debug("Provider 缓存失效", zap.String("provider_id", id))
+}
+
+func (c *RoutingCache) InvalidateModel(providerID, modelName string) {
+	key := providerID + "/" + modelName
+	c.models.Delete(key)
+	c.logger.Debug("Model 缓存失效",
+		zap.String("provider_id", providerID),
+		zap.String("model_name", modelName))
+}
+
+func (c *RoutingCache) invalidateModelsByProvider(providerID string) {
+	prefix := providerID + "/"
+	count := 0
+	c.models.Range(func(key, value interface{}) bool {
+		if strings.HasPrefix(key.(string), prefix) {
+			c.models.Delete(key)
+			count++
+		}
+		return true
+	})
+	if count > 0 {
+		c.logger.Debug("清除 Provider 相关 Model 缓存",
+			zap.String("provider_id", providerID),
+			zap.Int("count", count))
+	}
+}
+
+func (c *RoutingCache) Preload() error {
+	providers, err := c.providerRepo.List()
+	if err != nil {
+		return err
+	}
+	for i := range providers {
+		c.providers.Store(providers[i].ID, &providers[i])
+	}
+
+	models, err := c.modelRepo.List("")
+	if err != nil {
+		return err
+	}
+	for i := range models {
+		key := models[i].ProviderID + "/" + models[i].ModelName
+		c.models.Store(key, &models[i])
+	}
+
+	c.logger.Info("缓存预热完成",
+		zap.Int("providers", len(providers)),
+		zap.Int("models", len(models)))
+	return nil
+}

backend/internal/service/routing_cache_test.go

@@ -0,0 +1,273 @@
+package service
+
+import (
+	"errors"
+	"sync"
+	"testing"
+
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
+	"go.uber.org/zap"
+
+	"nex/backend/internal/domain"
+)
+
+type mockModelRepo struct {
+	models map[string]*domain.Model
+}
+
+func (m *mockModelRepo) Create(model *domain.Model) error {
+	return nil
+}
+
+func (m *mockModelRepo) GetByID(id string) (*domain.Model, error) {
+	return nil, nil
+}
+
+func (m *mockModelRepo) FindByProviderAndModelName(providerID, modelName string) (*domain.Model, error) {
+	key := providerID + "/" + modelName
+	if model, ok := m.models[key]; ok {
+		return model, nil
+	}
+	return nil, errors.New("not found")
+}
+
+func (m *mockModelRepo) List(providerID string) ([]domain.Model, error) {
+	var result []domain.Model
+	for _, model := range m.models {
+		if providerID == "" || model.ProviderID == providerID {
+			result = append(result, *model)
+		}
+	}
+	return result, nil
+}
+
+func (m *mockModelRepo) ListEnabled() ([]domain.Model, error) {
+	return nil, nil
+}
+
+func (m *mockModelRepo) Update(id string, updates map[string]interface{}) error {
+	return nil
+}
+
+func (m *mockModelRepo) Delete(id string) error {
+	return nil
+}
+
+type mockProviderRepo struct {
+	providers map[string]*domain.Provider
+}
+
+func (m *mockProviderRepo) Create(provider *domain.Provider) error {
+	return nil
+}
+
+func (m *mockProviderRepo) GetByID(id string) (*domain.Provider, error) {
+	if provider, ok := m.providers[id]; ok {
+		return provider, nil
+	}
+	return nil, errors.New("not found")
+}
+
+func (m *mockProviderRepo) List() ([]domain.Provider, error) {
+	var result []domain.Provider
+	for _, provider := range m.providers {
+		result = append(result, *provider)
+	}
+	return result, nil
+}
+
+func (m *mockProviderRepo) Update(id string, updates map[string]interface{}) error {
+	return nil
+}
+
+func (m *mockProviderRepo) Delete(id string) error {
+	return nil
+}
+
+func TestRoutingCache_GetProvider_CacheHit(t *testing.T) {
+	logger := zap.NewNop()
+	modelRepo := &mockModelRepo{models: make(map[string]*domain.Model)}
+	providerRepo := &mockProviderRepo{providers: map[string]*domain.Provider{
+		"openai": {ID: "openai", Name: "OpenAI", Enabled: true},
+	}}
+
+	cache := NewRoutingCache(modelRepo, providerRepo, logger)
+
+	provider, err := cache.GetProvider("openai")
+	require.NoError(t, err)
+	assert.Equal(t, "openai", provider.ID)
+
+	provider2, err := cache.GetProvider("openai")
+	require.NoError(t, err)
+	assert.Equal(t, provider, provider2)
+}
+
+func TestRoutingCache_GetProvider_CacheMiss(t *testing.T) {
+	logger := zap.NewNop()
+	modelRepo := &mockModelRepo{models: make(map[string]*domain.Model)}
+	providerRepo := &mockProviderRepo{providers: make(map[string]*domain.Provider)}
+
+	cache := NewRoutingCache(modelRepo, providerRepo, logger)
+
+	_, err := cache.GetProvider("notexist")
+	assert.Error(t, err)
+}
+
+func TestRoutingCache_GetModel_CacheHit(t *testing.T) {
+	logger := zap.NewNop()
+	modelRepo := &mockModelRepo{models: map[string]*domain.Model{
+		"openai/gpt-4": {ID: "1", ProviderID: "openai", ModelName: "gpt-4", Enabled: true},
+	}}
+	providerRepo := &mockProviderRepo{providers: make(map[string]*domain.Provider)}
+
+	cache := NewRoutingCache(modelRepo, providerRepo, logger)
+
+	model, err := cache.GetModel("openai", "gpt-4")
+	require.NoError(t, err)
+	assert.Equal(t, "gpt-4", model.ModelName)
+
+	model2, err := cache.GetModel("openai", "gpt-4")
+	require.NoError(t, err)
+	assert.Equal(t, model, model2)
+}
+
+func TestRoutingCache_GetModel_CacheMiss(t *testing.T) {
+	logger := zap.NewNop()
+	modelRepo := &mockModelRepo{models: make(map[string]*domain.Model)}
+	providerRepo := &mockProviderRepo{providers: make(map[string]*domain.Provider)}
+
+	cache := NewRoutingCache(modelRepo, providerRepo, logger)
+
+	_, err := cache.GetModel("openai", "notexist")
+	assert.Error(t, err)
+}
+
+func TestRoutingCache_DoubleCheck(t *testing.T) {
+	logger := zap.NewNop()
+	modelRepo := &mockModelRepo{models: make(map[string]*domain.Model)}
+	providerRepo := &mockProviderRepo{providers: map[string]*domain.Provider{
+		"openai": {ID: "openai", Name: "OpenAI", Enabled: true},
+	}}
+
+	cache := NewRoutingCache(modelRepo, providerRepo, logger)
+
+	var wg sync.WaitGroup
+	for i := 0; i < 10; i++ {
+		wg.Add(1)
+		go func() {
+			defer wg.Done()
+			_, err := cache.GetProvider("openai")
+			assert.NoError(t, err)
+		}()
+	}
+	wg.Wait()
+}
+
+func TestRoutingCache_InvalidateProvider_CascadingModels(t *testing.T) {
+	logger := zap.NewNop()
+	modelRepo := &mockModelRepo{models: map[string]*domain.Model{
+		"openai/gpt-4":     {ID: "1", ProviderID: "openai", ModelName: "gpt-4", Enabled: true},
+		"openai/gpt-3.5":   {ID: "2", ProviderID: "openai", ModelName: "gpt-3.5", Enabled: true},
+		"anthropic/claude": {ID: "3", ProviderID: "anthropic", ModelName: "claude", Enabled: true},
+	}}
+	providerRepo := &mockProviderRepo{providers: map[string]*domain.Provider{
+		"openai":    {ID: "openai", Name: "OpenAI", Enabled: true},
+		"anthropic": {ID: "anthropic", Name: "Anthropic", Enabled: true},
+	}}
+
+	cache := NewRoutingCache(modelRepo, providerRepo, logger)
+
+	_, err := cache.GetModel("openai", "gpt-4")
+	require.NoError(t, err)
+	_, err = cache.GetModel("openai", "gpt-3.5")
+	require.NoError(t, err)
+	_, err = cache.GetModel("anthropic", "claude")
+	require.NoError(t, err)
+
+	cache.InvalidateProvider("openai")
+
+	var openaiCount, anthropicCount int
+	cache.models.Range(func(key, value interface{}) bool {
+		if key.(string) == "anthropic/claude" {
+			anthropicCount++
+		}
+		return true
+	})
+	assert.Equal(t, 0, openaiCount)
+	assert.Equal(t, 1, anthropicCount)
+}
+
+func TestRoutingCache_InvalidateModel(t *testing.T) {
+	logger := zap.NewNop()
+	modelRepo := &mockModelRepo{models: map[string]*domain.Model{
+		"openai/gpt-4": {ID: "1", ProviderID: "openai", ModelName: "gpt-4", Enabled: true},
+	}}
+	providerRepo := &mockProviderRepo{providers: make(map[string]*domain.Provider)}
+
+	cache := NewRoutingCache(modelRepo, providerRepo, logger)
+
+	_, err := cache.GetModel("openai", "gpt-4")
+	require.NoError(t, err)
+
+	cache.InvalidateModel("openai", "gpt-4")
+
+	var count int
+	cache.models.Range(func(key, value interface{}) bool {
+		count++
+		return true
+	})
+	assert.Equal(t, 0, count)
+}
+
+func TestRoutingCache_Preload_Success(t *testing.T) {
+	logger := zap.NewNop()
+	modelRepo := &mockModelRepo{models: map[string]*domain.Model{
+		"openai/gpt-4": {ID: "1", ProviderID: "openai", ModelName: "gpt-4", Enabled: true},
+	}}
+	providerRepo := &mockProviderRepo{providers: map[string]*domain.Provider{
+		"openai": {ID: "openai", Name: "OpenAI", Enabled: true},
+	}}
+
+	cache := NewRoutingCache(modelRepo, providerRepo, logger)
+
+	err := cache.Preload()
+	require.NoError(t, err)
+
+	var providerCount, modelCount int
+	cache.providers.Range(func(key, value interface{}) bool {
+		providerCount++
+		return true
+	})
+	cache.models.Range(func(key, value interface{}) bool {
+		modelCount++
+		return true
+	})
+	assert.Equal(t, 1, providerCount)
+	assert.Equal(t, 1, modelCount)
+}
+
+func TestRoutingCache_ConcurrentAccess(t *testing.T) {
+	logger := zap.NewNop()
+	modelRepo := &mockModelRepo{models: map[string]*domain.Model{
+		"openai/gpt-4": {ID: "1", ProviderID: "openai", ModelName: "gpt-4", Enabled: true},
+	}}
+	providerRepo := &mockProviderRepo{providers: map[string]*domain.Provider{
+		"openai": {ID: "openai", Name: "OpenAI", Enabled: true},
+	}}
+
+	cache := NewRoutingCache(modelRepo, providerRepo, logger)
+
+	var wg sync.WaitGroup
+	for i := 0; i < 100; i++ {
+		wg.Add(1)
+		go func() {
+			defer wg.Done()
+			_, _ = cache.GetProvider("openai")
+			_, _ = cache.GetModel("openai", "gpt-4")
+			cache.InvalidateProvider("openai")
+			cache.InvalidateModel("openai", "gpt-4")
+		}()
+	}
+	wg.Wait()
+}

