Merge pull request #798 from touwaeriol/feature/account-load-factor

feat: add account load_factor for scheduling load calculation

backend/internal/service/account.go

@@ -28,6 +28,7 @@ type Account struct {
 	// RateMultiplier 账号计费倍率（>=0，允许 0 表示该账号计费为 0）。
 	// 使用指针用于兼容旧版本调度缓存（Redis）中缺字段的情况：nil 表示按 1.0 处理。
 	RateMultiplier     *float64
+	LoadFactor         *int // 调度负载因子；nil 表示使用 Concurrency
 	Status             string
 	ErrorMessage       string
 	LastUsedAt         *time.Time
@@ -88,6 +89,19 @@ func (a *Account) BillingRateMultiplier() float64 {
 	return *a.RateMultiplier
 }
 
+func (a *Account) EffectiveLoadFactor() int {
+	if a == nil {
+		return 1
+	}
+	if a.LoadFactor != nil && *a.LoadFactor > 0 {
+		return *a.LoadFactor
+	}
+	if a.Concurrency > 0 {
+		return a.Concurrency
+	}
+	return 1
+}
+
 func (a *Account) IsSchedulable() bool {
 	if !a.IsActive() || !a.Schedulable {
 		return false

backend/internal/service/account_load_factor_test.go

@@ -0,0 +1,46 @@
+//go:build unit
+
+package service
+
+import (
+	"testing"
+
+	"github.com/stretchr/testify/require"
+)
+
+func intPtrHelper(v int) *int { return &v }
+
+func TestEffectiveLoadFactor_NilAccount(t *testing.T) {
+	var a *Account
+	require.Equal(t, 1, a.EffectiveLoadFactor())
+}
+
+func TestEffectiveLoadFactor_NilLoadFactor_PositiveConcurrency(t *testing.T) {
+	a := &Account{Concurrency: 5}
+	require.Equal(t, 5, a.EffectiveLoadFactor())
+}
+
+func TestEffectiveLoadFactor_NilLoadFactor_ZeroConcurrency(t *testing.T) {
+	a := &Account{Concurrency: 0}
+	require.Equal(t, 1, a.EffectiveLoadFactor())
+}
+
+func TestEffectiveLoadFactor_PositiveLoadFactor(t *testing.T) {
+	a := &Account{Concurrency: 5, LoadFactor: intPtrHelper(20)}
+	require.Equal(t, 20, a.EffectiveLoadFactor())
+}
+
+func TestEffectiveLoadFactor_ZeroLoadFactor_FallbackToConcurrency(t *testing.T) {
+	a := &Account{Concurrency: 5, LoadFactor: intPtrHelper(0)}
+	require.Equal(t, 5, a.EffectiveLoadFactor())
+}
+
+func TestEffectiveLoadFactor_NegativeLoadFactor_FallbackToConcurrency(t *testing.T) {
+	a := &Account{Concurrency: 3, LoadFactor: intPtrHelper(-1)}
+	require.Equal(t, 3, a.EffectiveLoadFactor())
+}
+
+func TestEffectiveLoadFactor_ZeroLoadFactor_ZeroConcurrency(t *testing.T) {
+	a := &Account{Concurrency: 0, LoadFactor: intPtrHelper(0)}
+	require.Equal(t, 1, a.EffectiveLoadFactor())
+}

backend/internal/service/account_service.go

@@ -82,6 +82,7 @@ type AccountBulkUpdate struct {
 	Concurrency    *int
 	Priority       *int
 	RateMultiplier *float64
+	LoadFactor     *int
 	Status         *string
 	Schedulable    *bool
 	Credentials    map[string]any

backend/internal/service/admin_service.go

@@ -196,6 +196,7 @@ type CreateAccountInput struct {
 	Concurrency        int
 	Priority           int
 	RateMultiplier     *float64 // 账号计费倍率（>=0，允许 0）
+	LoadFactor         *int
 	GroupIDs           []int64
 	ExpiresAt          *int64
 	AutoPauseOnExpired *bool
@@ -216,6 +217,7 @@ type UpdateAccountInput struct {
 	Concurrency           *int     // 使用指针区分"未提供"和"设置为0"
 	Priority              *int     // 使用指针区分"未提供"和"设置为0"
 	RateMultiplier        *float64 // 账号计费倍率（>=0，允许 0）
+	LoadFactor            *int
 	Status                string
 	GroupIDs              *[]int64
 	ExpiresAt             *int64
@@ -231,6 +233,7 @@ type BulkUpdateAccountsInput struct {
 	Concurrency    *int
 	Priority       *int
 	RateMultiplier *float64 // 账号计费倍率（>=0，允许 0）
+	LoadFactor     *int
 	Status         string
 	Schedulable    *bool
 	GroupIDs       *[]int64
@@ -1414,6 +1417,12 @@ func (s *adminServiceImpl) CreateAccount(ctx context.Context, input *CreateAccou
 		}
 		account.RateMultiplier = input.RateMultiplier
 	}
+	if input.LoadFactor != nil && *input.LoadFactor > 0 {
+		if *input.LoadFactor > 10000 {
+			return nil, errors.New("load_factor must be <= 10000")
+		}
+		account.LoadFactor = input.LoadFactor
+	}
 	if err := s.accountRepo.Create(ctx, account); err != nil {
 		return nil, err
 	}
@@ -1488,6 +1497,15 @@ func (s *adminServiceImpl) UpdateAccount(ctx context.Context, id int64, input *U
 		}
 		account.RateMultiplier = input.RateMultiplier
 	}
+	if input.LoadFactor != nil {
+		if *input.LoadFactor <= 0 {
+			account.LoadFactor = nil // 0 或负数表示清除
+		} else if *input.LoadFactor > 10000 {
+			return nil, errors.New("load_factor must be <= 10000")
+		} else {
+			account.LoadFactor = input.LoadFactor
+		}
+	}
 	if input.Status != "" {
 		account.Status = input.Status
 	}
@@ -1621,6 +1639,15 @@ func (s *adminServiceImpl) BulkUpdateAccounts(ctx context.Context, input *BulkUp
 	if input.RateMultiplier != nil {
 		repoUpdates.RateMultiplier = input.RateMultiplier
 	}
+	if input.LoadFactor != nil {
+		if *input.LoadFactor <= 0 {
+			repoUpdates.LoadFactor = nil // 0 或负数表示清除
+		} else if *input.LoadFactor > 10000 {
+			return nil, errors.New("load_factor must be <= 10000")
+		} else {
+			repoUpdates.LoadFactor = input.LoadFactor
+		}
+	}
 	if input.Status != "" {
 		repoUpdates.Status = &input.Status
 	}

