Merge branch 'main' of https://github.com/mt21625457/aicodex2api

backend/internal/service/account.go

@@ -576,6 +576,44 @@ func (a *Account) IsAnthropicOAuthOrSetupToken() bool {
 	return a.Platform == PlatformAnthropic && (a.Type == AccountTypeOAuth || a.Type == AccountTypeSetupToken)
 }
 
+// IsTLSFingerprintEnabled 检查是否启用 TLS 指纹伪装
+// 仅适用于 Anthropic OAuth/SetupToken 类型账号
+// 启用后将模拟 Claude Code (Node.js) 客户端的 TLS 握手特征
+func (a *Account) IsTLSFingerprintEnabled() bool {
+	// 仅支持 Anthropic OAuth/SetupToken 账号
+	if !a.IsAnthropicOAuthOrSetupToken() {
+		return false
+	}
+	if a.Extra == nil {
+		return false
+	}
+	if v, ok := a.Extra["enable_tls_fingerprint"]; ok {
+		if enabled, ok := v.(bool); ok {
+			return enabled
+		}
+	}
+	return false
+}
+
+// IsSessionIDMaskingEnabled 检查是否启用会话ID伪装
+// 仅适用于 Anthropic OAuth/SetupToken 类型账号
+// 启用后将在一段时间内（15分钟）固定 metadata.user_id 中的 session ID，
+// 使上游认为请求来自同一个会话
+func (a *Account) IsSessionIDMaskingEnabled() bool {
+	if !a.IsAnthropicOAuthOrSetupToken() {
+		return false
+	}
+	if a.Extra == nil {
+		return false
+	}
+	if v, ok := a.Extra["session_id_masking_enabled"]; ok {
+		if enabled, ok := v.(bool); ok {
+			return enabled
+		}
+	}
+	return false
+}
+
 // GetWindowCostLimit 获取 5h 窗口费用阈值（美元）
 // 返回 0 表示未启用
 func (a *Account) GetWindowCostLimit() float64 {
@@ -652,6 +690,23 @@ func (a *Account) CheckWindowCostSchedulability(currentWindowCost float64) Windo
 	return WindowCostNotSchedulable
 }
 
+// GetCurrentWindowStartTime 获取当前有效的窗口开始时间
+// 逻辑：
+// 1. 如果窗口未过期（SessionWindowEnd 存在且在当前时间之后），使用记录的 SessionWindowStart
+// 2. 否则（窗口过期或未设置），使用新的预测窗口开始时间（从当前整点开始）
+func (a *Account) GetCurrentWindowStartTime() time.Time {
+	now := time.Now()
+
+	// 窗口未过期，使用记录的窗口开始时间
+	if a.SessionWindowStart != nil && a.SessionWindowEnd != nil && now.Before(*a.SessionWindowEnd) {
+		return *a.SessionWindowStart
+	}
+
+	// 窗口已过期或未设置，预测新的窗口开始时间（从当前整点开始）
+	// 与 ratelimit_service.go 中 UpdateSessionWindow 的预测逻辑保持一致
+	return time.Date(now.Year(), now.Month(), now.Day(), now.Hour(), 0, 0, 0, now.Location())
+}
+
 // parseExtraFloat64 从 extra 字段解析 float64 值
 func parseExtraFloat64(value any) float64 {
 	switch v := value.(type) {

backend/internal/service/account_service.go

@@ -37,6 +37,7 @@ type AccountRepository interface {
 	UpdateLastUsed(ctx context.Context, id int64) error
 	BatchUpdateLastUsed(ctx context.Context, updates map[int64]time.Time) error
 	SetError(ctx context.Context, id int64, errorMsg string) error
+	ClearError(ctx context.Context, id int64) error
 	SetSchedulable(ctx context.Context, id int64, schedulable bool) error
 	AutoPauseExpiredAccounts(ctx context.Context, now time.Time) (int64, error)
 	BindGroups(ctx context.Context, accountID int64, groupIDs []int64) error

backend/internal/service/account_service_delete_test.go

@@ -99,6 +99,10 @@ func (s *accountRepoStub) SetError(ctx context.Context, id int64, errorMsg strin
 	panic("unexpected SetError call")
 }
 
+func (s *accountRepoStub) ClearError(ctx context.Context, id int64) error {
+	panic("unexpected ClearError call")
+}
+
 func (s *accountRepoStub) SetSchedulable(ctx context.Context, id int64, schedulable bool) error {
 	panic("unexpected SetSchedulable call")
 }

backend/internal/service/account_test_service.go

@@ -265,7 +265,7 @@ func (s *AccountTestService) testClaudeAccountConnection(c *gin.Context, account
 		proxyURL = account.Proxy.URL()
 	}
 
-	resp, err := s.httpUpstream.Do(req, proxyURL, account.ID, account.Concurrency)
+	resp, err := s.httpUpstream.DoWithTLS(req, proxyURL, account.ID, account.Concurrency, account.IsTLSFingerprintEnabled())
 	if err != nil {
 		return s.sendErrorAndEnd(c, fmt.Sprintf("Request failed: %s", err.Error()))
 	}
@@ -375,7 +375,7 @@ func (s *AccountTestService) testOpenAIAccountConnection(c *gin.Context, account
 		proxyURL = account.Proxy.URL()
 	}
 
-	resp, err := s.httpUpstream.Do(req, proxyURL, account.ID, account.Concurrency)
+	resp, err := s.httpUpstream.DoWithTLS(req, proxyURL, account.ID, account.Concurrency, account.IsTLSFingerprintEnabled())
 	if err != nil {
 		return s.sendErrorAndEnd(c, fmt.Sprintf("Request failed: %s", err.Error()))
 	}
@@ -446,7 +446,7 @@ func (s *AccountTestService) testGeminiAccountConnection(c *gin.Context, account
 		proxyURL = account.Proxy.URL()
 	}
 
-	resp, err := s.httpUpstream.Do(req, proxyURL, account.ID, account.Concurrency)
+	resp, err := s.httpUpstream.DoWithTLS(req, proxyURL, account.ID, account.Concurrency, account.IsTLSFingerprintEnabled())
 	if err != nil {
 		return s.sendErrorAndEnd(c, fmt.Sprintf("Request failed: %s", err.Error()))
 	}

backend/internal/service/account_usage_service.go

@@ -369,12 +369,8 @@ func (s *AccountUsageService) addWindowStats(ctx context.Context, account *Accou
 
 	// 如果没有缓存，从数据库查询
 	if windowStats == nil {
-		var startTime time.Time
-		if account.SessionWindowStart != nil {
-			startTime = *account.SessionWindowStart
-		} else {
-			startTime = time.Now().Add(-5 * time.Hour)
-		}
+		// 使用统一的窗口开始时间计算逻辑（考虑窗口过期情况）
+		startTime := account.GetCurrentWindowStartTime()
 
 		stats, err := s.usageLogRepo.GetAccountWindowStats(ctx, account.ID, startTime)
 		if err != nil {

backend/internal/service/admin_service.go

@@ -42,6 +42,7 @@ type AdminService interface {
 	DeleteAccount(ctx context.Context, id int64) error
 	RefreshAccountCredentials(ctx context.Context, id int64) (*Account, error)
 	ClearAccountError(ctx context.Context, id int64) (*Account, error)
+	SetAccountError(ctx context.Context, id int64, errorMsg string) error
 	SetAccountSchedulable(ctx context.Context, id int64, schedulable bool) (*Account, error)
 	BulkUpdateAccounts(ctx context.Context, input *BulkUpdateAccountsInput) (*BulkUpdateAccountsResult, error)
 
@@ -1101,6 +1102,10 @@ func (s *adminServiceImpl) ClearAccountError(ctx context.Context, id int64) (*Ac
 	return account, nil
 }
 
+func (s *adminServiceImpl) SetAccountError(ctx context.Context, id int64, errorMsg string) error {
+	return s.accountRepo.SetError(ctx, id, errorMsg)
+}
+
 func (s *adminServiceImpl) SetAccountSchedulable(ctx context.Context, id int64, schedulable bool) (*Account, error) {
 	if err := s.accountRepo.SetSchedulable(ctx, id, schedulable); err != nil {
 		return nil, err

backend/internal/service/antigravity_oauth_service.go

@@ -82,13 +82,14 @@ type AntigravityExchangeCodeInput struct {
 
 // AntigravityTokenInfo token 信息
 type AntigravityTokenInfo struct {
-	AccessToken  string `json:"access_token"`
-	RefreshToken string `json:"refresh_token"`
-	ExpiresIn    int64  `json:"expires_in"`
-	ExpiresAt    int64  `json:"expires_at"`
-	TokenType    string `json:"token_type"`
-	Email        string `json:"email,omitempty"`
-	ProjectID    string `json:"project_id,omitempty"`
+	AccessToken      string `json:"access_token"`
+	RefreshToken     string `json:"refresh_token"`
+	ExpiresIn        int64  `json:"expires_in"`
+	ExpiresAt        int64  `json:"expires_at"`
+	TokenType        string `json:"token_type"`
+	Email            string `json:"email,omitempty"`
+	ProjectID        string `json:"project_id,omitempty"`
+	ProjectIDMissing bool   `json:"-"` // LoadCodeAssist 未返回 project_id
 }
 