backend/internal/service/routing_service_impl.go

@@ -4,20 +4,18 @@ import (
 	appErrors "nex/backend/pkg/errors"
 
 	"nex/backend/internal/domain"
-	"nex/backend/internal/repository"
 )
 
 type routingService struct {
-	modelRepo    repository.ModelRepository
-	providerRepo repository.ProviderRepository
+	cache *RoutingCache
 }
 
-func NewRoutingService(modelRepo repository.ModelRepository, providerRepo repository.ProviderRepository) RoutingService {
-	return &routingService{modelRepo: modelRepo, providerRepo: providerRepo}
+func NewRoutingService(cache *RoutingCache) RoutingService {
+	return &routingService{cache: cache}
 }
 
 func (s *routingService) RouteByModelName(providerID, modelName string) (*domain.RouteResult, error) {
-	model, err := s.modelRepo.FindByProviderAndModelName(providerID, modelName)
+	model, err := s.cache.GetModel(providerID, modelName)
 	if err != nil {
 		return nil, appErrors.ErrModelNotFound
 	}
@@ -26,7 +24,7 @@ func (s *routingService) RouteByModelName(providerID, modelName string) (*domain
 		return nil, appErrors.ErrModelDisabled
 	}
 
-	provider, err := s.providerRepo.GetByID(model.ProviderID)
+	provider, err := s.cache.GetProvider(model.ProviderID)
 	if err != nil {
 		return nil, appErrors.ErrProviderNotFound
 	}

backend/internal/service/service_supplemental_test.go

@@ -14,7 +14,8 @@ func TestProviderService_Update(t *testing.T) {
 	db := setupServiceTestDB(t)
 	repo := repository.NewProviderRepository(db)
 	modelRepo := repository.NewModelRepository(db)
-	svc := NewProviderService(repo, modelRepo)
+	cache := setupRoutingCache(t, db)
+	svc := NewProviderService(repo, modelRepo, cache)
 
 	require.NoError(t, svc.Create(&domain.Provider{ID: "p1", Name: "Original", APIKey: "key", BaseURL: "https://test.com"}))
 
@@ -30,7 +31,8 @@ func TestProviderService_Update_NotFound(t *testing.T) {
 	db := setupServiceTestDB(t)
 	repo := repository.NewProviderRepository(db)
 	modelRepo := repository.NewModelRepository(db)
-	svc := NewProviderService(repo, modelRepo)
+	cache := setupRoutingCache(t, db)
+	svc := NewProviderService(repo, modelRepo, cache)
 
 	err := svc.Update("nonexistent", map[string]interface{}{"name": "test"})
 	assert.Error(t, err)
@@ -40,7 +42,8 @@ func TestModelService_Get(t *testing.T) {
 	db := setupServiceTestDB(t)
 	providerRepo := repository.NewProviderRepository(db)
 	modelRepo := repository.NewModelRepository(db)
-	svc := NewModelService(modelRepo, providerRepo)
+	cache := setupRoutingCache(t, db)
+	svc := NewModelService(modelRepo, providerRepo, cache)
 
 	require.NoError(t, providerRepo.Create(&domain.Provider{ID: "p1", Name: "P1", APIKey: "key", BaseURL: "https://test.com"}))
 	model := &domain.Model{ProviderID: "p1", ModelName: "gpt-4"}
@@ -55,7 +58,8 @@ func TestModelService_Update(t *testing.T) {
 	db := setupServiceTestDB(t)
 	providerRepo := repository.NewProviderRepository(db)
 	modelRepo := repository.NewModelRepository(db)
-	svc := NewModelService(modelRepo, providerRepo)
+	cache := setupRoutingCache(t, db)
+	svc := NewModelService(modelRepo, providerRepo, cache)
 
 	require.NoError(t, providerRepo.Create(&domain.Provider{ID: "p1", Name: "P1", APIKey: "key", BaseURL: "https://test.com"}))
 	model := &domain.Model{ProviderID: "p1", ModelName: "gpt-4"}
@@ -73,7 +77,8 @@ func TestModelService_Update_ProviderID_Invalid(t *testing.T) {
 	db := setupServiceTestDB(t)
 	providerRepo := repository.NewProviderRepository(db)
 	modelRepo := repository.NewModelRepository(db)
-	svc := NewModelService(modelRepo, providerRepo)
+	cache := setupRoutingCache(t, db)
+	svc := NewModelService(modelRepo, providerRepo, cache)
 
 	require.NoError(t, providerRepo.Create(&domain.Provider{ID: "p1", Name: "P1", APIKey: "key", BaseURL: "https://test.com"}))
 	model := &domain.Model{ProviderID: "p1", ModelName: "gpt-4"}
@@ -87,7 +92,8 @@ func TestModelService_Delete(t *testing.T) {
 	db := setupServiceTestDB(t)
 	providerRepo := repository.NewProviderRepository(db)
 	modelRepo := repository.NewModelRepository(db)
-	svc := NewModelService(modelRepo, providerRepo)
+	cache := setupRoutingCache(t, db)
+	svc := NewModelService(modelRepo, providerRepo, cache)
 
 	require.NoError(t, providerRepo.Create(&domain.Provider{ID: "p1", Name: "P1", APIKey: "key", BaseURL: "https://test.com"}))
 	model := &domain.Model{ProviderID: "p1", ModelName: "gpt-4"}
@@ -104,7 +110,8 @@ func TestModelService_Delete_NotFound(t *testing.T) {
 	db := setupServiceTestDB(t)
 	modelRepo := repository.NewModelRepository(db)
 	providerRepo := repository.NewProviderRepository(db)
-	svc := NewModelService(modelRepo, providerRepo)
+	cache := setupRoutingCache(t, db)
+	svc := NewModelService(modelRepo, providerRepo, cache)
 
 	err := svc.Delete("nonexistent")
 	assert.Error(t, err)
@@ -112,7 +119,8 @@ func TestModelService_Delete_NotFound(t *testing.T) {
 
 func TestStatsService_Aggregate_Default(t *testing.T) {
 	statsRepo := repository.NewStatsRepository(nil)
-	svc := NewStatsService(statsRepo)
+	buffer := NewStatsBuffer(statsRepo, nil)
+	svc := NewStatsService(statsRepo, buffer)
 
 	stats := []domain.UsageStats{
 		{ProviderID: "p1", RequestCount: 10},
@@ -133,7 +141,8 @@ func TestModelService_Update_NotFound(t *testing.T) {
 	db := setupServiceTestDB(t)
 	modelRepo := repository.NewModelRepository(db)
 	providerRepo := repository.NewProviderRepository(db)
-	svc := NewModelService(modelRepo, providerRepo)
+	cache := setupRoutingCache(t, db)
+	svc := NewModelService(modelRepo, providerRepo, cache)
 
 	err := svc.Update("nonexistent", map[string]interface{}{"model_name": "test"})
 	assert.Error(t, err)

backend/internal/service/service_test.go

@@ -8,6 +8,7 @@ import (
 	"github.com/google/uuid"
 	"github.com/stretchr/testify/assert"
 	"github.com/stretchr/testify/require"
+	"go.uber.org/zap"
 	"gorm.io/gorm"
 
 	testHelpers "nex/backend/tests"
@@ -22,13 +23,21 @@ func setupServiceTestDB(t *testing.T) *gorm.DB {
 	return testHelpers.SetupTestDB(t)
 }
 
+func setupRoutingCache(t *testing.T, db *gorm.DB) *RoutingCache {
+	t.Helper()
+	providerRepo := repository.NewProviderRepository(db)
+	modelRepo := repository.NewModelRepository(db)
+	return NewRoutingCache(modelRepo, providerRepo, zap.NewNop())
+}
+
 // ============ RoutingService - RouteByModelName 测试 ============
 
 func TestRoutingService_RouteByModelName_Success(t *testing.T) {
 	db := setupServiceTestDB(t)
 	providerRepo := repository.NewProviderRepository(db)
 	modelRepo := repository.NewModelRepository(db)
-	svc := NewRoutingService(modelRepo, providerRepo)
+	cache := setupRoutingCache(t, db)
+	svc := NewRoutingService(cache)
 
 	require.NoError(t, providerRepo.Create(&domain.Provider{ID: "openai", Name: "OpenAI", APIKey: "key", BaseURL: "https://api.openai.com", Enabled: true}))
 	require.NoError(t, modelRepo.Create(&domain.Model{ID: "m1", ProviderID: "openai", ModelName: "gpt-4", Enabled: true}))
@@ -41,9 +50,8 @@ func TestRoutingService_RouteByModelName_Success(t *testing.T) {
 
 func TestRoutingService_RouteByModelName_NotFound(t *testing.T) {
 	db := setupServiceTestDB(t)
-	providerRepo := repository.NewProviderRepository(db)
-	modelRepo := repository.NewModelRepository(db)
-	svc := NewRoutingService(modelRepo, providerRepo)
+	cache := setupRoutingCache(t, db)
+	svc := NewRoutingService(cache)
 
 	_, err := svc.RouteByModelName("openai", "nonexistent-model")
 	assert.True(t, errors.Is(err, appErrors.ErrModelNotFound))
@@ -53,7 +61,8 @@ func TestRoutingService_RouteByModelName_DisabledModel(t *testing.T) {
 	db := setupServiceTestDB(t)
 	providerRepo := repository.NewProviderRepository(db)
 	modelRepo := repository.NewModelRepository(db)
-	svc := NewRoutingService(modelRepo, providerRepo)
+	cache := setupRoutingCache(t, db)
+	svc := NewRoutingService(cache)
 
 	require.NoError(t, providerRepo.Create(&domain.Provider{ID: "openai", Name: "OpenAI", APIKey: "key", BaseURL: "https://api.openai.com", Enabled: true}))
 	require.NoError(t, modelRepo.Create(&domain.Model{ID: "m1", ProviderID: "openai", ModelName: "gpt-4", Enabled: true}))
@@ -67,7 +76,8 @@ func TestRoutingService_RouteByModelName_DisabledProvider(t *testing.T) {
 	db := setupServiceTestDB(t)
 	providerRepo := repository.NewProviderRepository(db)
 	modelRepo := repository.NewModelRepository(db)
-	svc := NewRoutingService(modelRepo, providerRepo)
+	cache := setupRoutingCache(t, db)
+	svc := NewRoutingService(cache)
 
 	require.NoError(t, providerRepo.Create(&domain.Provider{ID: "openai", Name: "OpenAI", APIKey: "key", BaseURL: "https://api.openai.com", Enabled: true}))
 	require.NoError(t, modelRepo.Create(&domain.Model{ID: "m1", ProviderID: "openai", ModelName: "gpt-4", Enabled: true}))
@@ -83,7 +93,8 @@ func TestModelService_Create_GeneratesUUID(t *testing.T) {
 	db := setupServiceTestDB(t)
 	providerRepo := repository.NewProviderRepository(db)
 	modelRepo := repository.NewModelRepository(db)
-	svc := NewModelService(modelRepo, providerRepo)
+	cache := setupRoutingCache(t, db)
+	svc := NewModelService(modelRepo, providerRepo, cache)
 
 	require.NoError(t, providerRepo.Create(&domain.Provider{ID: "openai", Name: "OpenAI", APIKey: "key", BaseURL: "https://api.openai.com"}))
 
@@ -91,12 +102,10 @@ func TestModelService_Create_GeneratesUUID(t *testing.T) {
 	err := svc.Create(model)
 	require.NoError(t, err)
 
-	// 验证返回的 model 拥有有效的 UUID
 	assert.NotEmpty(t, model.ID)
 	_, err = uuid.Parse(model.ID)
 	assert.NoError(t, err, "model.ID should be a valid UUID")
 
-	// 通过 Get 验证持久化
 	stored, err := svc.Get(model.ID)
 	require.NoError(t, err)
 	assert.Equal(t, model.ID, stored.ID)
@@ -107,7 +116,8 @@ func TestModelService_Create_DuplicateModelName(t *testing.T) {
 	db := setupServiceTestDB(t)
 	providerRepo := repository.NewProviderRepository(db)
 	modelRepo := repository.NewModelRepository(db)
-	svc := NewModelService(modelRepo, providerRepo)
+	cache := setupRoutingCache(t, db)
+	svc := NewModelService(modelRepo, providerRepo, cache)
 
 	require.NoError(t, providerRepo.Create(&domain.Provider{ID: "openai", Name: "OpenAI", APIKey: "key", BaseURL: "https://api.openai.com"}))
 
@@ -115,7 +125,6 @@ func TestModelService_Create_DuplicateModelName(t *testing.T) {
 	err := svc.Create(model1)
 	require.NoError(t, err)
 
