Skip to content

GitLab

"backend/internal/vscode:/vscode.git/clone" did not exist on "005d0c5f53615e474e2dbb437008044c1b777fa4"
Commit a07174c1 authored by shaw's avatar shaw
Browse files

fix: 修复会话限制功能并在创建账号时支持配额控制

parent 32fff379
Show whitespace changes
Inline Side-by-side
backend/internal/service/gateway_service.go
View file @ a07174c1
......@@ -410,11 +410,9 @@ 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) {
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 {
......@@ -440,18 +438,39 @@ func (s *GatewayService) SelectAccountWithLoadAwareness(ctx context.Context, gro
}
if s.concurrencyService == nil || !cfg.LoadBatchEnabled {
account, err := s.SelectAccountForModelWithExclusions(ctx, groupID, sessionHash, requestedModel, excludedIDs)
// 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
}
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,
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 {
......@@ -476,6 +495,7 @@ func (s *GatewayService) SelectAccountWithLoadAwareness(ctx context.Context, gro
},
}, nil
}
}
platform, hasForcePlatform, err := s.resolvePlatform(ctx, groupID, group)
if err != nil {
......@@ -590,7 +610,7 @@ 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 {
......@@ -608,6 +628,10 @@ func (s *GatewayService) SelectAccountWithLoadAwareness(ctx context.Context, gro
waitingCount, _ := s.concurrencyService.GetAccountWaitingCount(ctx, stickyAccountID)
if waitingCount < cfg.StickySessionMaxWaiting {
// 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{
......@@ -618,6 +642,7 @@ func (s *GatewayService) SelectAccountWithLoadAwareness(ctx context.Context, gro
},
}, 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
}
}
......@@ -677,7 +702,7 @@ 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
}
......@@ -695,21 +720,27 @@ 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
// 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), acc.ID)
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: acc,
Account: item.account,
WaitPlan: &AccountWaitPlan{
AccountID: acc.ID,
MaxConcurrency: acc.Concurrency,
AccountID: item.account.ID,
MaxConcurrency: item.account.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)
}
......@@ -728,7 +759,7 @@ func (s *GatewayService) SelectAccountWithLoadAwareness(ctx context.Context, gro
result, err := s.tryAcquireAccountSlot(ctx, accountID, account.Concurrency)
if err == nil && result.Acquired {
// 会话数量限制检查
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)
......@@ -742,6 +773,10 @@ func (s *GatewayService) SelectAccountWithLoadAwareness(ctx context.Context, gro
waitingCount, _ := s.concurrencyService.GetAccountWaitingCount(ctx, accountID)
if waitingCount < cfg.StickySessionMaxWaiting {
// 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, account, sessionHash) {
// 会话限制已满，继续到 Layer 2
} else {
return &AccountSelectionResult{
Account: account,
WaitPlan: &AccountWaitPlan{
......@@ -755,6 +790,7 @@ func (s *GatewayService) SelectAccountWithLoadAwareness(ctx context.Context, gro
}
}
}
}
// ============ Layer 2: 负载感知选择 ============
candidates := make([]*Account, 0, len(accounts))
......@@ -799,7 +835,7 @@ 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 {
......@@ -849,7 +885,7 @@ 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
}
......@@ -869,6 +905,10 @@ func (s *GatewayService) SelectAccountWithLoadAwareness(ctx context.Context, gro
// ============ Layer 3: 兜底排队 ============
sortAccountsByPriorityAndLastUsed(candidates, preferOAuth)
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{
......@@ -882,7 +922,7 @@ 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)
......@@ -890,7 +930,7 @@ 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
}
......@@ -1188,15 +1228,16 @@ 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
}
......@@ -1206,7 +1247,7 @@ 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
......