 // ExchangeCode 用 authorization code 交换 token
@@ -149,12 +150,6 @@ func (s *AntigravityOAuthService) ExchangeCode(ctx context.Context, input *Antig
 		result.ProjectID = loadResp.CloudAICompanionProject
 	}
 
-	// 兜底：随机生成 project_id
-	if result.ProjectID == "" {
-		result.ProjectID = antigravity.GenerateMockProjectID()
-		fmt.Printf("[AntigravityOAuth] 使用随机生成的 project_id: %s\n", result.ProjectID)
-	}
-
 	return result, nil
 }
 
@@ -236,16 +231,24 @@ func (s *AntigravityOAuthService) RefreshAccountToken(ctx context.Context, accou
 		return nil, err
 	}
 
-	// 保留原有的 project_id 和 email
-	existingProjectID := strings.TrimSpace(account.GetCredential("project_id"))
-	if existingProjectID != "" {
-		tokenInfo.ProjectID = existingProjectID
-	}
+	// 保留原有的 email
 	existingEmail := strings.TrimSpace(account.GetCredential("email"))
 	if existingEmail != "" {
 		tokenInfo.Email = existingEmail
 	}
 
+	// 每次刷新都调用 LoadCodeAssist 获取 project_id
+	client := antigravity.NewClient(proxyURL)
+	loadResp, _, err := client.LoadCodeAssist(ctx, tokenInfo.AccessToken)
+	if err != nil || loadResp == nil || loadResp.CloudAICompanionProject == "" {
+		// LoadCodeAssist 失败或返回空，保留原有 project_id，标记缺失
+		existingProjectID := strings.TrimSpace(account.GetCredential("project_id"))
+		tokenInfo.ProjectID = existingProjectID
+		tokenInfo.ProjectIDMissing = true
+	} else {
+		tokenInfo.ProjectID = loadResp.CloudAICompanionProject
+	}
+
 	return tokenInfo, nil
 }
 

backend/internal/service/antigravity_quota_fetcher.go

@@ -31,11 +31,6 @@ func (f *AntigravityQuotaFetcher) FetchQuota(ctx context.Context, account *Accou
 	accessToken := account.GetCredential("access_token")
 	projectID := account.GetCredential("project_id")
 
-	// 如果没有 project_id，生成一个随机的
-	if projectID == "" {
-		projectID = antigravity.GenerateMockProjectID()
-	}
-
 	client := antigravity.NewClient(proxyURL)
 
 	// 调用 API 获取配额

backend/internal/service/antigravity_rate_limit_test.go

@@ -0,0 +1,190 @@
+//go:build unit
+
+package service
+
+import (
+	"context"
+	"fmt"
+	"io"
+	"net/http"
+	"strings"
+	"testing"
+	"time"
+
+	"github.com/Wei-Shaw/sub2api/internal/pkg/antigravity"
+	"github.com/stretchr/testify/require"
+)
+
+type stubAntigravityUpstream struct {
+	firstBase  string
+	secondBase string
+	calls      []string
+}
+
+func (s *stubAntigravityUpstream) Do(req *http.Request, proxyURL string, accountID int64, accountConcurrency int) (*http.Response, error) {
+	url := req.URL.String()
+	s.calls = append(s.calls, url)
+	if strings.HasPrefix(url, s.firstBase) {
+		return &http.Response{
+			StatusCode: http.StatusTooManyRequests,
+			Header:     http.Header{},
+			Body:       io.NopCloser(strings.NewReader(`{"error":{"message":"Resource has been exhausted"}}`)),
+		}, nil
+	}
+	return &http.Response{
+		StatusCode: http.StatusOK,
+		Header:     http.Header{},
+		Body:       io.NopCloser(strings.NewReader("ok")),
+	}, nil
+}
+
+func (s *stubAntigravityUpstream) DoWithTLS(req *http.Request, proxyURL string, accountID int64, accountConcurrency int, enableTLSFingerprint bool) (*http.Response, error) {
+	return s.Do(req, proxyURL, accountID, accountConcurrency)
+}
+
+type scopeLimitCall struct {
+	accountID int64
+	scope     AntigravityQuotaScope
+	resetAt   time.Time
+}
+
+type rateLimitCall struct {
+	accountID int64
+	resetAt   time.Time
+}
+
+type stubAntigravityAccountRepo struct {
+	AccountRepository
+	scopeCalls []scopeLimitCall
+	rateCalls  []rateLimitCall
+}
+
+func (s *stubAntigravityAccountRepo) SetAntigravityQuotaScopeLimit(ctx context.Context, id int64, scope AntigravityQuotaScope, resetAt time.Time) error {
+	s.scopeCalls = append(s.scopeCalls, scopeLimitCall{accountID: id, scope: scope, resetAt: resetAt})
+	return nil
+}
+
+func (s *stubAntigravityAccountRepo) SetRateLimited(ctx context.Context, id int64, resetAt time.Time) error {
+	s.rateCalls = append(s.rateCalls, rateLimitCall{accountID: id, resetAt: resetAt})
+	return nil
+}
+
+func TestAntigravityRetryLoop_URLFallback_UsesLatestSuccess(t *testing.T) {
+	oldBaseURLs := append([]string(nil), antigravity.BaseURLs...)
+	oldAvailability := antigravity.DefaultURLAvailability
+	defer func() {
+		antigravity.BaseURLs = oldBaseURLs
+		antigravity.DefaultURLAvailability = oldAvailability
+	}()
+
+	base1 := "https://ag-1.test"
+	base2 := "https://ag-2.test"
+	antigravity.BaseURLs = []string{base1, base2}
+	antigravity.DefaultURLAvailability = antigravity.NewURLAvailability(time.Minute)
+
+	upstream := &stubAntigravityUpstream{firstBase: base1, secondBase: base2}
+	account := &Account{
+		ID:          1,
+		Name:        "acc-1",
+		Platform:    PlatformAntigravity,
+		Schedulable: true,
+		Status:      StatusActive,
+		Concurrency: 1,
+	}
+
+	var handleErrorCalled bool
+	result, err := antigravityRetryLoop(antigravityRetryLoopParams{
+		prefix:      "[test]",
+		ctx:         context.Background(),
+		account:     account,
+		proxyURL:    "",
+		accessToken: "token",
+		action:      "generateContent",
+		body:        []byte(`{"input":"test"}`),
+		quotaScope:  AntigravityQuotaScopeClaude,
+		httpUpstream: upstream,
+		handleError: func(ctx context.Context, prefix string, account *Account, statusCode int, headers http.Header, body []byte, quotaScope AntigravityQuotaScope) {
+			handleErrorCalled = true
+		},
+	})
+
+	require.NoError(t, err)
+	require.NotNil(t, result)
+	require.NotNil(t, result.resp)
+	defer func() { _ = result.resp.Body.Close() }()
+	require.Equal(t, http.StatusOK, result.resp.StatusCode)
+	require.False(t, handleErrorCalled)
+	require.Len(t, upstream.calls, 2)
+	require.True(t, strings.HasPrefix(upstream.calls[0], base1))
+	require.True(t, strings.HasPrefix(upstream.calls[1], base2))
+
+	available := antigravity.DefaultURLAvailability.GetAvailableURLs()
+	require.NotEmpty(t, available)
+	require.Equal(t, base2, available[0])
+}
+
+func TestAntigravityHandleUpstreamError_UsesScopeLimitWhenEnabled(t *testing.T) {
+	t.Setenv(antigravityScopeRateLimitEnv, "true")
+	repo := &stubAntigravityAccountRepo{}
+	svc := &AntigravityGatewayService{accountRepo: repo}
+	account := &Account{ID: 9, Name: "acc-9", Platform: PlatformAntigravity}
+
+	body := buildGeminiRateLimitBody("3s")
+	svc.handleUpstreamError(context.Background(), "[test]", account, http.StatusTooManyRequests, http.Header{}, body, AntigravityQuotaScopeClaude)
+
+	require.Len(t, repo.scopeCalls, 1)
+	require.Empty(t, repo.rateCalls)
+	call := repo.scopeCalls[0]
+	require.Equal(t, account.ID, call.accountID)
+	require.Equal(t, AntigravityQuotaScopeClaude, call.scope)
+	require.WithinDuration(t, time.Now().Add(3*time.Second), call.resetAt, 2*time.Second)
+}
+
+func TestAntigravityHandleUpstreamError_UsesAccountLimitWhenScopeDisabled(t *testing.T) {
+	t.Setenv(antigravityScopeRateLimitEnv, "false")
+	repo := &stubAntigravityAccountRepo{}
+	svc := &AntigravityGatewayService{accountRepo: repo}
+	account := &Account{ID: 10, Name: "acc-10", Platform: PlatformAntigravity}
+
+	body := buildGeminiRateLimitBody("2s")
+	svc.handleUpstreamError(context.Background(), "[test]", account, http.StatusTooManyRequests, http.Header{}, body, AntigravityQuotaScopeClaude)
+
+	require.Len(t, repo.rateCalls, 1)
+	require.Empty(t, repo.scopeCalls)
+	call := repo.rateCalls[0]
+	require.Equal(t, account.ID, call.accountID)
+	require.WithinDuration(t, time.Now().Add(2*time.Second), call.resetAt, 2*time.Second)
+}
+
+func TestAccountIsSchedulableForModel_AntigravityRateLimits(t *testing.T) {
+	now := time.Now()
+	future := now.Add(10 * time.Minute)
+
+	account := &Account{
+		ID:          1,
+		Name:        "acc",
+		Platform:    PlatformAntigravity,
+		Status:      StatusActive,
+		Schedulable: true,
+	}
+
+	account.RateLimitResetAt = &future
+	require.False(t, account.IsSchedulableForModel("claude-sonnet-4-5"))
+	require.False(t, account.IsSchedulableForModel("gemini-3-flash"))
+
+	account.RateLimitResetAt = nil
+	account.Extra = map[string]any{
+		antigravityQuotaScopesKey: map[string]any{
+			"claude": map[string]any{
+				"rate_limit_reset_at": future.Format(time.RFC3339),
+			},
+		},
+	}
+
+	require.False(t, account.IsSchedulableForModel("claude-sonnet-4-5"))
+	require.True(t, account.IsSchedulableForModel("gemini-3-flash"))
+}
+
+func buildGeminiRateLimitBody(delay string) []byte {
+	return []byte(fmt.Sprintf(`{"error":{"message":"too many requests","details":[{"metadata":{"quotaResetDelay":%q}}]}}`, delay))
+}