-	// 使用相同的 (providerID, modelName) 创建第二个模型应失败
 	model2 := &domain.Model{ProviderID: "openai", ModelName: "gpt-4"}
 	err = svc.Create(model2)
 	assert.True(t, errors.Is(err, appErrors.ErrDuplicateModel))
@@ -125,7 +134,8 @@ func TestModelService_Create_ProviderNotFound(t *testing.T) {
 	db := setupServiceTestDB(t)
 	providerRepo := repository.NewProviderRepository(db)
 	modelRepo := repository.NewModelRepository(db)
-	svc := NewModelService(modelRepo, providerRepo)
+	 cache := setupRoutingCache(t, db)
+	svc := NewModelService(modelRepo, providerRepo, cache)
 
 	model := &domain.Model{ProviderID: "nonexistent", ModelName: "gpt-4"}
 	err := svc.Create(model)
@@ -138,7 +148,8 @@ func TestProviderService_Create_InvalidID(t *testing.T) {
 	db := setupServiceTestDB(t)
 	repo := repository.NewProviderRepository(db)
 	modelRepo := repository.NewModelRepository(db)
-	svc := NewProviderService(repo, modelRepo)
+	 cache := setupRoutingCache(t, db)
+	svc := NewProviderService(repo, modelRepo, cache)
 
 	provider := &domain.Provider{ID: "open-ai", Name: "OpenAI", APIKey: "key", BaseURL: "https://api.openai.com"}
 	err := svc.Create(provider)
@@ -149,7 +160,8 @@ func TestProviderService_Create_ValidID(t *testing.T) {
 	db := setupServiceTestDB(t)
 	repo := repository.NewProviderRepository(db)
 	modelRepo := repository.NewModelRepository(db)
-	svc := NewProviderService(repo, modelRepo)
+	 cache := setupRoutingCache(t, db)
+	svc := NewProviderService(repo, modelRepo, cache)
 
 	provider := &domain.Provider{ID: "openai", Name: "OpenAI", APIKey: "key", BaseURL: "https://api.openai.com"}
 	err := svc.Create(provider)
@@ -164,7 +176,8 @@ func TestModelService_Update_DuplicateModelName(t *testing.T) {
 	db := setupServiceTestDB(t)
 	providerRepo := repository.NewProviderRepository(db)
 	modelRepo := repository.NewModelRepository(db)
-	svc := NewModelService(modelRepo, providerRepo)
+	 cache := setupRoutingCache(t, db)
+	svc := NewModelService(modelRepo, providerRepo, cache)
 
 	require.NoError(t, providerRepo.Create(&domain.Provider{ID: "openai", Name: "OpenAI", APIKey: "key", BaseURL: "https://api.openai.com"}))
 	require.NoError(t, providerRepo.Create(&domain.Provider{ID: "anthropic", Name: "Anthropic", APIKey: "key", BaseURL: "https://api.anthropic.com"}))
@@ -189,7 +202,8 @@ func TestModelService_Update_ModelNotFound(t *testing.T) {
 	db := setupServiceTestDB(t)
 	providerRepo := repository.NewProviderRepository(db)
 	modelRepo := repository.NewModelRepository(db)
-	svc := NewModelService(modelRepo, providerRepo)
+	 cache := setupRoutingCache(t, db)
+	svc := NewModelService(modelRepo, providerRepo, cache)
 
 	err := svc.Update("nonexistent-id", map[string]interface{}{
 		"model_name": "gpt-4",
@@ -201,7 +215,8 @@ func TestModelService_Update_Success(t *testing.T) {
 	db := setupServiceTestDB(t)
 	providerRepo := repository.NewProviderRepository(db)
 	modelRepo := repository.NewModelRepository(db)
-	svc := NewModelService(modelRepo, providerRepo)
+	 cache := setupRoutingCache(t, db)
+	svc := NewModelService(modelRepo, providerRepo, cache)
 
 	require.NoError(t, providerRepo.Create(&domain.Provider{ID: "openai", Name: "OpenAI", APIKey: "key", BaseURL: "https://api.openai.com"}))
 
@@ -226,7 +241,8 @@ func TestProviderService_Update_ImmutableID(t *testing.T) {
 	db := setupServiceTestDB(t)
 	repo := repository.NewProviderRepository(db)
 	modelRepo := repository.NewModelRepository(db)
-	svc := NewProviderService(repo, modelRepo)
+	 cache := setupRoutingCache(t, db)
+	svc := NewProviderService(repo, modelRepo, cache)
 
 	provider := &domain.Provider{ID: "openai", Name: "OpenAI", APIKey: "key", BaseURL: "https://api.openai.com"}
 	err := svc.Create(provider)
@@ -243,7 +259,8 @@ func TestProviderService_Update_Success(t *testing.T) {
 	db := setupServiceTestDB(t)
 	repo := repository.NewProviderRepository(db)
 	modelRepo := repository.NewModelRepository(db)
-	svc := NewProviderService(repo, modelRepo)
+	 cache := setupRoutingCache(t, db)
+	svc := NewProviderService(repo, modelRepo, cache)
 
 	provider := &domain.Provider{ID: "openai", Name: "OpenAI", APIKey: "key", BaseURL: "https://api.openai.com"}
 	err := svc.Create(provider)
@@ -301,7 +318,7 @@ func TestStatsService_Aggregate_ByModel(t *testing.T) {
 		t.Run(tt.name, func(t *testing.T) {
 			db := setupServiceTestDB(t)
 			statsRepo := repository.NewStatsRepository(db)
-			svc := NewStatsService(statsRepo)
+			buffer := NewStatsBuffer(statsRepo, nil); svc := NewStatsService(statsRepo, buffer)
 
 			result := svc.Aggregate(tt.stats, "model")
 
@@ -362,7 +379,7 @@ func TestStatsService_Aggregate_ByDate(t *testing.T) {
 		t.Run(tt.name, func(t *testing.T) {
 			db := setupServiceTestDB(t)
 			statsRepo := repository.NewStatsRepository(db)
-			svc := NewStatsService(statsRepo)
+			buffer := NewStatsBuffer(statsRepo, nil); svc := NewStatsService(statsRepo, buffer)
 
 			result := svc.Aggregate(tt.stats, "date")
 
@@ -431,7 +448,8 @@ func TestProviderService_List_APIKeyNotMasked(t *testing.T) {
 	db := setupServiceTestDB(t)
 	repo := repository.NewProviderRepository(db)
 	modelRepo := repository.NewModelRepository(db)
-	svc := NewProviderService(repo, modelRepo)
+	 cache := setupRoutingCache(t, db)
+	svc := NewProviderService(repo, modelRepo, cache)
 
 	provider1 := &domain.Provider{ID: "openai", Name: "OpenAI", APIKey: "sk-1234567890", BaseURL: "https://api.openai.com"}
 	provider2 := &domain.Provider{ID: "anthropic", Name: "Anthropic", APIKey: "sk-anthropic1234", BaseURL: "https://api.anthropic.com"}
@@ -456,7 +474,8 @@ func TestModelService_ConcurrentCreate(t *testing.T) {
 	db := setupServiceTestDB(t)
 	providerRepo := repository.NewProviderRepository(db)
 	modelRepo := repository.NewModelRepository(db)
-	svc := NewModelService(modelRepo, providerRepo)
+	 cache := setupRoutingCache(t, db)
+	svc := NewModelService(modelRepo, providerRepo, cache)
 
 	require.NoError(t, providerRepo.Create(&domain.Provider{ID: "openai", Name: "OpenAI", APIKey: "key", BaseURL: "https://api.openai.com"}))
 

backend/internal/service/stats_buffer.go

@@ -0,0 +1,169 @@
+package service
+
+import (
+	"strings"
+	"sync"
+	"sync/atomic"
+	"time"
+
+	"go.uber.org/zap"
+
+	"nex/backend/internal/repository"
+)
+
+type StatsBuffer struct {
+	counters sync.Map
+
+	flushInterval  time.Duration
+	flushThreshold int
+	totalCount     atomic.Int64
+
+	statsRepo repository.StatsRepository
+	logger    *zap.Logger
+
+	stopCh chan struct{}
+	doneCh chan struct{}
+}
+
+type StatsBufferOption func(*StatsBuffer)
+
+func WithFlushInterval(d time.Duration) StatsBufferOption {
+	return func(b *StatsBuffer) {
+		b.flushInterval = d
+	}
+}
+
+func WithFlushThreshold(threshold int) StatsBufferOption {
+	return func(b *StatsBuffer) {
+		b.flushThreshold = threshold
+	}
+}
+
+func NewStatsBuffer(
+	statsRepo repository.StatsRepository,
+	logger *zap.Logger,
+	opts ...StatsBufferOption,
+) *StatsBuffer {
+	b := &StatsBuffer{
+		statsRepo:      statsRepo,
+		logger:         logger,
+		flushInterval:  5 * time.Second,
+		flushThreshold: 100,
+		stopCh:         make(chan struct{}),
+		doneCh:         make(chan struct{}),
+	}
+
+	for _, opt := range opts {
+		opt(b)
+	}
+
+	return b
+}
+
+func (b *StatsBuffer) Increment(providerID, modelName string) {
+	today := time.Now().Format("2006-01-02")
+	key := providerID + "/" + modelName + "/" + today
+
+	var counter *int64
+	if v, ok := b.counters.Load(key); ok {
+		counter = v.(*int64)
+	} else {
+		val := int64(0)
+		counter = &val
+		actual, loaded := b.counters.LoadOrStore(key, counter)
+		if loaded {
+			counter = actual.(*int64)
+		}
+	}
+
+	atomic.AddInt64(counter, 1)
+
+	if b.totalCount.Add(1) >= int64(b.flushThreshold) {
+		go b.flush()
+	}
+}
+
+func (b *StatsBuffer) Start() {
+	go func() {
+		ticker := time.NewTicker(b.flushInterval)
+		defer ticker.Stop()
+
+		for {
+			select {
+			case <-ticker.C:
+				b.flush()
+			case <-b.stopCh:
+				b.flush()
+				close(b.doneCh)
+				return
+			}
+		}
+	}()
+}
+
+func (b *StatsBuffer) Stop() {
+	close(b.stopCh)
+	<-b.doneCh
+}
+
+func (b *StatsBuffer) flush() {
+	type statEntry struct {
+		providerID string
+		modelName  string
+		date       string
+		count      int64
+	}
+
+	var entries []statEntry
+	b.counters.Range(func(key, value interface{}) bool {
+		keyStr := key.(string)
+		parts := strings.Split(keyStr, "/")
+		if len(parts) != 3 {
+			return true
+		}
+
+		counter := value.(*int64)
+		count := atomic.SwapInt64(counter, 0)
+
+		if count > 0 {
+			entries = append(entries, statEntry{
+				providerID: parts[0],
+				modelName:  parts[1],
+				date:       parts[2],
+				count:      count,
+			})
+		}
+		return true
+	})
+
+	if len(entries) == 0 {
+		return
+	}
+
+	success := 0
+	for _, entry := range entries {
+		date, _ := time.Parse("2006-01-02", entry.date)
+		err := b.statsRepo.BatchUpdate(entry.providerID, entry.modelName, date, int(entry.count))
+		if err != nil {
+			b.logger.Error("批量更新统计失败",
+				zap.String("provider_id", entry.providerID),
+				zap.String("model_name", entry.modelName),
+				zap.Int64("count", entry.count),
+				zap.Error(err))
+
+			key := entry.providerID + "/" + entry.modelName + "/" + entry.date
+			if v, ok := b.counters.Load(key); ok {
+				counter := v.(*int64)
+				atomic.AddInt64(counter, entry.count)
+			}
+		} else {
+			success++
+		}
+	}
+
+	b.totalCount.Store(0)
+	b.logger.Debug("统计刷新完成",
+		zap.Int("total", len(entries)),
+		zap.Int("success", success),
+		zap.Int("failed", len(entries)-success))
+}

backend/internal/service/stats_buffer_test.go

@@ -0,0 +1,251 @@
+package service
+
+import (
+	"errors"
+	"sync"
+	"sync/atomic"
+	"testing"
+	"time"
+
+	"github.com/stretchr/testify/assert"
+	"go.uber.org/zap"
+
+	"nex/backend/internal/domain"
+)
+
+type mockStatsRepo struct {
+	records []struct {
+		providerID string
+		modelName  string
+		date       time.Time
+		delta      int
+	}
+	fail bool
+	mu   sync.Mutex
+}
+
+func (m *mockStatsRepo) Record(providerID, modelName string) error {
+	return nil
+}
+
+func (m *mockStatsRepo) BatchUpdate(providerID, modelName string, date time.Time, delta int) error {
+	m.mu.Lock()
+	defer m.mu.Unlock()
+	if m.fail {
+		return errors.New("db error")
+	}
+	m.records = append(m.records, struct {
+		providerID string
+		modelName  string
+		date       time.Time
+		delta      int
+	}{providerID, modelName, date, delta})
+	return nil
+}
+
+func (m *mockStatsRepo) Query(providerID, modelName string, startDate, endDate *time.Time) ([]domain.UsageStats, error) {
+	return nil, nil
+}
+
+func TestStatsBuffer_Increment(t *testing.T) {
+	logger := zap.NewNop()
+	statsRepo := &mockStatsRepo{}
+	buffer := NewStatsBuffer(statsRepo, logger)
+
+	buffer.Increment("openai", "gpt-4")
+	buffer.Increment("openai", "gpt-4")
+	buffer.Increment("openai", "gpt-3.5")
+
+	var count int64
+	buffer.counters.Range(func(key, value interface{}) bool {
+		counter := value.(*int64)
+		count += atomic.LoadInt64(counter)
+		return true
+	})
+	assert.Equal(t, int64(3), count)
+}
+
+func TestStatsBuffer_ConcurrentIncrement(t *testing.T) {
+	logger := zap.NewNop()
+	statsRepo := &mockStatsRepo{}
+	buffer := NewStatsBuffer(statsRepo, logger)
+
+	var wg sync.WaitGroup
+	for i := 0; i < 100; i++ {
+		wg.Add(1)
+		go func() {
+			defer wg.Done()
+			buffer.Increment("openai", "gpt-4")
+		}()
+	}
+	wg.Wait()
+
+	var count int64
+	buffer.counters.Range(func(key, value interface{}) bool {
+		counter := value.(*int64)
+		count = atomic.LoadInt64(counter)
+		return true
+	})
+	assert.Equal(t, int64(100), count)
+}
+
+func TestStatsBuffer_LoadOrStore(t *testing.T) {
+	logger := zap.NewNop()
+	statsRepo := &mockStatsRepo{}
+	buffer := NewStatsBuffer(statsRepo, logger)
+
+	var wg sync.WaitGroup
+	for i := 0; i < 10; i++ {
+		wg.Add(1)
+		go func() {
+			defer wg.Done()
+			buffer.Increment("openai", "gpt-4")
+		}()
+	}
+	wg.Wait()
+
+	var counterCount int
+	buffer.counters.Range(func(key, value interface{}) bool {
+		counterCount++
+		return true
+	})
+	assert.Equal(t, 1, counterCount)
+}
+
+func TestStatsBuffer_FlushByInterval(t *testing.T) {
+	logger := zap.NewNop()
+	statsRepo := &mockStatsRepo{}
+	buffer := NewStatsBuffer(statsRepo, logger,
+		WithFlushInterval(100*time.Millisecond))
+
+	buffer.Start()
+	defer buffer.Stop()
+
+	buffer.Increment("openai", "gpt-4")
+	buffer.Increment("openai", "gpt-4")
+
+	time.Sleep(200 * time.Millisecond)
+
+	statsRepo.mu.Lock()
+	assert.GreaterOrEqual(t, len(statsRepo.records), 1)
+	statsRepo.mu.Unlock()
+}
+
+func TestStatsBuffer_FlushByThreshold(t *testing.T) {
+	logger := zap.NewNop()
+	statsRepo := &mockStatsRepo{}
+	buffer := NewStatsBuffer(statsRepo, logger,
+		WithFlushThreshold(10))
+
+	buffer.Start()
+	defer buffer.Stop()
+
+	for i := 0; i < 10; i++ {
+		buffer.Increment("openai", "gpt-4")
+	}
+
+	time.Sleep(50 * time.Millisecond)
+
+	statsRepo.mu.Lock()
+	assert.GreaterOrEqual(t, len(statsRepo.records), 1)
+	statsRepo.mu.Unlock()
+}
+
+func TestStatsBuffer_SwapInt64(t *testing.T) {
+	logger := zap.NewNop()
+	statsRepo := &mockStatsRepo{}
+	buffer := NewStatsBuffer(statsRepo, logger)
+
+	buffer.Increment("openai", "gpt-4")
+	buffer.Increment("openai", "gpt-4")
+
+	var beforeCount int64
+	buffer.counters.Range(func(key, value interface{}) bool {
+		counter := value.(*int64)
+		beforeCount = atomic.LoadInt64(counter)
+		return true
+	})
+	assert.Equal(t, int64(2), beforeCount)
+
+	buffer.flush()
+
+	var afterCount int64
+	buffer.counters.Range(func(key, value interface{}) bool {
+		counter := value.(*int64)
+		afterCount = atomic.LoadInt64(counter)
+		return true
+	})
+	assert.Equal(t, int64(0), afterCount)
+}
+
+func TestStatsBuffer_FailRetry(t *testing.T) {
+	logger := zap.NewNop()
+	statsRepo := &mockStatsRepo{fail: true}
+	buffer := NewStatsBuffer(statsRepo, logger)
+
+	buffer.Increment("openai", "gpt-4")
+	buffer.Increment("openai", "gpt-4")
+
+	buffer.flush()
+
+	var count int64
+	buffer.counters.Range(func(key, value interface{}) bool {
+		counter := value.(*int64)
+		count = atomic.LoadInt64(counter)
+		return true
+	})
+	assert.Equal(t, int64(2), count)
+}
+
+func TestStatsBuffer_Stop(t *testing.T) {
+	logger := zap.NewNop()
+	statsRepo := &mockStatsRepo{}
+	buffer := NewStatsBuffer(statsRepo, logger,
+		WithFlushInterval(10*time.Second))
+
+	buffer.Start()
+
+	buffer.Increment("openai", "gpt-4")
+	buffer.Increment("openai", "gpt-4")
+
+	start := time.Now()
+	buffer.Stop()
+	elapsed := time.Since(start)
+
+	assert.Less(t, elapsed, 1*time.Second)
+
+	statsRepo.mu.Lock()
+	assert.GreaterOrEqual(t, len(statsRepo.records), 1)
+	statsRepo.mu.Unlock()
+}
+
+func TestStatsBuffer_ConcurrentIncrementAndFlush(t *testing.T) {
+	logger := zap.NewNop()
+	statsRepo := &mockStatsRepo{}
+	buffer := NewStatsBuffer(statsRepo, logger,
+		WithFlushInterval(50*time.Millisecond))
+
+	buffer.Start()
+	defer buffer.Stop()
+
+	var wg sync.WaitGroup
+	for i := 0; i < 100; i++ {
+		wg.Add(1)
+		go func() {
+			defer wg.Done()
+			buffer.Increment("openai", "gpt-4")
+		}()
+	}
+	wg.Wait()
+
+	time.Sleep(100 * time.Millisecond)
+
+	statsRepo.mu.Lock()
+	totalDelta := 0
+	for _, r := range statsRepo.records {
+		totalDelta += r.delta
+	}
+	statsRepo.mu.Unlock()
+
+	assert.Equal(t, 100, totalDelta)
+}

backend/internal/service/stats_service_impl.go

@@ -10,14 +10,16 @@ import (
 
 type statsService struct {
 	statsRepo repository.StatsRepository
+	buffer    *StatsBuffer
 }
 
-func NewStatsService(statsRepo repository.StatsRepository) StatsService {
-	return &statsService{statsRepo: statsRepo}
+func NewStatsService(statsRepo repository.StatsRepository, buffer *StatsBuffer) StatsService {
+	return &statsService{statsRepo: statsRepo, buffer: buffer}
 }
 
 func (s *statsService) Record(providerID, modelName string) error {
-	return s.statsRepo.Record(providerID, modelName)
+	s.buffer.Increment(providerID, modelName)
+	return nil
 }
 
 func (s *statsService) Get(providerID, modelName string, startDate, endDate *time.Time) ([]domain.UsageStats, error) {

backend/tests/integration/conversion_test.go

@@ -14,6 +14,7 @@ import (
 	"github.com/gin-gonic/gin"
 	"github.com/stretchr/testify/assert"
 	"github.com/stretchr/testify/require"
+	"go.uber.org/zap"
 	"gorm.io/gorm"
 
 	"nex/backend/internal/conversion"
@@ -54,10 +55,14 @@ func setupConversionTest(t *testing.T) (*gin.Engine, *gorm.DB, *httptest.Server)
 	modelRepo := repository.NewModelRepository(db)
 	statsRepo := repository.NewStatsRepository(db)
 
-	providerService := service.NewProviderService(providerRepo, modelRepo)
-	modelService := service.NewModelService(modelRepo, providerRepo)
-	routingService := service.NewRoutingService(modelRepo, providerRepo)
-	statsService := service.NewStatsService(statsRepo)
+	logger := zap.NewNop()
+	routingCache := service.NewRoutingCache(modelRepo, providerRepo, logger)
+	statsBuffer := service.NewStatsBuffer(statsRepo, logger)
+
+	providerService := service.NewProviderService(providerRepo, modelRepo, routingCache)
+	modelService := service.NewModelService(modelRepo, providerRepo, routingCache)
+	routingService := service.NewRoutingService(routingCache)
+	statsService := service.NewStatsService(statsRepo, statsBuffer)
 
 	// 创建 ConversionEngine
 	registry := conversion.NewMemoryRegistry()

backend/tests/integration/e2e_conversion_test.go

@@ -15,6 +15,7 @@ import (
 	"github.com/gin-gonic/gin"
 	"github.com/stretchr/testify/assert"
 	"github.com/stretchr/testify/require"
+	"go.uber.org/zap"
 
 	"nex/backend/internal/conversion"
 	"nex/backend/internal/conversion/anthropic"
@@ -48,10 +49,14 @@ func setupE2ETest(t *testing.T) (*gin.Engine, *httptest.Server) {
 	modelRepo := repository.NewModelRepository(db)
 	statsRepo := repository.NewStatsRepository(db)
 
-	providerService := service.NewProviderService(providerRepo, modelRepo)
-	modelService := service.NewModelService(modelRepo, providerRepo)
-	routingService := service.NewRoutingService(modelRepo, providerRepo)
-	statsService := service.NewStatsService(statsRepo)
+	logger := zap.NewNop()
+	routingCache := service.NewRoutingCache(modelRepo, providerRepo, logger)
+	statsBuffer := service.NewStatsBuffer(statsRepo, logger)
+
+	providerService := service.NewProviderService(providerRepo, modelRepo, routingCache)
+	modelService := service.NewModelService(modelRepo, providerRepo, routingCache)
+	routingService := service.NewRoutingService(routingCache)
+	statsService := service.NewStatsService(statsRepo, statsBuffer)
 
 	registry := conversion.NewMemoryRegistry()
 	require.NoError(t, registry.Register(openaiConv.NewAdapter()))

backend/tests/integration/integration_test.go

@@ -9,6 +9,7 @@ import (
 
 	"github.com/gin-gonic/gin"
 	"github.com/stretchr/testify/assert"
+	"go.uber.org/zap"
 	"gorm.io/gorm"
 
 	"nex/backend/internal/domain"
@@ -30,10 +31,14 @@ func setupIntegrationTest(t *testing.T) (*gin.Engine, *gorm.DB) {
 	modelRepo := repository.NewModelRepository(db)
 	statsRepo := repository.NewStatsRepository(db)
 
-	providerService := service.NewProviderService(providerRepo, modelRepo)
-	modelService := service.NewModelService(modelRepo, providerRepo)
-	_ = service.NewRoutingService(modelRepo, providerRepo)
-	statsService := service.NewStatsService(statsRepo)
+	logger := zap.NewNop()
+	routingCache := service.NewRoutingCache(modelRepo, providerRepo, logger)
+	statsBuffer := service.NewStatsBuffer(statsRepo, logger)
+
+	providerService := service.NewProviderService(providerRepo, modelRepo, routingCache)
+	modelService := service.NewModelService(modelRepo, providerRepo, routingCache)
+	_ = service.NewRoutingService(routingCache)
+	statsService := service.NewStatsService(statsRepo, statsBuffer)
 
 	providerHandler := handler.NewProviderHandler(providerService)
 	modelHandler := handler.NewModelHandler(modelService)

backend/tests/mocks/mock_stats_repository.go

@@ -41,6 +41,20 @@ func (m *MockStatsRepository) EXPECT() *MockStatsRepositoryMockRecorder {
 	return m.recorder
 }
 
