refactor: simplify sticky session rate limit handling — switch immediately on any rate limit

Remove threshold-based waiting in both sticky session and antigravity
pre-check paths. When a model is rate-limited, immediately clear the
sticky session and switch accounts instead of waiting for short durations.

backend/internal/service/antigravity_gateway_service.go

@@ -264,27 +264,15 @@ func (s *AntigravityGatewayService) handleSmartRetry(p antigravityRetryLoopParam
 
 // antigravityRetryLoop 执行带 URL fallback 的重试循环
 func (s *AntigravityGatewayService) antigravityRetryLoop(p antigravityRetryLoopParams) (*antigravityRetryLoopResult, error) {
-	// 预检查：如果账号已限流，根据剩余时间决定等待或切换
+	// 预检查：如果账号已限流，直接返回切换信号
 	if p.requestedModel != "" {
 		if remaining := p.account.GetRateLimitRemainingTimeWithContext(p.ctx, p.requestedModel); remaining > 0 {
-			if remaining < antigravityRateLimitThreshold {
+			log.Printf("%s pre_check: rate_limit_switch remaining=%v model=%s account=%d",
-				// 限流剩余时间较短，等待后继续
+				p.prefix, remaining.Truncate(time.Millisecond), p.requestedModel, p.account.ID)
-				log.Printf("%s pre_check: rate_limit_wait remaining=%v model=%s account=%d",
+			return nil, &AntigravityAccountSwitchError{
-					p.prefix, remaining.Truncate(time.Millisecond), p.requestedModel, p.account.ID)
+				OriginalAccountID: p.account.ID,
-				select {
+				RateLimitedModel:  p.requestedModel,
-				case <-p.ctx.Done():
+				IsStickySession:   p.isStickySession,
-					return nil, p.ctx.Err()
-				case <-time.After(remaining):
-				}
-			} else {
-				// 限流剩余时间较长，返回账号切换信号
-				log.Printf("%s pre_check: rate_limit_switch remaining=%v model=%s account=%d",
-					p.prefix, remaining.Truncate(time.Second), p.requestedModel, p.account.ID)
-				return nil, &AntigravityAccountSwitchError{
-					OriginalAccountID: p.account.ID,
-					RateLimitedModel:  p.requestedModel,
-					IsStickySession:   p.isStickySession,
-				}
 			}
 		}
 	}

backend/internal/service/antigravity_rate_limit_test.go

@@ -803,7 +803,7 @@ func TestSetModelRateLimitByModelName_NotConvertToScope(t *testing.T) {
 	require.NotEqual(t, "claude_sonnet", call.modelKey, "should NOT be scope")
 }
 
-func TestAntigravityRetryLoop_PreCheck_WaitsWhenRemainingBelowThreshold(t *testing.T) {
+func TestAntigravityRetryLoop_PreCheck_SwitchesWhenRateLimited(t *testing.T) {
 	upstream := &recordingOKUpstream{}
 	account := &Account{
 		ID:          1,
@@ -815,19 +815,15 @@ func TestAntigravityRetryLoop_PreCheck_WaitsWhenRemainingBelowThreshold(t *testi
 		Extra: map[string]any{
 			modelRateLimitsKey: map[string]any{
 				"claude-sonnet-4-5": map[string]any{
-					// RFC3339 here is second-precision; keep it safely in the future.
 					"rate_limit_reset_at": time.Now().Add(2 * time.Second).Format(time.RFC3339),
 				},
 			},
 		},
 	}
 
-	ctx, cancel := context.WithTimeout(context.Background(), 30*time.Millisecond)
-	defer cancel()
-
 	svc := &AntigravityGatewayService{}
 	result, err := svc.antigravityRetryLoop(antigravityRetryLoopParams{
-		ctx:             ctx,
+		ctx:             context.Background(),
 		prefix:          "[test]",
 		account:         account,
 		accessToken:     "token",
@@ -841,12 +837,16 @@ func TestAntigravityRetryLoop_PreCheck_WaitsWhenRemainingBelowThreshold(t *testi
 		},
 	})
 
-	require.ErrorIs(t, err, context.DeadlineExceeded)
 	require.Nil(t, result)
-	require.Equal(t, 0, upstream.calls, "should not call upstream while waiting on pre-check")
+	var switchErr *AntigravityAccountSwitchError
+	require.ErrorAs(t, err, &switchErr)
+	require.Equal(t, account.ID, switchErr.OriginalAccountID)
+	require.Equal(t, "claude-sonnet-4-5", switchErr.RateLimitedModel)
+	require.True(t, switchErr.IsStickySession)
+	require.Equal(t, 0, upstream.calls, "should not call upstream when switching on pre-check")
 }
 