backend/internal/service/antigravity_token_refresher.go

@@ -61,5 +61,10 @@ func (r *AntigravityTokenRefresher) Refresh(ctx context.Context, account *Accoun
 		}
 	}
 
+	// 如果 project_id 获取失败，返回 credentials 但同时返回错误让账户被标记
+	if tokenInfo.ProjectIDMissing {
+		return newCredentials, fmt.Errorf("missing_project_id: 账户缺少project id，可能无法使用Antigravity")
+	}
+
 	return newCredentials, nil
 }

backend/internal/service/api_key_auth_cache_impl.go

@@ -94,6 +94,20 @@ func (s *APIKeyService) initAuthCache(cfg *config.Config) {
 	s.authCacheL1 = cache
 }
 
+// StartAuthCacheInvalidationSubscriber starts the Pub/Sub subscriber for L1 cache invalidation.
+// This should be called after the service is fully initialized.
+func (s *APIKeyService) StartAuthCacheInvalidationSubscriber(ctx context.Context) {
+	if s.cache == nil || s.authCacheL1 == nil {
+		return
+	}
+	if err := s.cache.SubscribeAuthCacheInvalidation(ctx, func(cacheKey string) {
+		s.authCacheL1.Del(cacheKey)
+	}); err != nil {
+		// Log but don't fail - L1 cache will still work, just without cross-instance invalidation
+		println("[Service] Warning: failed to start auth cache invalidation subscriber:", err.Error())
+	}
+}
+
 func (s *APIKeyService) authCacheKey(key string) string {
 	sum := sha256.Sum256([]byte(key))
 	return hex.EncodeToString(sum[:])
@@ -149,6 +163,8 @@ func (s *APIKeyService) deleteAuthCache(ctx context.Context, cacheKey string) {
 		return
 	}
 	_ = s.cache.DeleteAuthCache(ctx, cacheKey)
+	// Publish invalidation message to other instances
+	_ = s.cache.PublishAuthCacheInvalidation(ctx, cacheKey)
 }
 
 func (s *APIKeyService) loadAuthCacheEntry(ctx context.Context, key, cacheKey string) (*APIKeyAuthCacheEntry, error) {

backend/internal/service/api_key_service.go

@@ -65,6 +65,10 @@ type APIKeyCache interface {
 	GetAuthCache(ctx context.Context, key string) (*APIKeyAuthCacheEntry, error)
 	SetAuthCache(ctx context.Context, key string, entry *APIKeyAuthCacheEntry, ttl time.Duration) error
 	DeleteAuthCache(ctx context.Context, key string) error
+
+	// Pub/Sub for L1 cache invalidation across instances
+	PublishAuthCacheInvalidation(ctx context.Context, cacheKey string) error
+	SubscribeAuthCacheInvalidation(ctx context.Context, handler func(cacheKey string)) error
 }
 
 // APIKeyAuthCacheInvalidator 提供认证缓存失效能力

backend/internal/service/api_key_service_cache_test.go

@@ -142,6 +142,14 @@ func (s *authCacheStub) DeleteAuthCache(ctx context.Context, key string) error {
 	return nil
 }
 
+func (s *authCacheStub) PublishAuthCacheInvalidation(ctx context.Context, cacheKey string) error {
+	return nil
+}
+
+func (s *authCacheStub) SubscribeAuthCacheInvalidation(ctx context.Context, handler func(cacheKey string)) error {
+	return nil
+}
+
 func TestAPIKeyService_GetByKey_UsesL2Cache(t *testing.T) {
 	cache := &authCacheStub{}
 	repo := &authRepoStub{

backend/internal/service/api_key_service_delete_test.go

@@ -168,6 +168,14 @@ func (s *apiKeyCacheStub) DeleteAuthCache(ctx context.Context, key string) error
 	return nil
 }
 
+func (s *apiKeyCacheStub) PublishAuthCacheInvalidation(ctx context.Context, cacheKey string) error {
+	return nil
+}
+
+func (s *apiKeyCacheStub) SubscribeAuthCacheInvalidation(ctx context.Context, handler func(cacheKey string)) error {
+	return nil
+}
+
 // TestApiKeyService_Delete_OwnerMismatch 测试非所有者尝试删除时返回权限错误。
 // 预期行为：
 //   - GetKeyAndOwnerID 返回所有者 ID 为 1

backend/internal/service/gateway_multiplatform_test.go

@@ -105,6 +105,9 @@ func (m *mockAccountRepoForPlatform) BatchUpdateLastUsed(ctx context.Context, up
 func (m *mockAccountRepoForPlatform) SetError(ctx context.Context, id int64, errorMsg string) error {
 	return nil
 }
+func (m *mockAccountRepoForPlatform) ClearError(ctx context.Context, id int64) error {
+	return nil
+}
 func (m *mockAccountRepoForPlatform) SetSchedulable(ctx context.Context, id int64, schedulable bool) error {
 	return nil
 }

backend/internal/service/gateway_service.go

@@ -11,6 +11,8 @@ import (
 	"fmt"
 	"io"
 	"log"
+	"log/slog"
+	mathrand "math/rand"
 	"net/http"
 	"os"
 	"regexp"
@@ -445,11 +447,20 @@ func (s *GatewayService) SelectAccountForModelWithExclusions(ctx context.Context
 }
 
 // SelectAccountWithLoadAwareness selects account with load-awareness and wait plan.
-// metadataUserID: 原始 metadata.user_id 字段（用于提取会话 UUID 进行会话数量限制）
+// metadataUserID: 已废弃参数，会话限制现在统一使用 sessionHash
 func (s *GatewayService) SelectAccountWithLoadAwareness(ctx context.Context, groupID *int64, sessionHash string, requestedModel string, excludedIDs map[int64]struct{}, metadataUserID string) (*AccountSelectionResult, error) {
+	// 调试日志：记录调度入口参数
+	excludedIDsList := make([]int64, 0, len(excludedIDs))
+	for id := range excludedIDs {
+		excludedIDsList = append(excludedIDsList, id)
+	}
+	slog.Debug("account_scheduling_starting",
+		"group_id", derefGroupID(groupID),
+		"model", requestedModel,
+		"session", shortSessionHash(sessionHash),
+		"excluded_ids", excludedIDsList)
+
 	cfg := s.schedulingConfig()
-	// 提取会话 UUID（用于会话数量限制）
-	sessionUUID := extractSessionUUID(metadataUserID)
 
 	var stickyAccountID int64
 	if sessionHash != "" && s.cache != nil {
@@ -475,41 +486,63 @@ func (s *GatewayService) SelectAccountWithLoadAwareness(ctx context.Context, gro
 	}
 