+// BatchUpdate mocks base method.
+func (m *MockStatsRepository) BatchUpdate(providerID, modelName string, date time.Time, delta int) error {
+	m.ctrl.T.Helper()
+	ret := m.ctrl.Call(m, "BatchUpdate", providerID, modelName, date, delta)
+	ret0, _ := ret[0].(error)
+	return ret0
+}
+
+// BatchUpdate indicates an expected call of BatchUpdate.
+func (mr *MockStatsRepositoryMockRecorder) BatchUpdate(providerID, modelName, date, delta any) *gomock.Call {
+	mr.mock.ctrl.T.Helper()
+	return mr.mock.ctrl.RecordCallWithMethodType(mr.mock, "BatchUpdate", reflect.TypeOf((*MockStatsRepository)(nil).BatchUpdate), providerID, modelName, date, delta)
+}
+
 // Query mocks base method.
 func (m *MockStatsRepository) Query(providerID, modelName string, startDate, endDate *time.Time) ([]domain.UsageStats, error) {
 	m.ctrl.T.Helper()

openspec/specs/routing-cache/spec.md

@@ -0,0 +1,136 @@
+# Routing Cache
+
+## Purpose
+
+TBD - 为 Provider 和 Model 提供内存缓存，优化路由查询性能。
+
+## Requirements
+
+### Requirement: RoutingCache 缓存 Provider 和 Model
+
+系统 SHALL 为 Provider 和 Model 提供内存缓存，使用 sync.Map 作为缓存数据结构。
+
+#### Scenario: 缓存数据结构
+
+- **WHEN** 创建 RoutingCache
+- **THEN** SHALL 使用 sync.Map 存储 Provider（key = providerID）
+- **THEN** SHALL 使用 sync.Map 存储 Model（key = providerID/modelName）
+
+### Requirement: 缓存查询优先
+
+RoutingCache SHALL 优先从缓存查询，缓存未命中时查询数据库并更新缓存。
+
+#### Scenario: 缓存命中
+
+- **WHEN** 查询 Provider 或 Model
+- **THEN** SHALL 先从缓存查询
+- **THEN** 如果缓存命中，SHALL 直接返回缓存值
+- **THEN** SHALL 不查询数据库
+
+#### Scenario: 缓存未命中
+
+- **WHEN** 查询 Provider 或 Model
+- **THEN** SHALL 先从缓存查询
+- **THEN** 如果缓存未命中，SHALL 查询数据库
+- **THEN** SHALL 将查询结果存入缓存
+- **THEN** SHALL 返回查询结果
+
+#### Scenario: 双重检查防止竞态
+
+- **WHEN** 并发查询同一个 Provider 或 Model
+- **THEN** SHALL 在查询数据库后再次检查缓存
+- **THEN** 如果已有其他 goroutine 存入缓存，SHALL 使用缓存值
+- **THEN** SHALL 防止存入旧值
+
+### Requirement: 缓存更新
+
+RoutingCache SHALL 支持 Create 操作后更新缓存。
+
+#### Scenario: Create Provider 后更新缓存
+
+- **WHEN** 创建 Provider 成功
+- **THEN** SHALL 调用 RoutingCache.SetProvider
+- **THEN** SHALL 将新 Provider 存入缓存
+
+#### Scenario: Create Model 后更新缓存
+
+- **WHEN** 创建 Model 成功
+- **THEN** SHALL 调用 RoutingCache.SetModel
+- **THEN** SHALL 将新 Model 存入缓存
+
+### Requirement: 缓存失效
+
+RoutingCache SHALL 支持 Update/Delete 操作后清除缓存。
+
+#### Scenario: Update Provider 后清除缓存
+
+- **WHEN** 更新 Provider 成功
+- **THEN** SHALL 调用 RoutingCache.InvalidateProvider
+- **THEN** SHALL 清除该 Provider 的缓存
+- **THEN** SHALL 级联清除该 Provider 的所有 Model 缓存
+
+#### Scenario: Delete Provider 后清除缓存
+
+- **WHEN** 删除 Provider 成功
+- **THEN** SHALL 调用 RoutingCache.InvalidateProvider
+- **THEN** SHALL 清除该 Provider 的缓存
+- **THEN** SHALL 级联清除该 Provider 的所有 Model 缓存
+
+#### Scenario: Update Model 后清除缓存
+
+- **WHEN** 更新 Model 成功
+- **THEN** SHALL 调用 RoutingCache.InvalidateModel
+- **THEN** SHALL 清除旧位置的 Model 缓存
+- **THEN** 如果 provider_id 或 model_name 变化，SHALL 也清除新位置的缓存
+
+#### Scenario: Delete Model 后清除缓存
+
+- **WHEN** 删除 Model 成功
+- **THEN** SHALL 调用 RoutingCache.InvalidateModel
+- **THEN** SHALL 清除该 Model 的缓存
+
+### Requirement: 缓存预热
+
+RoutingCache SHALL 支持启动时预热缓存。
+
+#### Scenario: 预热成功
+
+- **WHEN** 服务启动时
+- **THEN** SHALL 调用 RoutingCache.Preload
+- **THEN** SHALL 从数据库加载所有 Provider 到缓存
+- **THEN** SHALL 从数据库加载所有 Model 到缓存
+- **THEN** SHALL 记录预热完成的日志
+
+#### Scenario: 预热失败
+
+- **WHEN** 预热失败时
+- **THEN** SHALL 记录警告日志
+- **THEN** SHALL 继续启动服务
+- **THEN** SHALL 使用懒加载（首次查询时加载）
+
+### Requirement: RoutingService 使用缓存
+
+RoutingService SHALL 使用 RoutingCache 进行路由查询。
+
+#### Scenario: RouteByModelName 使用缓存
+
+- **WHEN** 调用 RoutingService.RouteByModelName
+- **THEN** SHALL 调用 RoutingCache.GetModel 获取 Model
+- **THEN** SHALL 调用 RoutingCache.GetProvider 获取 Provider
+- **THEN** SHALL 不直接调用 Repository
+
+### Requirement: 并发安全
+
+RoutingCache SHALL 支持并发访问。
+
+#### Scenario: 并发查询
+
+- **WHEN** 多个 goroutine 并发查询缓存
+- **THEN** SHALL 无竞态条件
+- **THEN** SHALL 无 panic
+
+#### Scenario: 并发查询和失效
+
+- **WHEN** 并发查询和失效缓存
+- **THEN** SHALL 无竞态条件
+- **THEN** SHALL 保证一致性