backend/internal/service/gateway_service.go

@@ -1316,7 +1316,7 @@ func (s *GatewayService) SelectAccountWithLoadAwareness(ctx context.Context, gro
 			for _, acc := range routingCandidates {
 				routingLoads = append(routingLoads, AccountWithConcurrency{
 					ID:             acc.ID,
-					MaxConcurrency: acc.Concurrency,
+					MaxConcurrency: acc.EffectiveLoadFactor(),
 				})
 			}
 			routingLoadMap, _ := s.concurrencyService.GetAccountsLoadBatch(ctx, routingLoads)
@@ -1509,7 +1509,7 @@ func (s *GatewayService) SelectAccountWithLoadAwareness(ctx context.Context, gro
 	for _, acc := range candidates {
 		accountLoads = append(accountLoads, AccountWithConcurrency{
 			ID:             acc.ID,
-			MaxConcurrency: acc.Concurrency,
+			MaxConcurrency: acc.EffectiveLoadFactor(),
 		})
 	}
 

backend/internal/service/openai_account_scheduler.go

@@ -342,6 +342,7 @@ func (s *defaultOpenAIAccountScheduler) selectBySessionHash(
 	}
 
 	cfg := s.service.schedulingConfig()
+	// WaitPlan.MaxConcurrency 使用 Concurrency（非 EffectiveLoadFactor），因为 WaitPlan 控制的是 Redis 实际并发槽位等待。
 	if s.service.concurrencyService != nil {
 		return &AccountSelectionResult{
 			Account: account,
@@ -590,7 +591,7 @@ func (s *defaultOpenAIAccountScheduler) selectByLoadBalance(
 		filtered = append(filtered, account)
 		loadReq = append(loadReq, AccountWithConcurrency{
 			ID:             account.ID,
-			MaxConcurrency: account.Concurrency,
+			MaxConcurrency: account.EffectiveLoadFactor(),
 		})
 	}
 	if len(filtered) == 0 {
@@ -703,6 +704,7 @@ func (s *defaultOpenAIAccountScheduler) selectByLoadBalance(
 	}
 
 	cfg := s.service.schedulingConfig()
+	// WaitPlan.MaxConcurrency 使用 Concurrency（非 EffectiveLoadFactor），因为 WaitPlan 控制的是 Redis 实际并发槽位等待。
 	candidate := selectionOrder[0]
 	return &AccountSelectionResult{
 		Account: candidate.account,

backend/internal/service/openai_gateway_service.go

@@ -1252,7 +1252,7 @@ func (s *OpenAIGatewayService) SelectAccountWithLoadAwareness(ctx context.Contex
 	for _, acc := range candidates {
 		accountLoads = append(accountLoads, AccountWithConcurrency{
 			ID:             acc.ID,
-			MaxConcurrency: acc.Concurrency,
+			MaxConcurrency: acc.EffectiveLoadFactor(),
 		})
 	}
 

backend/internal/service/ops_concurrency.go

@@ -64,8 +64,12 @@ func (s *OpsService) getAccountsLoadMapBestEffort(ctx context.Context, accounts
 		if acc.ID <= 0 {
 			continue
 		}
-		if prev, ok := unique[acc.ID]; !ok || acc.Concurrency > prev {
-			unique[acc.ID] = acc.Concurrency
+		c := acc.Concurrency
+		if c <= 0 {
+			c = 1
+		}
+		if prev, ok := unique[acc.ID]; !ok || c > prev {
+			unique[acc.ID] = c
 		}
 	}
 

backend/internal/service/ops_metrics_collector.go

@@ -389,13 +389,9 @@ func (c *OpsMetricsCollector) collectConcurrencyQueueDepth(parentCtx context.Con
 		if acc.ID <= 0 {
 			continue
 		}
-		maxConc := acc.Concurrency
-		if maxConc < 0 {
-			maxConc = 0
-		}
 		batch = append(batch, AccountWithConcurrency{
 			ID:             acc.ID,
-			MaxConcurrency: maxConc,
+			MaxConcurrency: acc.Concurrency,
 		})
 	}
 	if len(batch) == 0 {