 	if s.concurrencyService == nil || !cfg.LoadBatchEnabled {
-		account, err := s.SelectAccountForModelWithExclusions(ctx, groupID, sessionHash, requestedModel, excludedIDs)
-		if err != nil {
-			return nil, err
+		// 复制排除列表，用于会话限制拒绝时的重试
+		localExcluded := make(map[int64]struct{})
+		for k, v := range excludedIDs {
+			localExcluded[k] = v
 		}
-		result, err := s.tryAcquireAccountSlot(ctx, account.ID, account.Concurrency)
-		if err == nil && result.Acquired {
-			return &AccountSelectionResult{
-				Account:     account,
-				Acquired:    true,
-				ReleaseFunc: result.ReleaseFunc,
-			}, nil
-		}
-		if stickyAccountID > 0 && stickyAccountID == account.ID && s.concurrencyService != nil {
-			waitingCount, _ := s.concurrencyService.GetAccountWaitingCount(ctx, account.ID)
-			if waitingCount < cfg.StickySessionMaxWaiting {
+
+		for {
+			account, err := s.SelectAccountForModelWithExclusions(ctx, groupID, sessionHash, requestedModel, localExcluded)
+			if err != nil {
+				return nil, err
+			}
+
+			result, err := s.tryAcquireAccountSlot(ctx, account.ID, account.Concurrency)
+			if err == nil && result.Acquired {
+				// 获取槽位后检查会话限制（使用 sessionHash 作为会话标识符）
+				if !s.checkAndRegisterSession(ctx, account, sessionHash) {
+					result.ReleaseFunc()                   // 释放槽位
+					localExcluded[account.ID] = struct{}{} // 排除此账号
+					continue                               // 重新选择
+				}
 				return &AccountSelectionResult{
-					Account: account,
-					WaitPlan: &AccountWaitPlan{
-						AccountID:      account.ID,
-						MaxConcurrency: account.Concurrency,
-						Timeout:        cfg.StickySessionWaitTimeout,
-						MaxWaiting:     cfg.StickySessionMaxWaiting,
-					},
+					Account:     account,
+					Acquired:    true,
+					ReleaseFunc: result.ReleaseFunc,
 				}, nil
 			}
+
+			// 对于等待计划的情况，也需要先检查会话限制
+			if !s.checkAndRegisterSession(ctx, account, sessionHash) {
+				localExcluded[account.ID] = struct{}{}
+				continue
+			}
+
+			if stickyAccountID > 0 && stickyAccountID == account.ID && s.concurrencyService != nil {
+				waitingCount, _ := s.concurrencyService.GetAccountWaitingCount(ctx, account.ID)
+				if waitingCount < cfg.StickySessionMaxWaiting {
+					return &AccountSelectionResult{
+						Account: account,
+						WaitPlan: &AccountWaitPlan{
+							AccountID:      account.ID,
+							MaxConcurrency: account.Concurrency,
+							Timeout:        cfg.StickySessionWaitTimeout,
+							MaxWaiting:     cfg.StickySessionMaxWaiting,
+						},
+					}, nil
+				}
+			}
+			return &AccountSelectionResult{
+				Account: account,
+				WaitPlan: &AccountWaitPlan{
+					AccountID:      account.ID,
+					MaxConcurrency: account.Concurrency,
+					Timeout:        cfg.FallbackWaitTimeout,
+					MaxWaiting:     cfg.FallbackMaxWaiting,
+				},
+			}, nil
 		}
-		return &AccountSelectionResult{
-			Account: account,
-			WaitPlan: &AccountWaitPlan{
-				AccountID:      account.ID,
-				MaxConcurrency: account.Concurrency,
-				Timeout:        cfg.FallbackWaitTimeout,
-				MaxWaiting:     cfg.FallbackMaxWaiting,
-			},
-		}, nil
 	}
 
 	platform, hasForcePlatform, err := s.resolvePlatform(ctx, groupID, group)
@@ -625,7 +658,7 @@ func (s *GatewayService) SelectAccountWithLoadAwareness(ctx context.Context, gro
 							result, err := s.tryAcquireAccountSlot(ctx, stickyAccountID, stickyAccount.Concurrency)
 							if err == nil && result.Acquired {
 								// 会话数量限制检查
-								if !s.checkAndRegisterSession(ctx, stickyAccount, sessionUUID) {
+								if !s.checkAndRegisterSession(ctx, stickyAccount, sessionHash) {
 									result.ReleaseFunc() // 释放槽位
 									// 继续到负载感知选择
 								} else {
@@ -643,15 +676,20 @@ func (s *GatewayService) SelectAccountWithLoadAwareness(ctx context.Context, gro
 
 							waitingCount, _ := s.concurrencyService.GetAccountWaitingCount(ctx, stickyAccountID)
 							if waitingCount < cfg.StickySessionMaxWaiting {
-								return &AccountSelectionResult{
-									Account: stickyAccount,
-									WaitPlan: &AccountWaitPlan{
-										AccountID:      stickyAccountID,
-										MaxConcurrency: stickyAccount.Concurrency,
-										Timeout:        cfg.StickySessionWaitTimeout,
-										MaxWaiting:     cfg.StickySessionMaxWaiting,
-									},
-								}, nil
+								// 会话数量限制检查（等待计划也需要占用会话配额）
+								if !s.checkAndRegisterSession(ctx, stickyAccount, sessionHash) {
+									// 会话限制已满，继续到负载感知选择
+								} else {
+									return &AccountSelectionResult{
+										Account: stickyAccount,
+										WaitPlan: &AccountWaitPlan{
+											AccountID:      stickyAccountID,
+											MaxConcurrency: stickyAccount.Concurrency,
+											Timeout:        cfg.StickySessionWaitTimeout,
+											MaxWaiting:     cfg.StickySessionMaxWaiting,
+										},
+									}, nil
+								}
 							}
 							// 粘性账号槽位满且等待队列已满，继续使用负载感知选择
 						}
@@ -714,7 +752,7 @@ func (s *GatewayService) SelectAccountWithLoadAwareness(ctx context.Context, gro
 					result, err := s.tryAcquireAccountSlot(ctx, item.account.ID, item.account.Concurrency)
 					if err == nil && result.Acquired {
 						// 会话数量限制检查
-						if !s.checkAndRegisterSession(ctx, item.account, sessionUUID) {
+						if !s.checkAndRegisterSession(ctx, item.account, sessionHash) {
 							result.ReleaseFunc() // 释放槽位，继续尝试下一个账号
 							continue
 						}
@@ -732,20 +770,26 @@ func (s *GatewayService) SelectAccountWithLoadAwareness(ctx context.Context, gro
 					}
 				}
 
-				// 5. 所有路由账号槽位满，返回等待计划（选择负载最低的）
-				acc := routingAvailable[0].account
-				if s.debugModelRoutingEnabled() {
-					log.Printf("[ModelRoutingDebug] routed wait: group_id=%v model=%s session=%s account=%d", derefGroupID(groupID), requestedModel, shortSessionHash(sessionHash), acc.ID)
+				// 5. 所有路由账号槽位满，尝试返回等待计划（选择负载最低的）
+				// 遍历找到第一个满足会话限制的账号
+				for _, item := range routingAvailable {
+					if !s.checkAndRegisterSession(ctx, item.account, sessionHash) {
+						continue // 会话限制已满，尝试下一个
+					}
+					if s.debugModelRoutingEnabled() {
+						log.Printf("[ModelRoutingDebug] routed wait: group_id=%v model=%s session=%s account=%d", derefGroupID(groupID), requestedModel, shortSessionHash(sessionHash), item.account.ID)
+					}
+					return &AccountSelectionResult{
+						Account: item.account,
+						WaitPlan: &AccountWaitPlan{
+							AccountID:      item.account.ID,
+							MaxConcurrency: item.account.Concurrency,
+							Timeout:        cfg.StickySessionWaitTimeout,
+							MaxWaiting:     cfg.StickySessionMaxWaiting,
+						},
+					}, nil
 				}
-				return &AccountSelectionResult{
-					Account: acc,
-					WaitPlan: &AccountWaitPlan{
-						AccountID:      acc.ID,
-						MaxConcurrency: acc.Concurrency,
-						Timeout:        cfg.StickySessionWaitTimeout,
-						MaxWaiting:     cfg.StickySessionMaxWaiting,
-					},
-				}, nil
+				// 所有路由账号会话限制都已满，继续到 Layer 2 回退
 			}
 			// 路由列表中的账号都不可用（负载率 >= 100），继续到 Layer 2 回退
 			log.Printf("[ModelRouting] All routed accounts unavailable for model=%s, falling back to normal selection", requestedModel)
@@ -773,7 +817,7 @@ func (s *GatewayService) SelectAccountWithLoadAwareness(ctx context.Context, gro
 					if err == nil && result.Acquired {
 						// 会话数量限制检查
 						// Session count limit check
-						if !s.checkAndRegisterSession(ctx, account, sessionUUID) {
+						if !s.checkAndRegisterSession(ctx, account, sessionHash) {
 							result.ReleaseFunc() // 释放槽位，继续到 Layer 2
 						} else {
 							_ = s.cache.RefreshSessionTTL(ctx, derefGroupID(groupID), sessionHash, stickySessionTTL)
@@ -787,15 +831,22 @@ func (s *GatewayService) SelectAccountWithLoadAwareness(ctx context.Context, gro
 
 					waitingCount, _ := s.concurrencyService.GetAccountWaitingCount(ctx, accountID)
 					if waitingCount < cfg.StickySessionMaxWaiting {
-						return &AccountSelectionResult{
-							Account: account,
-							WaitPlan: &AccountWaitPlan{
-								AccountID:      accountID,
-								MaxConcurrency: account.Concurrency,
-								Timeout:        cfg.StickySessionWaitTimeout,
-								MaxWaiting:     cfg.StickySessionMaxWaiting,
-							},
-						}, nil
+						// 会话数量限制检查（等待计划也需要占用会话配额）
+						// Session count limit check (wait plan also requires session quota)
+						if !s.checkAndRegisterSession(ctx, account, sessionHash) {
+							// 会话限制已满，继续到 Layer 2
+							// Session limit full, continue to Layer 2
+						} else {
+							return &AccountSelectionResult{
+								Account: account,
+								WaitPlan: &AccountWaitPlan{
+									AccountID:      accountID,
+									MaxConcurrency: account.Concurrency,
+									Timeout:        cfg.StickySessionWaitTimeout,
+									MaxWaiting:     cfg.StickySessionMaxWaiting,
+								},
+							}, nil
+						}
 					}
 				}
 			}
@@ -845,7 +896,7 @@ func (s *GatewayService) SelectAccountWithLoadAwareness(ctx context.Context, gro
 
 	loadMap, err := s.concurrencyService.GetAccountsLoadBatch(ctx, accountLoads)
 	if err != nil {
-		if result, ok := s.tryAcquireByLegacyOrder(ctx, candidates, groupID, sessionHash, preferOAuth, sessionUUID); ok {
+		if result, ok := s.tryAcquireByLegacyOrder(ctx, candidates, groupID, sessionHash, preferOAuth); ok {
 			return result, nil
 		}
 	} else {
@@ -895,7 +946,7 @@ func (s *GatewayService) SelectAccountWithLoadAwareness(ctx context.Context, gro
 				result, err := s.tryAcquireAccountSlot(ctx, item.account.ID, item.account.Concurrency)
 				if err == nil && result.Acquired {
 					// 会话数量限制检查
-					if !s.checkAndRegisterSession(ctx, item.account, sessionUUID) {
+					if !s.checkAndRegisterSession(ctx, item.account, sessionHash) {
 						result.ReleaseFunc() // 释放槽位，继续尝试下一个账号
 						continue
 					}
@@ -913,8 +964,12 @@ func (s *GatewayService) SelectAccountWithLoadAwareness(ctx context.Context, gro
 	}
 
 	// ============ Layer 3: 兜底排队 ============
-	sortAccountsByPriorityAndLastUsed(candidates, preferOAuth)
+	s.sortCandidatesForFallback(candidates, preferOAuth, cfg.FallbackSelectionMode)
 	for _, acc := range candidates {
+		// 会话数量限制检查（等待计划也需要占用会话配额）
+		if !s.checkAndRegisterSession(ctx, acc, sessionHash) {
+			continue // 会话限制已满，尝试下一个账号
+		}
 		return &AccountSelectionResult{
 			Account: acc,
 			WaitPlan: &AccountWaitPlan{
@@ -928,7 +983,7 @@ func (s *GatewayService) SelectAccountWithLoadAwareness(ctx context.Context, gro
 	return nil, errors.New("no available accounts")
 }
 
-func (s *GatewayService) tryAcquireByLegacyOrder(ctx context.Context, candidates []*Account, groupID *int64, sessionHash string, preferOAuth bool, sessionUUID string) (*AccountSelectionResult, bool) {
+func (s *GatewayService) tryAcquireByLegacyOrder(ctx context.Context, candidates []*Account, groupID *int64, sessionHash string, preferOAuth bool) (*AccountSelectionResult, bool) {
 	ordered := append([]*Account(nil), candidates...)
 	sortAccountsByPriorityAndLastUsed(ordered, preferOAuth)
 
@@ -936,7 +991,7 @@ func (s *GatewayService) tryAcquireByLegacyOrder(ctx context.Context, candidates
 		result, err := s.tryAcquireAccountSlot(ctx, acc.ID, acc.Concurrency)
 		if err == nil && result.Acquired {
 			// 会话数量限制检查
-			if !s.checkAndRegisterSession(ctx, acc, sessionUUID) {
+			if !s.checkAndRegisterSession(ctx, acc, sessionHash) {
 				result.ReleaseFunc() // 释放槽位，继续尝试下一个账号
 				continue
 			}
@@ -1093,7 +1148,24 @@ func (s *GatewayService) resolvePlatform(ctx context.Context, groupID *int64, gr
 
 func (s *GatewayService) listSchedulableAccounts(ctx context.Context, groupID *int64, platform string, hasForcePlatform bool) ([]Account, bool, error) {
 	if s.schedulerSnapshot != nil {
-		return s.schedulerSnapshot.ListSchedulableAccounts(ctx, groupID, platform, hasForcePlatform)
+		accounts, useMixed, err := s.schedulerSnapshot.ListSchedulableAccounts(ctx, groupID, platform, hasForcePlatform)
+		if err == nil {
+			slog.Debug("account_scheduling_list_snapshot",
+				"group_id", derefGroupID(groupID),
+				"platform", platform,
+				"use_mixed", useMixed,
+				"count", len(accounts))
+			for _, acc := range accounts {
+				slog.Debug("account_scheduling_account_detail",
+					"account_id", acc.ID,
+					"name", acc.Name,
+					"platform", acc.Platform,
+					"type", acc.Type,
+					"status", acc.Status,
+					"tls_fingerprint", acc.IsTLSFingerprintEnabled())
+			}
+		}
+		return accounts, useMixed, err
 	}
 	useMixed := (platform == PlatformAnthropic || platform == PlatformGemini) && !hasForcePlatform
 	if useMixed {
@@ -1106,6 +1178,10 @@ func (s *GatewayService) listSchedulableAccounts(ctx context.Context, groupID *i
 			accounts, err = s.accountRepo.ListSchedulableByPlatforms(ctx, platforms)
 		}
 		if err != nil {
+			slog.Debug("account_scheduling_list_failed",
+				"group_id", derefGroupID(groupID),
+				"platform", platform,
+				"error", err)
 			return nil, useMixed, err
 		}
 		filtered := make([]Account, 0, len(accounts))
@@ -1115,6 +1191,20 @@ func (s *GatewayService) listSchedulableAccounts(ctx context.Context, groupID *i
 			}
 			filtered = append(filtered, acc)
 		}
+		slog.Debug("account_scheduling_list_mixed",
+			"group_id", derefGroupID(groupID),
+			"platform", platform,
+			"raw_count", len(accounts),
+			"filtered_count", len(filtered))
+		for _, acc := range filtered {
+			slog.Debug("account_scheduling_account_detail",
+				"account_id", acc.ID,
+				"name", acc.Name,
+				"platform", acc.Platform,
+				"type", acc.Type,
+				"status", acc.Status,
+				"tls_fingerprint", acc.IsTLSFingerprintEnabled())
+		}
 		return filtered, useMixed, nil
 	}
 
@@ -1129,8 +1219,25 @@ func (s *GatewayService) listSchedulableAccounts(ctx context.Context, groupID *i
 		accounts, err = s.accountRepo.ListSchedulableByPlatform(ctx, platform)
 	}
 	if err != nil {
+		slog.Debug("account_scheduling_list_failed",
+			"group_id", derefGroupID(groupID),
+			"platform", platform,
+			"error", err)
 		return nil, useMixed, err
 	}
+	slog.Debug("account_scheduling_list_single",
+		"group_id", derefGroupID(groupID),
+		"platform", platform,
+		"count", len(accounts))
+	for _, acc := range accounts {
+		slog.Debug("account_scheduling_account_detail",
+			"account_id", acc.ID,
+			"name", acc.Name,
+			"platform", acc.Platform,
+			"type", acc.Type,
+			"status", acc.Status,
+			"tls_fingerprint", acc.IsTLSFingerprintEnabled())
+	}
 	return accounts, useMixed, nil
 }
 
@@ -1196,12 +1303,8 @@ func (s *GatewayService) isAccountSchedulableForWindowCost(ctx context.Context,
 
