feat: add independent load_factor field for scheduling load calculation

- Separate load factor from concurrency: concurrency controls actual
  slot acquisition, load_factor controls load rate calculation
- Add EffectiveLoadFactor() method: LoadFactor > Concurrency > 1
- Add load_factor field to Create/Edit/BulkEdit account forms
- Fix RPM default value: auto-fill 15 when RPM enabled but not set
- Fix stale test compilation errors in server and handler packages
This commit is contained in:
erio
2026-03-06 03:42:24 +08:00
parent c746964936
commit 440c3f46a7
34 changed files with 602 additions and 86 deletions

View File

@@ -102,6 +102,7 @@ type CreateAccountRequest struct {
Concurrency int `json:"concurrency"`
Priority int `json:"priority"`
RateMultiplier *float64 `json:"rate_multiplier"`
LoadFactor *int `json:"load_factor"`
GroupIDs []int64 `json:"group_ids"`
ExpiresAt *int64 `json:"expires_at"`
AutoPauseOnExpired *bool `json:"auto_pause_on_expired"`
@@ -120,6 +121,7 @@ type UpdateAccountRequest struct {
Concurrency *int `json:"concurrency"`
Priority *int `json:"priority"`
RateMultiplier *float64 `json:"rate_multiplier"`
LoadFactor *int `json:"load_factor"`
Status string `json:"status" binding:"omitempty,oneof=active inactive"`
GroupIDs *[]int64 `json:"group_ids"`
ExpiresAt *int64 `json:"expires_at"`
@@ -135,6 +137,7 @@ type BulkUpdateAccountsRequest struct {
Concurrency *int `json:"concurrency"`
Priority *int `json:"priority"`
RateMultiplier *float64 `json:"rate_multiplier"`
LoadFactor *int `json:"load_factor"`
Status string `json:"status" binding:"omitempty,oneof=active inactive error"`
Schedulable *bool `json:"schedulable"`
GroupIDs *[]int64 `json:"group_ids"`
@@ -506,6 +509,7 @@ func (h *AccountHandler) Create(c *gin.Context) {
Concurrency: req.Concurrency,
Priority: req.Priority,
RateMultiplier: req.RateMultiplier,
LoadFactor: req.LoadFactor,
GroupIDs: req.GroupIDs,
ExpiresAt: req.ExpiresAt,
AutoPauseOnExpired: req.AutoPauseOnExpired,
@@ -575,6 +579,7 @@ func (h *AccountHandler) Update(c *gin.Context) {
Concurrency: req.Concurrency, // 指针类型nil 表示未提供
Priority: req.Priority, // 指针类型nil 表示未提供
RateMultiplier: req.RateMultiplier,
LoadFactor: req.LoadFactor,
Status: req.Status,
GroupIDs: req.GroupIDs,
ExpiresAt: req.ExpiresAt,
@@ -1101,6 +1106,7 @@ func (h *AccountHandler) BulkUpdate(c *gin.Context) {
req.Concurrency != nil ||
req.Priority != nil ||
req.RateMultiplier != nil ||
req.LoadFactor != nil ||
req.Status != "" ||
req.Schedulable != nil ||
req.GroupIDs != nil ||
@@ -1119,6 +1125,7 @@ func (h *AccountHandler) BulkUpdate(c *gin.Context) {
Concurrency: req.Concurrency,
Priority: req.Priority,
RateMultiplier: req.RateMultiplier,
LoadFactor: req.LoadFactor,
Status: req.Status,
Schedulable: req.Schedulable,
GroupIDs: req.GroupIDs,

View File

@@ -184,6 +184,7 @@ func AccountFromServiceShallow(a *service.Account) *Account {
Extra: a.Extra,
ProxyID: a.ProxyID,
Concurrency: a.Concurrency,
LoadFactor: a.LoadFactor,
Priority: a.Priority,
RateMultiplier: a.BillingRateMultiplier(),
Status: a.Status,

View File

@@ -133,6 +133,7 @@ type Account struct {
Extra map[string]any `json:"extra"`
ProxyID *int64 `json:"proxy_id"`
Concurrency int `json:"concurrency"`
LoadFactor *int `json:"load_factor,omitempty"`
Priority int `json:"priority"`
RateMultiplier float64 `json:"rate_multiplier"`
Status string `json:"status"`

View File

@@ -337,9 +337,6 @@ func (s *stubUsageLogRepo) GetModelStatsWithFilters(ctx context.Context, startTi
func (s *stubUsageLogRepo) GetGroupStatsWithFilters(ctx context.Context, startTime, endTime time.Time, userID, apiKeyID, accountID, groupID int64, requestType *int16, stream *bool, billingType *int8) ([]usagestats.GroupStat, error) {
return nil, nil
}
func (s *stubUsageLogRepo) GetGroupStatsWithFilters(ctx context.Context, startTime, endTime time.Time, userID, apiKeyID, accountID, groupID int64, stream *bool, billingType *int8) ([]usagestats.GroupStat, error) {
return nil, nil
}
func (s *stubUsageLogRepo) GetAPIKeyUsageTrend(ctx context.Context, startTime, endTime time.Time, granularity string, limit int) ([]usagestats.APIKeyUsageTrendPoint, error) {
return nil, nil
}

View File

@@ -84,6 +84,9 @@ func (r *accountRepository) Create(ctx context.Context, account *service.Account
if account.RateMultiplier != nil {
builder.SetRateMultiplier(*account.RateMultiplier)
}
if account.LoadFactor != nil {
builder.SetLoadFactor(*account.LoadFactor)
}
if account.ProxyID != nil {
builder.SetProxyID(*account.ProxyID)
@@ -318,6 +321,11 @@ func (r *accountRepository) Update(ctx context.Context, account *service.Account
if account.RateMultiplier != nil {
builder.SetRateMultiplier(*account.RateMultiplier)
}
if account.LoadFactor != nil {
builder.SetLoadFactor(*account.LoadFactor)
} else {
builder.ClearLoadFactor()
}
if account.ProxyID != nil {
builder.SetProxyID(*account.ProxyID)
@@ -1205,6 +1213,15 @@ func (r *accountRepository) BulkUpdate(ctx context.Context, ids []int64, updates
args = append(args, *updates.RateMultiplier)
idx++
}
if updates.LoadFactor != nil {
if *updates.LoadFactor <= 0 {
setClauses = append(setClauses, "load_factor = NULL")
} else {
setClauses = append(setClauses, "load_factor = $"+itoa(idx))
args = append(args, *updates.LoadFactor)
idx++
}
}
if updates.Status != nil {
setClauses = append(setClauses, "status = $"+itoa(idx))
args = append(args, *updates.Status)
@@ -1527,6 +1544,7 @@ func accountEntityToService(m *dbent.Account) *service.Account {
Concurrency: m.Concurrency,
Priority: m.Priority,
RateMultiplier: &rateMultiplier,
LoadFactor: m.LoadFactor,
Status: m.Status,
ErrorMessage: derefString(m.ErrorMessage),
LastUsedAt: m.LastUsedAt,

View File

@@ -1096,6 +1096,14 @@ func (s *stubAccountRepo) UpdateExtra(ctx context.Context, id int64, updates map
return errors.New("not implemented")
}
func (s *stubAccountRepo) IncrementQuotaUsed(ctx context.Context, id int64, amount float64) error {
return errors.New("not implemented")
}
func (s *stubAccountRepo) ResetQuotaUsed(ctx context.Context, id int64) error {
return errors.New("not implemented")
}
func (s *stubAccountRepo) BulkUpdate(ctx context.Context, ids []int64, updates service.AccountBulkUpdate) (int64, error) {
s.bulkUpdateIDs = append([]int64{}, ids...)
return int64(len(ids)), nil
@@ -1617,10 +1625,6 @@ func (r *stubUsageLogRepo) GetGroupStatsWithFilters(ctx context.Context, startTi
return nil, errors.New("not implemented")
}
func (r *stubUsageLogRepo) GetGroupStatsWithFilters(ctx context.Context, startTime, endTime time.Time, userID, apiKeyID, accountID, groupID int64, stream *bool, billingType *int8) ([]usagestats.GroupStat, error) {
return nil, errors.New("not implemented")
}
func (r *stubUsageLogRepo) GetAPIKeyUsageTrend(ctx context.Context, startTime, endTime time.Time, granularity string, limit int) ([]usagestats.APIKeyUsageTrendPoint, error) {
return nil, errors.New("not implemented")
}

View File

@@ -1,36 +0,0 @@
//go:build unit
package server
import (
"testing"
"github.com/stretchr/testify/assert"
)
func TestExtractOrigin(t *testing.T) {
tests := []struct {
name string
input string
want string
}{
{"empty string", "", ""},
{"whitespace only", " ", ""},
{"valid https", "https://pay.example.com/checkout", "https://pay.example.com"},
{"valid http", "http://pay.example.com/checkout", "http://pay.example.com"},
{"https with port", "https://pay.example.com:8443/checkout", "https://pay.example.com:8443"},
{"protocol-relative //host", "//pay.example.com/path", ""},
{"no scheme", "pay.example.com/path", ""},
{"ftp scheme rejected", "ftp://pay.example.com/file", ""},
{"empty host after parse", "https:///path", ""},
{"invalid url", "://bad url", ""},
{"only scheme", "https://", ""},
}
for _, tt := range tests {
t.Run(tt.name, func(t *testing.T) {
got := extractOrigin(tt.input)
assert.Equal(t, tt.want, got)
})
}
}

View File

@@ -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

View File

@@ -0,0 +1,42 @@
package service
import (
"testing"
"github.com/stretchr/testify/require"
)
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: intPtr(20)}
require.Equal(t, 20, a.EffectiveLoadFactor())
}
func TestEffectiveLoadFactor_ZeroLoadFactor_FallbackToConcurrency(t *testing.T) {
a := &Account{Concurrency: 5, LoadFactor: intPtr(0)}
require.Equal(t, 5, a.EffectiveLoadFactor())
}
func TestEffectiveLoadFactor_NegativeLoadFactor_FallbackToConcurrency(t *testing.T) {
a := &Account{Concurrency: 3, LoadFactor: intPtr(-1)}
require.Equal(t, 3, a.EffectiveLoadFactor())
}
func TestEffectiveLoadFactor_ZeroLoadFactor_ZeroConcurrency(t *testing.T) {
a := &Account{Concurrency: 0, LoadFactor: intPtr(0)}
require.Equal(t, 1, a.EffectiveLoadFactor())
}

View File

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

View File

@@ -198,6 +198,7 @@ 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
@@ -218,6 +219,7 @@ 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
@@ -233,6 +235,7 @@ 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
@@ -1433,6 +1436,9 @@ func (s *adminServiceImpl) CreateAccount(ctx context.Context, input *CreateAccou
}
account.RateMultiplier = input.RateMultiplier
}
if input.LoadFactor != nil && *input.LoadFactor > 0 {
account.LoadFactor = input.LoadFactor
}
if err := s.accountRepo.Create(ctx, account); err != nil {
return nil, err
}
@@ -1507,6 +1513,13 @@ 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 {
account.LoadFactor = input.LoadFactor
}
}
if input.Status != "" {
account.Status = input.Status
}
@@ -1640,6 +1653,9 @@ func (s *adminServiceImpl) BulkUpdateAccounts(ctx context.Context, input *BulkUp
if input.RateMultiplier != nil {
repoUpdates.RateMultiplier = input.RateMultiplier
}
if input.LoadFactor != nil {
repoUpdates.LoadFactor = input.LoadFactor
}
if input.Status != "" {
repoUpdates.Status = &input.Status
}

View File

@@ -1322,7 +1322,7 @@ 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)
@@ -1515,7 +1515,7 @@ 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(),
})
}

View File

@@ -590,7 +590,7 @@ 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 {

View File

@@ -1250,7 +1250,7 @@ func (s *OpenAIGatewayService) SelectAccountWithLoadAwareness(ctx context.Contex
for _, acc := range candidates {
accountLoads = append(accountLoads, AccountWithConcurrency{
ID: acc.ID,
MaxConcurrency: acc.Concurrency,
MaxConcurrency: acc.EffectiveLoadFactor(),
})
}

View File

@@ -64,8 +64,9 @@ 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
lf := acc.EffectiveLoadFactor()
if prev, ok := unique[acc.ID]; !ok || lf > prev {
unique[acc.ID] = lf
}
}

View File

@@ -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.EffectiveLoadFactor(),
})
}
if len(batch) == 0 {