openspec/specs/stats-buffer/spec.md

@@ -0,0 +1,156 @@
+# Stats Buffer
+
+## Purpose
+
+TBD - 为统计数据提供内存缓冲，优化写入性能。
+
+## Requirements
+
+### Requirement: StatsBuffer 内存缓冲
+
+系统 SHALL 为统计数据提供内存缓冲，使用 sync.Map + atomic.Int64 进行计数。
+
+#### Scenario: 缓冲数据结构
+
+- **WHEN** 创建 StatsBuffer
+- **THEN** SHALL 使用 sync.Map 存储计数器（key = providerID/modelName/date）
+- **THEN** SHALL 使用 atomic.Int64 进行计数
+- **THEN** SHALL 支持配置刷新间隔和阈值
+
+### Requirement: 原子计数
+
+StatsBuffer SHALL 使用原子操作进行计数。
+
+#### Scenario: Increment 计数
+
+- **WHEN** 调用 StatsBuffer.Increment
+- **THEN** SHALL 使用 atomic.AddInt64 增加计数
+- **THEN** SHALL 无锁操作
+- **THEN** SHALL 线程安全
+
+#### Scenario: Increment 创建计数器
+
+- **WHEN** 调用 StatsBuffer.Increment 且计数器不存在
+- **THEN** SHALL 使用 sync.Map.LoadOrStore 创建计数器
+- **THEN** SHALL 初始化计数器为 0
+- **THEN** SHALL 原子增加计数
+
+#### Scenario: 并发计数
+
+- **WHEN** 多个 goroutine 并发 Increment
+- **THEN** SHALL 计数准确
+- **THEN** SHALL 无竞态条件
+
+### Requirement: 定时刷新
+
+StatsBuffer SHALL 支持定时刷新到数据库。
+
+#### Scenario: 定时刷新触发
+
+- **WHEN** 后台刷新协程运行
+- **THEN** SHALL 每隔 flushInterval 触发刷新
+- **THEN** SHALL 调用 StatsRepository.BatchUpdate 写入数据库
+
+#### Scenario: 刷新间隔配置
+
+- **WHEN** 创建 StatsBuffer
+- **THEN** 默认 flushInterval 为 5 秒
+- **THEN** 可通过 WithFlushInterval 选项配置
+
+### Requirement: 阈值触发刷新
+
+StatsBuffer SHALL 支持累计阈值触发刷新。
+
+#### Scenario: 阈值触发
+
+- **WHEN** 累计计数达到 flushThreshold
+- **THEN** SHALL 异步触发刷新
+- **THEN** SHALL 不阻塞请求
+
+#### Scenario: 阈值配置
+
+- **WHEN** 创建 StatsBuffer
+- **THEN** 默认 flushThreshold 为 100
+- **THEN** 可通过 WithFlushThreshold 选项配置
+
+### Requirement: 批量写入数据库
+
+StatsBuffer SHALL 批量写入统计数据到数据库。
+
+#### Scenario: 批量写入
+
+- **WHEN** 刷新触发
+- **THEN** SHALL 遍历所有计数器
+- **THEN** SHALL 使用 atomic.SwapInt64(counter, 0) 获取并清零计数器
+- **THEN** SHALL 调用 StatsRepository.BatchUpdate 批量写入
+- **THEN** SHALL 重置 totalCount 为 0
+
+#### Scenario: SwapInt64 清零计数器
+
+- **WHEN** flush 收集计数器
+- **THEN** SHALL 使用 SwapInt64 原子操作获取当前计数并清零
+- **THEN** SHALL 保证计数不丢失（新计数会累加到已清零的计数器）
+- **THEN** SHALL 不阻塞后续 Increment 操作
+
+#### Scenario: 写入失败保留计数器
+
+- **WHEN** BatchUpdate 失败
+- **THEN** SHALL 将计数加回计数器（使用 atomic.AddInt64）
+- **THEN** SHALL 记录错误日志
+- **THEN** SHALL 继续处理其他条目
+
+### Requirement: 优雅关闭
+
+StatsBuffer SHALL 支持优雅关闭，确保最后的统计写入数据库。
+
+#### Scenario: Stop 等待刷新完成
+
+- **WHEN** 调用 StatsBuffer.Stop
+- **THEN** SHALL 停止后台刷新协程
+- **THEN** SHALL 执行最后一次刷新
+- **THEN** SHALL 等待刷新完成
+- **THEN** SHALL 保证统计数据不丢失
+
+#### Scenario: 无超时
+
+- **WHEN** Stop 等待刷新
+- **THEN** SHALL 无超时限制
+- **THEN** SHALL 等待刷新完成
+
+### Requirement: StatsService 使用缓冲
+
+StatsService SHALL 使用 StatsBuffer 进行统计记录。
+
+#### Scenario: Record 使用缓冲
+
+- **WHEN** 调用 StatsService.Record
+- **THEN** SHALL 调用 StatsBuffer.Increment
+- **THEN** SHALL 不直接调用 StatsRepository.Record
+- **THEN** SHALL 立即返回，不阻塞
+
+### Requirement: StatsRepository 扩展
+
+StatsRepository SHALL 新增 BatchUpdate 方法。
+
+#### Scenario: BatchUpdate 方法
+
+- **WHEN** 调用 StatsRepository.BatchUpdate
+- **THEN** SHALL 使用事务更新或创建统计记录
+- **THEN** SHALL 使用 request_count + delta 更新
+- **THEN** SHALL 支持批量增量更新
+
+### Requirement: 并发安全
+
+StatsBuffer SHALL 支持并发访问。
+
+#### Scenario: 并发 Increment 和 flush
+
+- **WHEN** 并发 Increment 和 flush
+- **THEN** SHALL 无竞态条件
+- **THEN** SHALL 计数准确（可能延迟到下次 flush）
+
+#### Scenario: flush 期间 Increment
+
+- **WHEN** flush 正在执行
+- **THEN** 新的 Increment SHALL 继续计数
+- **THEN** SHALL 不会丢失计数

openspec/specs/usage-statistics/spec.md

@@ -13,14 +13,15 @@
 #### Scenario: 记录成功请求
 
 - **WHEN** 请求成功转发到供应商
-- **THEN** 网关 SHALL 增加该供应商和模型的请求计数
-- **THEN** 网关 SHALL 记录当前日期的统计
+- **THEN** 网关 SHALL 通过 StatsBuffer 增加该供应商和模型的请求计数
+- **THEN** 网关 SHALL 异步批量写入数据库（定时或阈值触发）
+- **THEN** 网关 SHALL 不阻塞响应
 
 #### Scenario: 记录流式请求
 
 - **WHEN** 流式请求成功完成
-- **THEN** 网关 SHALL 增加该供应商和模型的请求计数
-- **THEN** 网关 SHALL 在流结束后记录统计
+- **THEN** 网关 SHALL 通过 StatsBuffer 增加该供应商和模型的请求计数
+- **THEN** 网关 SHALL 在流结束后异步记录统计
 
 ### Requirement: 使用统计记录统一模型标识
 
@@ -90,10 +91,9 @@
 #### Scenario: 并发请求
 
 - **WHEN** 同时处理多个并发请求
-- **THEN** 网关 SHALL 正确为每个请求增加请求计数
+- **THEN** 网关 SHALL 使用原子操作正确增加每个请求的计数
 - **THEN** 不 SHALL 因并发写入而丢失统计
-
-**变更说明：** 并发控制在 StatsRepository 中通过数据库事务实现。API 接口保持不变。
+- **THEN** SHALL 使用 StatsBuffer 的内存计数器
 
 ### Requirement: 使用 service 层处理业务逻辑
 
@@ -121,6 +121,13 @@ Service SHALL 通过 StatsRepository 访问数据。
 - **THEN** SHALL 调用对应的 StatsRepository 方法
 - **THEN** SHALL 使用 domain.UsageStats 类型
 
+#### Scenario: 批量更新统计
+
+- **WHEN** StatsBuffer 刷新统计
+- **THEN** SHALL 调用 StatsRepository.BatchUpdate
+- **THEN** SHALL 使用事务更新或创建统计记录
+- **THEN** SHALL 支持增量更新（request_count + delta）
+
 #### Scenario: 事务处理
 
 - **WHEN** 记录统计
@@ -136,3 +143,36 @@ Service SHALL 通过 StatsRepository 访问数据。
 - **WHEN** 查询统计
 - **THEN** SHALL 使用 (provider_id, model_name, date) 复合索引
 - **THEN** SHALL 优化查询性能
+
+### Requirement: 统计数据可接受少量丢失
+
+统计记录方式改为内存缓冲，可接受少量丢失。
+
+#### Scenario: 进程崩溃丢失统计
+
+- **WHEN** 进程崩溃
+- **THEN** MAY 丢失最近 flushInterval 内的统计
+- **THEN** 统计数据用于监控，可接受少量丢失
+
+#### Scenario: 优雅关闭保证不丢失
+
+- **WHEN** 服务优雅关闭
+- **THEN** SHALL 调用 StatsBuffer.Stop
+- **THEN** SHALL 等待最后一次刷新完成
+- **THEN** SHALL 保证统计数据不丢失
+
+### Requirement: StatsRepository 支持批量更新
+
+StatsRepository SHALL 新增 BatchUpdate 方法支持批量增量更新。
+
+#### Scenario: BatchUpdate 更新现有记录
+
+- **WHEN** 调用 BatchUpdate 且当日记录存在
+- **THEN** SHALL 使用事务更新 request_count = request_count + delta
+- **THEN** SHALL 不创建新记录
+
+#### Scenario: BatchUpdate 创建新记录
+
+- **WHEN** 调用 BatchUpdate 且当日记录不存在
+- **THEN** SHALL 创建新记录，request_count = delta
+- **THEN** SHALL 使用事务保证原子性