 	// 缓存未命中，从数据库查询
 	{
-		var startTime time.Time
-		if account.SessionWindowStart != nil {
-			startTime = *account.SessionWindowStart
-		} else {
-			startTime = time.Now().Add(-5 * time.Hour)
-		}
+		// 使用统一的窗口开始时间计算逻辑（考虑窗口过期情况）
+		startTime := account.GetCurrentWindowStartTime()
 
 		stats, err := s.usageLogRepo.GetAccountWindowStats(ctx, account.ID, startTime)
 		if err != nil {
@@ -1234,15 +1337,16 @@ checkSchedulability:
 
 // checkAndRegisterSession 检查并注册会话，用于会话数量限制
 // 仅适用于 Anthropic OAuth/SetupToken 账号
+// sessionID: 会话标识符（使用粘性会话的 hash）
 // 返回 true 表示允许（在限制内或会话已存在），false 表示拒绝（超出限制且是新会话）
-func (s *GatewayService) checkAndRegisterSession(ctx context.Context, account *Account, sessionUUID string) bool {
+func (s *GatewayService) checkAndRegisterSession(ctx context.Context, account *Account, sessionID string) bool {
 	// 只检查 Anthropic OAuth/SetupToken 账号
 	if !account.IsAnthropicOAuthOrSetupToken() {
 		return true
 	}
 
 	maxSessions := account.GetMaxSessions()
-	if maxSessions <= 0 || sessionUUID == "" {
+	if maxSessions <= 0 || sessionID == "" {
 		return true // 未启用会话限制或无会话ID
 	}
 
@@ -1252,7 +1356,7 @@ func (s *GatewayService) checkAndRegisterSession(ctx context.Context, account *A
 
 	idleTimeout := time.Duration(account.GetSessionIdleTimeoutMinutes()) * time.Minute
 
-	allowed, err := s.sessionLimitCache.RegisterSession(ctx, account.ID, sessionUUID, maxSessions, idleTimeout)
+	allowed, err := s.sessionLimitCache.RegisterSession(ctx, account.ID, sessionID, maxSessions, idleTimeout)
 	if err != nil {
 		// 失败开放：缓存错误时允许通过
 		return true
@@ -1260,18 +1364,6 @@ func (s *GatewayService) checkAndRegisterSession(ctx context.Context, account *A
 	return allowed
 }
 
-// extractSessionUUID 从 metadata.user_id 中提取会话 UUID
-// 格式: user_{64位hex}_account__session_{uuid}
-func extractSessionUUID(metadataUserID string) string {
-	if metadataUserID == "" {
-		return ""
-	}
-	if match := sessionIDRegex.FindStringSubmatch(metadataUserID); len(match) > 1 {
-		return match[1]
-	}
-	return ""
-}
-
 func (s *GatewayService) getSchedulableAccount(ctx context.Context, accountID int64) (*Account, error) {
 	if s.schedulerSnapshot != nil {
 		return s.schedulerSnapshot.GetAccount(ctx, accountID)
@@ -1301,6 +1393,56 @@ func sortAccountsByPriorityAndLastUsed(accounts []*Account, preferOAuth bool) {
 	})
 }
 
+// sortCandidatesForFallback 根据配置选择排序策略
+// mode: "last_used"(按最后使用时间) 或 "random"(随机)
+func (s *GatewayService) sortCandidatesForFallback(accounts []*Account, preferOAuth bool, mode string) {
+	if mode == "random" {
+		// 先按优先级排序，然后在同优先级内随机打乱
+		sortAccountsByPriorityOnly(accounts, preferOAuth)
+		shuffleWithinPriority(accounts)
+	} else {
+		// 默认按最后使用时间排序
+		sortAccountsByPriorityAndLastUsed(accounts, preferOAuth)
+	}
+}
+
+// sortAccountsByPriorityOnly 仅按优先级排序
+func sortAccountsByPriorityOnly(accounts []*Account, preferOAuth bool) {
+	sort.SliceStable(accounts, func(i, j int) bool {
+		a, b := accounts[i], accounts[j]
+		if a.Priority != b.Priority {
+			return a.Priority < b.Priority
+		}
+		if preferOAuth && a.Type != b.Type {
+			return a.Type == AccountTypeOAuth
+		}
+		return false
+	})
+}
+
+// shuffleWithinPriority 在同优先级内随机打乱顺序
+func shuffleWithinPriority(accounts []*Account) {
+	if len(accounts) <= 1 {
+		return
+	}
+	r := mathrand.New(mathrand.NewSource(time.Now().UnixNano()))
+	start := 0
+	for start < len(accounts) {
+		priority := accounts[start].Priority
+		end := start + 1
+		for end < len(accounts) && accounts[end].Priority == priority {
+			end++
+		}
+		// 对 [start, end) 范围内的账户随机打乱
+		if end-start > 1 {
+			r.Shuffle(end-start, func(i, j int) {
+				accounts[start+i], accounts[start+j] = accounts[start+j], accounts[start+i]
+			})
+		}
+		start = end
+	}
+}
+
 // selectAccountForModelWithPlatform 选择单平台账户（完全隔离）
 func (s *GatewayService) selectAccountForModelWithPlatform(ctx context.Context, groupID *int64, sessionHash string, requestedModel string, excludedIDs map[int64]struct{}, platform string) (*Account, error) {
 	preferOAuth := platform == PlatformGemini
@@ -2158,6 +2300,10 @@ func (s *GatewayService) Forward(ctx context.Context, c *gin.Context, account *A
 		proxyURL = account.Proxy.URL()
 	}
 
+	// 调试日志：记录即将转发的账号信息
+	log.Printf("[Forward] Using account: ID=%d Name=%s Platform=%s Type=%s TLSFingerprint=%v Proxy=%s",
+		account.ID, account.Name, account.Platform, account.Type, account.IsTLSFingerprintEnabled(), proxyURL)
+
 	// 重试循环
 	var resp *http.Response
 	retryStart := time.Now()
@@ -2172,7 +2318,7 @@ func (s *GatewayService) Forward(ctx context.Context, c *gin.Context, account *A
 		}
 
 		// 发送请求
-		resp, err = s.httpUpstream.Do(upstreamReq, proxyURL, account.ID, account.Concurrency)
+		resp, err = s.httpUpstream.DoWithTLS(upstreamReq, proxyURL, account.ID, account.Concurrency, account.IsTLSFingerprintEnabled())
 		if err != nil {
 			if resp != nil && resp.Body != nil {
 				_ = resp.Body.Close()
@@ -2246,7 +2392,7 @@ func (s *GatewayService) Forward(ctx context.Context, c *gin.Context, account *A
 					filteredBody := FilterThinkingBlocksForRetry(body)
 					retryReq, buildErr := s.buildUpstreamRequest(ctx, c, account, filteredBody, token, tokenType, reqModel)
 					if buildErr == nil {
-						retryResp, retryErr := s.httpUpstream.Do(retryReq, proxyURL, account.ID, account.Concurrency)
+						retryResp, retryErr := s.httpUpstream.DoWithTLS(retryReq, proxyURL, account.ID, account.Concurrency, account.IsTLSFingerprintEnabled())
 						if retryErr == nil {
 							if retryResp.StatusCode < 400 {
 								log.Printf("Account %d: signature error retry succeeded (thinking downgraded)", account.ID)
@@ -2278,7 +2424,7 @@ func (s *GatewayService) Forward(ctx context.Context, c *gin.Context, account *A
 									filteredBody2 := FilterSignatureSensitiveBlocksForRetry(body)
 									retryReq2, buildErr2 := s.buildUpstreamRequest(ctx, c, account, filteredBody2, token, tokenType, reqModel)
 									if buildErr2 == nil {
-										retryResp2, retryErr2 := s.httpUpstream.Do(retryReq2, proxyURL, account.ID, account.Concurrency)
+										retryResp2, retryErr2 := s.httpUpstream.DoWithTLS(retryReq2, proxyURL, account.ID, account.Concurrency, account.IsTLSFingerprintEnabled())
 										if retryErr2 == nil {
 											resp = retryResp2
 											break
@@ -2393,6 +2539,10 @@ func (s *GatewayService) Forward(ctx context.Context, c *gin.Context, account *A
 			_ = resp.Body.Close()
 			resp.Body = io.NopCloser(bytes.NewReader(respBody))
 
+			// 调试日志：打印重试耗尽后的错误响应
+			log.Printf("[Forward] Upstream error (retry exhausted, failover): Account=%d(%s) Status=%d RequestID=%s Body=%s",
+				account.ID, account.Name, resp.StatusCode, resp.Header.Get("x-request-id"), truncateString(string(respBody), 1000))
+
 			s.handleRetryExhaustedSideEffects(ctx, resp, account)
 			appendOpsUpstreamError(c, OpsUpstreamErrorEvent{
 				Platform:           account.Platform,
@@ -2420,6 +2570,10 @@ func (s *GatewayService) Forward(ctx context.Context, c *gin.Context, account *A
 		_ = resp.Body.Close()
 		resp.Body = io.NopCloser(bytes.NewReader(respBody))
 
+		// 调试日志：打印上游错误响应
+		log.Printf("[Forward] Upstream error (failover): Account=%d(%s) Status=%d RequestID=%s Body=%s",
+			account.ID, account.Name, resp.StatusCode, resp.Header.Get("x-request-id"), truncateString(string(respBody), 1000))
+
 		s.handleFailoverSideEffects(ctx, resp, account)
 		appendOpsUpstreamError(c, OpsUpstreamErrorEvent{
 			Platform:           account.Platform,
@@ -2549,9 +2703,10 @@ func (s *GatewayService) buildUpstreamRequest(ctx context.Context, c *gin.Contex
 			fingerprint = fp
 
 			// 2. 重写metadata.user_id（需要指纹中的ClientID和账号的account_uuid）
+			// 如果启用了会话ID伪装，会在重写后替换 session 部分为固定值
 			accountUUID := account.GetExtraString("account_uuid")
 			if accountUUID != "" && fp.ClientID != "" {
-				if newBody, err := s.identityService.RewriteUserID(body, account.ID, accountUUID, fp.ClientID); err == nil && len(newBody) > 0 {
+				if newBody, err := s.identityService.RewriteUserIDWithMasking(ctx, body, account, accountUUID, fp.ClientID); err == nil && len(newBody) > 0 {
 					body = newBody
 				}
 			}
@@ -2770,6 +2925,10 @@ func extractUpstreamErrorMessage(body []byte) string {
 func (s *GatewayService) handleErrorResponse(ctx context.Context, resp *http.Response, c *gin.Context, account *Account) (*ForwardResult, error) {
 	body, _ := io.ReadAll(io.LimitReader(resp.Body, 2<<20))
 
+	// 调试日志：打印上游错误响应
+	log.Printf("[Forward] Upstream error (non-retryable): Account=%d(%s) Status=%d RequestID=%s Body=%s",
+		account.ID, account.Name, resp.StatusCode, resp.Header.Get("x-request-id"), truncateString(string(body), 1000))
+
 	upstreamMsg := strings.TrimSpace(extractUpstreamErrorMessage(body))
 	upstreamMsg = sanitizeUpstreamErrorMessage(upstreamMsg)
 
@@ -3478,7 +3637,7 @@ func (s *GatewayService) ForwardCountTokens(ctx context.Context, c *gin.Context,
 	}
 
 	// 发送请求
-	resp, err := s.httpUpstream.Do(upstreamReq, proxyURL, account.ID, account.Concurrency)
+	resp, err := s.httpUpstream.DoWithTLS(upstreamReq, proxyURL, account.ID, account.Concurrency, account.IsTLSFingerprintEnabled())
 	if err != nil {
 		setOpsUpstreamError(c, 0, sanitizeUpstreamErrorMessage(err.Error()), "")
 		s.countTokensError(c, http.StatusBadGateway, "upstream_error", "Request failed")
@@ -3500,7 +3659,7 @@ func (s *GatewayService) ForwardCountTokens(ctx context.Context, c *gin.Context,
 		filteredBody := FilterThinkingBlocksForRetry(body)
 		retryReq, buildErr := s.buildCountTokensRequest(ctx, c, account, filteredBody, token, tokenType, reqModel)
 		if buildErr == nil {
-			retryResp, retryErr := s.httpUpstream.Do(retryReq, proxyURL, account.ID, account.Concurrency)
+			retryResp, retryErr := s.httpUpstream.DoWithTLS(retryReq, proxyURL, account.ID, account.Concurrency, account.IsTLSFingerprintEnabled())
 			if retryErr == nil {
 				resp = retryResp
 				respBody, err = io.ReadAll(resp.Body)
@@ -3578,12 +3737,13 @@ func (s *GatewayService) buildCountTokensRequest(ctx context.Context, c *gin.Con
 	}
 
 	// OAuth 账号：应用统一指纹和重写 userID
+	// 如果启用了会话ID伪装，会在重写后替换 session 部分为固定值
 	if account.IsOAuth() && s.identityService != nil {
 		fp, err := s.identityService.GetOrCreateFingerprint(ctx, account.ID, c.Request.Header)
 		if err == nil {
 			accountUUID := account.GetExtraString("account_uuid")
 			if accountUUID != "" && fp.ClientID != "" {
-				if newBody, err := s.identityService.RewriteUserID(body, account.ID, accountUUID, fp.ClientID); err == nil && len(newBody) > 0 {
+				if newBody, err := s.identityService.RewriteUserIDWithMasking(ctx, body, account, accountUUID, fp.ClientID); err == nil && len(newBody) > 0 {
 					body = newBody
 				}
 			}

backend/internal/service/gemini_multiplatform_test.go

@@ -90,6 +90,9 @@ func (m *mockAccountRepoForGemini) BatchUpdateLastUsed(ctx context.Context, upda
 func (m *mockAccountRepoForGemini) SetError(ctx context.Context, id int64, errorMsg string) error {
 	return nil
 }
+func (m *mockAccountRepoForGemini) ClearError(ctx context.Context, id int64) error {
+	return nil
+}
 func (m *mockAccountRepoForGemini) SetSchedulable(ctx context.Context, id int64, schedulable bool) error {
 	return nil
 }

backend/internal/service/http_upstream_port.go

@@ -10,6 +10,7 @@ import "net/http"
 // - 支持可选代理配置
 // - 支持账户级连接池隔离
 // - 实现类负责连接池管理和复用
+// - 支持可选的 TLS 指纹伪装
 type HTTPUpstream interface {
 	// Do 执行 HTTP 请求
 	//
@@ -27,4 +28,28 @@ type HTTPUpstream interface {
 	//   - 调用方必须关闭 resp.Body，否则会导致连接泄漏
 	//   - 响应体可能已被包装以跟踪请求生命周期
 	Do(req *http.Request, proxyURL string, accountID int64, accountConcurrency int) (*http.Response, error)
+
+	// DoWithTLS 执行带 TLS 指纹伪装的 HTTP 请求
+	//
+	// 参数:
+	//   - req: HTTP 请求对象，由调用方构建
+	//   - proxyURL: 代理服务器地址，空字符串表示直连
+	//   - accountID: 账户 ID，用于连接池隔离和 TLS 指纹模板选择
+	//   - accountConcurrency: 账户并发限制，用于动态调整连接池大小
+	//   - enableTLSFingerprint: 是否启用 TLS 指纹伪装
+	//
+	// 返回:
+	//   - *http.Response: HTTP 响应，调用方必须关闭 Body
+	//   - error: 请求错误（网络错误、超时等）
+	//
+	// TLS 指纹说明:
+	//   - 当 enableTLSFingerprint=true 时，使用 utls 库模拟 Claude CLI 的 TLS 指纹
+	//   - TLS 指纹模板根据 accountID % len(profiles) 自动选择
+	//   - 支持直连、HTTP/HTTPS 代理、SOCKS5 代理三种场景
+	//   - 如果 enableTLSFingerprint=false，行为与 Do 方法相同
+	//
+	// 注意:
+	//   - 调用方必须关闭 resp.Body，否则会导致连接泄漏
+	//   - TLS 指纹客户端与普通客户端使用不同的缓存键，互不影响
+	DoWithTLS(req *http.Request, proxyURL string, accountID int64, accountConcurrency int, enableTLSFingerprint bool) (*http.Response, error)
 }

backend/internal/service/identity_service.go

@@ -8,9 +8,11 @@ import (
 	"encoding/json"
 	"fmt"
 	"log"
+	"log/slog"
 	"net/http"
 	"regexp"
 	"strconv"
+	"strings"
 	"time"
 )
 
@@ -49,6 +51,13 @@ type Fingerprint struct {
 type IdentityCache interface {
 	GetFingerprint(ctx context.Context, accountID int64) (*Fingerprint, error)
 	SetFingerprint(ctx context.Context, accountID int64, fp *Fingerprint) error
+	// GetMaskedSessionID 获取固定的会话ID（用于会话ID伪装功能）
+	// 返回的 sessionID 是一个 UUID 格式的字符串
+	// 如果不存在或已过期（15分钟无请求），返回空字符串
+	GetMaskedSessionID(ctx context.Context, accountID int64) (string, error)
+	// SetMaskedSessionID 设置固定的会话ID，TTL 为 15 分钟
+	// 每次调用都会刷新 TTL
+	SetMaskedSessionID(ctx context.Context, accountID int64, sessionID string) error
 }
 
 // IdentityService 管理OAuth账号的请求身份指纹
@@ -203,6 +212,94 @@ func (s *IdentityService) RewriteUserID(body []byte, accountID int64, accountUUI
 	return json.Marshal(reqMap)
 }
 
+// RewriteUserIDWithMasking 重写body中的metadata.user_id，支持会话ID伪装
+// 如果账号启用了会话ID伪装（session_id_masking_enabled），
+// 则在完成常规重写后，将 session 部分替换为固定的伪装ID（15分钟内保持不变）
+func (s *IdentityService) RewriteUserIDWithMasking(ctx context.Context, body []byte, account *Account, accountUUID, cachedClientID string) ([]byte, error) {
+	// 先执行常规的 RewriteUserID 逻辑
+	newBody, err := s.RewriteUserID(body, account.ID, accountUUID, cachedClientID)
+	if err != nil {
+		return newBody, err
+	}
+
+	// 检查是否启用会话ID伪装
+	if !account.IsSessionIDMaskingEnabled() {
+		return newBody, nil
+	}
+
+	// 解析重写后的 body，提取 user_id
+	var reqMap map[string]any
+	if err := json.Unmarshal(newBody, &reqMap); err != nil {
+		return newBody, nil
+	}
+
+	metadata, ok := reqMap["metadata"].(map[string]any)
+	if !ok {
+		return newBody, nil
+	}
+
+	userID, ok := metadata["user_id"].(string)
+	if !ok || userID == "" {
+		return newBody, nil
+	}
+
+	// 查找 _session_ 的位置，替换其后的内容
+	const sessionMarker = "_session_"
+	idx := strings.LastIndex(userID, sessionMarker)
+	if idx == -1 {
+		return newBody, nil
+	}
+
+	// 获取或生成固定的伪装 session ID
+	maskedSessionID, err := s.cache.GetMaskedSessionID(ctx, account.ID)
+	if err != nil {
+		log.Printf("Warning: failed to get masked session ID for account %d: %v", account.ID, err)
+		return newBody, nil
+	}
+
+	if maskedSessionID == "" {
+		// 首次或已过期，生成新的伪装 session ID
+		maskedSessionID = generateRandomUUID()
+		log.Printf("Generated new masked session ID for account %d: %s", account.ID, maskedSessionID)
+	}
+
+	// 刷新 TTL（每次请求都刷新，保持 15 分钟有效期）
+	if err := s.cache.SetMaskedSessionID(ctx, account.ID, maskedSessionID); err != nil {
+		log.Printf("Warning: failed to set masked session ID for account %d: %v", account.ID, err)
+	}
+
+	// 替换 session 部分：保留 _session_ 之前的内容，替换之后的内容
+	newUserID := userID[:idx+len(sessionMarker)] + maskedSessionID
+
+	slog.Debug("session_id_masking_applied",
+		"account_id", account.ID,
+		"before", userID,
+		"after", newUserID,
+	)
+
+	metadata["user_id"] = newUserID
+	reqMap["metadata"] = metadata
+
+	return json.Marshal(reqMap)
+}
+
+// generateRandomUUID 生成随机 UUID v4 格式字符串
+func generateRandomUUID() string {
+	b := make([]byte, 16)
+	if _, err := rand.Read(b); err != nil {
+		// fallback: 使用时间戳生成
+		h := sha256.Sum256([]byte(fmt.Sprintf("%d", time.Now().UnixNano())))
+		b = h[:16]
+	}
+
+	// 设置 UUID v4 版本和变体位
+	b[6] = (b[6] & 0x0f) | 0x40
+	b[8] = (b[8] & 0x3f) | 0x80
+
+	return fmt.Sprintf("%x-%x-%x-%x-%x",
+		b[0:4], b[4:6], b[6:8], b[8:10], b[10:16])
+}
+
 // generateClientID 生成64位十六进制客户端ID（32字节随机数）
 func generateClientID() string {
 	b := make([]byte, 32)

backend/internal/service/ratelimit_service.go

@@ -73,10 +73,14 @@ func (s *RateLimitService) HandleUpstreamError(ctx context.Context, account *Acc
 		return false
 	}
 
-	tempMatched := false
+	// 先尝试临时不可调度规则（401除外）
+	// 如果匹配成功，直接返回，不执行后续禁用逻辑
 	if statusCode != 401 {
-		tempMatched = s.tryTempUnschedulable(ctx, account, statusCode, responseBody)
+		if s.tryTempUnschedulable(ctx, account, statusCode, responseBody) {
+			return true
+		}
 	}
+
 	upstreamMsg := strings.TrimSpace(extractUpstreamErrorMessage(responseBody))
 	upstreamMsg = sanitizeUpstreamErrorMessage(upstreamMsg)
 	if upstreamMsg != "" {
@@ -84,6 +88,14 @@ func (s *RateLimitService) HandleUpstreamError(ctx context.Context, account *Acc
 	}
 
 	switch statusCode {
+	case 400:
+		// 只有当错误信息包含 "organization has been disabled" 时才禁用
+		if strings.Contains(strings.ToLower(upstreamMsg), "organization has been disabled") {
+			msg := "Organization disabled (400): " + upstreamMsg
+			s.handleAuthError(ctx, account, msg)
+			shouldDisable = true
+		}
+		// 其他 400 错误（如参数问题）不处理，不禁用账号
 	case 401:
 		// 对所有 OAuth 账号在 401 错误时调用缓存失效并强制下次刷新
 		if account.Type == AccountTypeOAuth {
@@ -148,9 +160,6 @@ func (s *RateLimitService) HandleUpstreamError(ctx context.Context, account *Acc
 		}
 	}
 
-	if tempMatched {
-		return true
-	}
 	return shouldDisable
 }
 

backend/internal/service/session_limit_cache.go

@@ -38,8 +38,9 @@ type SessionLimitCache interface {
 	GetActiveSessionCount(ctx context.Context, accountID int64) (int, error)
 
 	// GetActiveSessionCountBatch 批量获取多个账号的活跃会话数
+	// idleTimeouts: 每个账号的空闲超时时间配置，key 为 accountID；若为 nil 或某账号不在其中，则使用默认超时
 	// 返回 map[accountID]count，查询失败的账号不在 map 中
-	GetActiveSessionCountBatch(ctx context.Context, accountIDs []int64) (map[int64]int, error)
+	GetActiveSessionCountBatch(ctx context.Context, accountIDs []int64, idleTimeouts map[int64]time.Duration) (map[int64]int, error)
 
 	// IsSessionActive 检查特定会话是否活跃（未过期）
 	IsSessionActive(ctx context.Context, accountID int64, sessionUUID string) (bool, error)

backend/internal/service/token_refresh_service.go

@@ -166,11 +166,25 @@ func (s *TokenRefreshService) refreshWithRetry(ctx context.Context, account *Acc
 
 	for attempt := 1; attempt <= s.cfg.MaxRetries; attempt++ {
 		newCredentials, err := refresher.Refresh(ctx, account)
-		if err == nil {
-			// 刷新成功，更新账号credentials
+
+		// 如果有新凭证，先更新（即使有错误也要保存 token）
+		if newCredentials != nil {
 			account.Credentials = newCredentials
-			if err := s.accountRepo.Update(ctx, account); err != nil {
-				return fmt.Errorf("failed to save credentials: %w", err)
+			if saveErr := s.accountRepo.Update(ctx, account); saveErr != nil {
+				return fmt.Errorf("failed to save credentials: %w", saveErr)
+			}
+		}
+
+		if err == nil {
+			// Antigravity 账户：如果之前是因为缺少 project_id 而标记为 error，现在成功获取到了，清除错误状态
+			if account.Platform == PlatformAntigravity &&
+				account.Status == StatusError &&
+				strings.Contains(account.ErrorMessage, "missing_project_id:") {
+				if clearErr := s.accountRepo.ClearError(ctx, account.ID); clearErr != nil {
+					log.Printf("[TokenRefresh] Failed to clear error status for account %d: %v", account.ID, clearErr)
+				} else {
+					log.Printf("[TokenRefresh] Account %d: cleared missing_project_id error", account.ID)
+				}
 			}
 			// 对所有 OAuth 账号调用缓存失效（InvalidateToken 内部根据平台判断是否需要处理）
 			if s.cacheInvalidator != nil && account.Type == AccountTypeOAuth {
@@ -230,6 +244,7 @@ func isNonRetryableRefreshError(err error) bool {
 		"invalid_client",      // 客户端配置错误
 		"unauthorized_client", // 客户端未授权
 		"access_denied",       // 访问被拒绝
+		"missing_project_id",  // 缺少 project_id
 	}
 	for _, needle := range nonRetryable {
 		if strings.Contains(msg, needle) {

backend/internal/service/wire.go

@@ -1,6 +1,7 @@
 package service
 
 import (
+	"context"
 	"database/sql"
 	"time"
 
@@ -196,6 +197,8 @@ func ProvideOpsScheduledReportService(
 
 // ProvideAPIKeyAuthCacheInvalidator 提供 API Key 认证缓存失效能力
 func ProvideAPIKeyAuthCacheInvalidator(apiKeyService *APIKeyService) APIKeyAuthCacheInvalidator {
+	// Start Pub/Sub subscriber for L1 cache invalidation across instances
+	apiKeyService.StartAuthCacheInvalidationSubscriber(context.Background())
 	return apiKeyService
 }