frontend/src/components/account/CreateAccountModal.vue
View file @ a07174c1
......@@ -1191,6 +1191,136 @@
</div>
</div>
<!-- Quota Control Section (Anthropic OAuth/SetupToken only) -->
<div
v-if="form.platform === 'anthropic' && accountCategory === 'oauth-based'"
class="border-t border-gray-200 pt-4 dark:border-dark-600 space-y-4"
>
<div class="mb-3">
<h3 class="input-label mb-0 text-base font-semibold">{{ t('admin.accounts.quotaControl.title') }}</h3>
<p class="mt-1 text-xs text-gray-500 dark:text-gray-400">
{{ t('admin.accounts.quotaControl.hint') }}
</p>
</div>
<!-- Window Cost Limit -->
<div class="rounded-lg border border-gray-200 p-4 dark:border-dark-600">
<div class="mb-3 flex items-center justify-between">
<div>
<label class="input-label mb-0">{{ t('admin.accounts.quotaControl.windowCost.label') }}</label>
<p class="mt-1 text-xs text-gray-500 dark:text-gray-400">
{{ t('admin.accounts.quotaControl.windowCost.hint') }}
</p>
</div>
<button
type="button"
@click="windowCostEnabled = !windowCostEnabled"
:class="[
'relative inline-flex h-6 w-11 flex-shrink-0 cursor-pointer rounded-full border-2 border-transparent transition-colors duration-200 ease-in-out focus:outline-none focus:ring-2 focus:ring-primary-500 focus:ring-offset-2',
windowCostEnabled ? 'bg-primary-600' : 'bg-gray-200 dark:bg-dark-600'
]"
>
<span
:class="[
'pointer-events-none inline-block h-5 w-5 transform rounded-full bg-white shadow ring-0 transition duration-200 ease-in-out',
windowCostEnabled ? 'translate-x-5' : 'translate-x-0'
]"
/>
</button>
</div>
<div v-if="windowCostEnabled" class="grid grid-cols-2 gap-4">
<div>
<label class="input-label">{{ t('admin.accounts.quotaControl.windowCost.limit') }}</label>
<div class="relative">
<span class="absolute left-3 top-1/2 -translate-y-1/2 text-gray-500 dark:text-gray-400">$</span>
<input
v-model.number="windowCostLimit"
type="number"
min="0"
step="1"
class="input pl-7"
:placeholder="t('admin.accounts.quotaControl.windowCost.limitPlaceholder')"
/>
</div>
<p class="input-hint">{{ t('admin.accounts.quotaControl.windowCost.limitHint') }}</p>
</div>
<div>
<label class="input-label">{{ t('admin.accounts.quotaControl.windowCost.stickyReserve') }}</label>
<div class="relative">
<span class="absolute left-3 top-1/2 -translate-y-1/2 text-gray-500 dark:text-gray-400">$</span>
<input
v-model.number="windowCostStickyReserve"
type="number"
min="0"
step="1"
class="input pl-7"
:placeholder="t('admin.accounts.quotaControl.windowCost.stickyReservePlaceholder')"
/>
</div>
<p class="input-hint">{{ t('admin.accounts.quotaControl.windowCost.stickyReserveHint') }}</p>
</div>
</div>
</div>
<!-- Session Limit -->
<div class="rounded-lg border border-gray-200 p-4 dark:border-dark-600">
<div class="mb-3 flex items-center justify-between">
<div>
<label class="input-label mb-0">{{ t('admin.accounts.quotaControl.sessionLimit.label') }}</label>
<p class="mt-1 text-xs text-gray-500 dark:text-gray-400">
{{ t('admin.accounts.quotaControl.sessionLimit.hint') }}
</p>
</div>
<button
type="button"
@click="sessionLimitEnabled = !sessionLimitEnabled"
:class="[
'relative inline-flex h-6 w-11 flex-shrink-0 cursor-pointer rounded-full border-2 border-transparent transition-colors duration-200 ease-in-out focus:outline-none focus:ring-2 focus:ring-primary-500 focus:ring-offset-2',
sessionLimitEnabled ? 'bg-primary-600' : 'bg-gray-200 dark:bg-dark-600'
]"
>
<span
:class="[
'pointer-events-none inline-block h-5 w-5 transform rounded-full bg-white shadow ring-0 transition duration-200 ease-in-out',
sessionLimitEnabled ? 'translate-x-5' : 'translate-x-0'
]"
/>
</button>
</div>
<div v-if="sessionLimitEnabled" class="grid grid-cols-2 gap-4">
<div>
<label class="input-label">{{ t('admin.accounts.quotaControl.sessionLimit.maxSessions') }}</label>
<input
v-model.number="maxSessions"
type="number"
min="1"
step="1"
class="input"
:placeholder="t('admin.accounts.quotaControl.sessionLimit.maxSessionsPlaceholder')"