-func TestAntigravityRetryLoop_PreCheck_SwitchesWhenRemainingAtOrAboveThreshold(t *testing.T) {
+func TestAntigravityRetryLoop_PreCheck_SwitchesWhenRemainingLong(t *testing.T) {
 	upstream := &recordingOKUpstream{}
 	account := &Account{
 		ID:          2,

backend/internal/service/gateway_service.go

@@ -323,21 +323,15 @@ func derefGroupID(groupID *int64) int64 {
 	return *groupID
 }
 
-// stickySessionRateLimitThreshold 定义清除粘性会话的限流时间阈值。
-// 当账号限流剩余时间超过此阈值时，清除粘性会话以便切换到其他账号。
-// 低于此阈值时保持粘性会话，等待短暂限流结束。
-const stickySessionRateLimitThreshold = 10 * time.Second
-
 // shouldClearStickySession 检查账号是否处于不可调度状态，需要清理粘性会话绑定。
 // 当账号状态为错误、禁用、不可调度、处于临时不可调度期间，
-// 或模型限流剩余时间超过 stickySessionRateLimitThreshold 时，返回 true。
+// 或请求的模型处于限流状态时，返回 true。
 // 这确保后续请求不会继续使用不可用的账号。
 //
 // shouldClearStickySession checks if an account is in an unschedulable state
 // and the sticky session binding should be cleared.
 // Returns true when account status is error/disabled, schedulable is false,
-// within temporary unschedulable period, or model rate limit remaining time
+// within temporary unschedulable period, or the requested model is rate-limited.
-// exceeds stickySessionRateLimitThreshold.
 // This ensures subsequent requests won't continue using unavailable accounts.
 func shouldClearStickySession(account *Account, requestedModel string) bool {
 	if account == nil {
@@ -349,8 +343,8 @@ func shouldClearStickySession(account *Account, requestedModel string) bool {
 	if account.TempUnschedulableUntil != nil && time.Now().Before(*account.TempUnschedulableUntil) {
 		return true
 	}
-	// 检查模型限流和 scope 限流，只在超过阈值时清除粘性会话
+	// 检查模型限流和 scope 限流，有限流即清除粘性会话
-	if remaining := account.GetRateLimitRemainingTimeWithContext(context.Background(), requestedModel); remaining > stickySessionRateLimitThreshold {
+	if remaining := account.GetRateLimitRemainingTimeWithContext(context.Background(), requestedModel); remaining > 0 {
 		return true
 	}
 	return false

backend/internal/service/sticky_session_test.go

@@ -23,8 +23,7 @@ import (
 //   - 临时不可调度且未过期：清理
 //   - 临时不可调度已过期：不清理
 //   - 正常可调度状态：不清理
-//   - 模型限流超过阈值：清理
+//   - 模型限流（任意时长）：清理
-//   - 模型限流未超过阈值：不清理
 //
 // TestShouldClearStickySession tests the sticky session clearing logic.
 // Verifies correct behavior for various account states including:
@@ -35,9 +34,9 @@ func TestShouldClearStickySession(t *testing.T) {
 	future := now.Add(1 * time.Hour)
 	past := now.Add(-1 * time.Hour)
 
-	// 短限流时间（低于阈值，不应清除粘性会话）
+	// 短限流时间（有限流即清除粘性会话）
 	shortRateLimitReset := now.Add(5 * time.Second).Format(time.RFC3339)
-	// 长限流时间（超过阈值，应清除粘性会话）
+	// 长限流时间（有限流即清除粘性会话）
 	longRateLimitReset := now.Add(30 * time.Second).Format(time.RFC3339)
 
 	tests := []struct {
@@ -53,7 +52,7 @@ func TestShouldClearStickySession(t *testing.T) {
 		{name: "temp unschedulable", account: &Account{Status: StatusActive, Schedulable: true, TempUnschedulableUntil: &future}, requestedModel: "", want: true},
 		{name: "temp unschedulable expired", account: &Account{Status: StatusActive, Schedulable: true, TempUnschedulableUntil: &past}, requestedModel: "", want: false},
 		{name: "active schedulable", account: &Account{Status: StatusActive, Schedulable: true}, requestedModel: "", want: false},
-		// 模型限流测试
+		// 模型限流测试：有限流即清除
 		{
 			name: "model rate limited short duration",
 			account: &Account{
@@ -68,7 +67,7 @@ func TestShouldClearStickySession(t *testing.T) {
 				},
 			},
 			requestedModel: "claude-sonnet-4",
-			want:           false, // 低于阈值，不清除
+			want:           true, // 有限流即清除
 		},
 		{
 			name: "model rate limited long duration",
@@ -84,7 +83,7 @@ func TestShouldClearStickySession(t *testing.T) {
 				},
 			},
 			requestedModel: "claude-sonnet-4",
-			want:           true, // 超过阈值，清除
+			want:           true, // 有限流即清除
 		},
 		{
 			name: "model rate limited different model",