/>
<p class="input-hint">{{ t('admin.accounts.quotaControl.sessionLimit.maxSessionsHint') }}</p>
</div>
<div>
<label class="input-label">{{ t('admin.accounts.quotaControl.sessionLimit.idleTimeout') }}</label>
<div class="relative">
<input
v-model.number="sessionIdleTimeout"
type="number"
min="1"
step="1"
class="input pr-12"
:placeholder="t('admin.accounts.quotaControl.sessionLimit.idleTimeoutPlaceholder')"
/>
<span class="absolute right-3 top-1/2 -translate-y-1/2 text-gray-500 dark:text-gray-400">{{ t('common.minutes') }}</span>
</div>
<p class="input-hint">{{ t('admin.accounts.quotaControl.sessionLimit.idleTimeoutHint') }}</p>
</div>
</div>
</div>
</div>
<div>
<label class="input-label">{{ t('admin.accounts.proxy') }}</label>
<ProxySelector v-model="form.proxy_id" :proxies="proxies" />
......@@ -1763,6 +1893,14 @@ const geminiAIStudioOAuthEnabled = ref(false)
const showAdvancedOAuth = ref(false)
const showGeminiHelpDialog = ref(false)
// Quota control state (Anthropic OAuth/SetupToken only)
const windowCostEnabled = ref(false)
const windowCostLimit = ref<number | null>(null)
const windowCostStickyReserve = ref<number | null>(null)
const sessionLimitEnabled = ref(false)
const maxSessions = ref<number | null>(null)
const sessionIdleTimeout = ref<number | null>(null)
// Gemini tier selection (used as fallback when auto-detection is unavailable/fails)
const geminiTierGoogleOne = ref<'google_one_free' | 'google_ai_pro' | 'google_ai_ultra'>('google_one_free')
const geminiTierGcp = ref<'gcp_standard' | 'gcp_enterprise'>('gcp_standard')
......@@ -2140,6 +2278,13 @@ const resetForm = () => {
customErrorCodeInput.value = null
interceptWarmupRequests.value = false
autoPauseOnExpired.value = true
// Reset quota control state
windowCostEnabled.value = false
windowCostLimit.value = null
windowCostStickyReserve.value = null
sessionLimitEnabled.value = false
maxSessions.value = null
sessionIdleTimeout.value = null
tempUnschedEnabled.value = false
tempUnschedRules.value = []
geminiOAuthType.value = 'code_assist'
......@@ -2407,7 +2552,22 @@ const handleAnthropicExchange = async (authCode: string) => {
...proxyConfig
})
const extra = oauth.buildExtraInfo(tokenInfo)
// Build extra with quota control settings
const baseExtra = oauth.buildExtraInfo(tokenInfo) || {}
const extra: Record<string, unknown> = { ...baseExtra }
// Add window cost limit settings
if (windowCostEnabled.value && windowCostLimit.value != null && windowCostLimit.value > 0) {
extra.window_cost_limit = windowCostLimit.value
extra.window_cost_sticky_reserve = windowCostStickyReserve.value ?? 10
}
// Add session limit settings
if (sessionLimitEnabled.value && maxSessions.value != null && maxSessions.value > 0) {
extra.max_sessions = maxSessions.value
extra.session_idle_timeout_minutes = sessionIdleTimeout.value ?? 5
}
const credentials = {
...tokenInfo,
...(interceptWarmupRequests.value ? { intercept_warmup_requests: true } : {})
......@@ -2475,7 +2635,22 @@ const handleCookieAuth = async (sessionKey: string) => {
...proxyConfig
})
const extra = oauth.buildExtraInfo(tokenInfo)
// Build extra with quota control settings
const baseExtra = oauth.buildExtraInfo(tokenInfo) || {}
const extra: Record<string, unknown> = { ...baseExtra }
// Add window cost limit settings
if (windowCostEnabled.value && windowCostLimit.value != null && windowCostLimit.value > 0) {
extra.window_cost_limit = windowCostLimit.value
extra.window_cost_sticky_reserve = windowCostStickyReserve.value ?? 10
}
// Add session limit settings
if (sessionLimitEnabled.value && maxSessions.value != null && maxSessions.value > 0) {
extra.max_sessions = maxSessions.value
extra.session_idle_timeout_minutes = sessionIdleTimeout.value ?? 5
}
const accountName = keys.length > 1 ? `${form.name} #${i + 1}` : form.name
// Merge interceptWarmupRequests into credentials
......
Markdown is supported
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment