perf: 全栈查询优化、前端缓存去重与页面可见性优化

后端:
- SQL count 查询统一改用 func.count() 子查询替代 query.count()
- Dashboard/Audit 等页面多次独立查询合并为单次聚合查询
- Provider summary 列表改为批量查询消除 N+1 问题
- DailyStats 逐天循环查询改为 CASE 分桶单次查询
- 使用 load_only() 减少不必要的列加载
- cache_decorator 支持嵌套属性路径解析(dotted vary_by)
- 多个管理/公共端点新增 @cache_result 缓存装饰

前端:
- cache.ts 新增 in-flight 请求复用、dedupedRequest、buildCacheKey
- 大量 API 调用添加前端缓存或去重
- 多个页面定时器在标签页隐藏时暂停、可见时恢复
- Auth 检查从 setInterval 改为 storage + visibilitychange 事件驱动
- 请求竞态防护(requestId 模式)

数据库:
- Usage 表新增 idx_usage_status_user_created 复合索引

alembic/versions/20260303_1730_5f1d2e3c4b5a_add_idx_usage_status_user_created.py

@@ -0,0 +1,45 @@
+"""add_idx_usage_status_user_created
+
+Add composite index on usage(status, user_id, created_at) to speed up
+interval timeline and active usage analytics queries.
+
+Revision ID: 5f1d2e3c4b5a
+Revises: 0ba031f328de
+Create Date: 2026-03-03 17:30:00.000000+00:00
+"""
+
+from __future__ import annotations
+
+import sqlalchemy as sa
+from alembic import op
+
+revision = "5f1d2e3c4b5a"
+down_revision = "0ba031f328de"
+branch_labels = None
+depends_on = None
+
+INDEX_NAME = "idx_usage_status_user_created"
+TABLE = "usage"
+COLUMNS = ["status", "user_id", "created_at"]
+
+
+def upgrade() -> None:
+    bind = op.get_bind()
+    result = bind.execute(
+        sa.text("SELECT 1 FROM pg_indexes WHERE indexname = :name"),
+        {"name": INDEX_NAME},
+    ).fetchone()
+    if result:
+        return
+    op.create_index(INDEX_NAME, TABLE, COLUMNS)
+
+
+def downgrade() -> None:
+    bind = op.get_bind()
+    result = bind.execute(
+        sa.text("SELECT 1 FROM pg_indexes WHERE indexname = :name"),
+        {"name": INDEX_NAME},
+    ).fetchone()
+    if not result:
+        return
+    op.drop_index(INDEX_NAME, table_name=TABLE)

frontend/src/api/admin.ts

@@ -1,4 +1,5 @@
 import apiClient from './client'
+import { cachedRequest, buildCacheKey } from '@/utils/cache'
 
 // LDAP 配置导出结构
 export interface LDAPConfigExport {
@@ -743,10 +744,17 @@ export const adminApi = {
     include_inactive?: boolean
     exclude_admin?: boolean
   }): Promise<LeaderboardResponse> {
-    const response = await apiClient.get<LeaderboardResponse>('/api/admin/stats/leaderboard/users', {
+    const cacheKey = buildCacheKey('admin:stats:leaderboard:users', params)
-      params
+    return cachedRequest(
-    })
+      cacheKey,
-    return response.data
+      async () => {
+        const response = await apiClient.get<LeaderboardResponse>('/api/admin/stats/leaderboard/users', {
+          params
+        })
+        return response.data
+      },
+      20 * 1000
+    )
   },
 
   async getLeaderboardApiKeys(params?: {
@@ -764,10 +772,17 @@ export const adminApi = {
     include_inactive?: boolean
     exclude_admin?: boolean
   }): Promise<LeaderboardResponse> {
-    const response = await apiClient.get<LeaderboardResponse>('/api/admin/stats/leaderboard/api-keys', {
+    const cacheKey = buildCacheKey('admin:stats:leaderboard:api-keys', params)
-      params
+    return cachedRequest(
-    })
+      cacheKey,
-    return response.data
+      async () => {
+        const response = await apiClient.get<LeaderboardResponse>('/api/admin/stats/leaderboard/api-keys', {
+          params
+        })
+        return response.data
+      },
+      20 * 1000
+    )
   },
 
   async getLeaderboardModels(params?: {
@@ -783,10 +798,17 @@ export const adminApi = {
     provider_name?: string
     model?: string
   }): Promise<LeaderboardResponse> {
-    const response = await apiClient.get<LeaderboardResponse>('/api/admin/stats/leaderboard/models', {
+    const cacheKey = buildCacheKey('admin:stats:leaderboard:models', params)
-      params
+    return cachedRequest(
-    })
+      cacheKey,
-    return response.data
+      async () => {
+        const response = await apiClient.get<LeaderboardResponse>('/api/admin/stats/leaderboard/models', {
+          params
+        })
+        return response.data
+      },
+      20 * 1000
+    )
   },
 
   async getCostForecast(params?: {
@@ -798,10 +820,17 @@ export const adminApi = {
     days?: number
     forecast_days?: number
   }): Promise<CostForecastResponse> {
-    const response = await apiClient.get<CostForecastResponse>('/api/admin/stats/cost/forecast', {
+    const cacheKey = buildCacheKey('admin:stats:cost:forecast', params)
-      params
+    return cachedRequest(
-    })
+      cacheKey,
-    return response.data
+      async () => {
+        const response = await apiClient.get<CostForecastResponse>('/api/admin/stats/cost/forecast', {
+          params
+        })
+        return response.data
+      },
+      30 * 1000
+    )
   },
 
   async getCostSavings(params?: {
@@ -813,15 +842,28 @@ export const adminApi = {
     provider_name?: string
     model?: string
   }): Promise<CostSavingsResponse> {
-    const response = await apiClient.get<CostSavingsResponse>('/api/admin/stats/cost/savings', {
+    const cacheKey = buildCacheKey('admin:stats:cost:savings', params)
-      params
+    return cachedRequest(
-    })
+      cacheKey,
-    return response.data
+      async () => {
+        const response = await apiClient.get<CostSavingsResponse>('/api/admin/stats/cost/savings', {
+          params
+        })
+        return response.data
+      },
+      30 * 1000
+    )
   },
 
   async getQuotaUsage(): Promise<QuotaUsageResponse> {
-    const response = await apiClient.get<QuotaUsageResponse>('/api/admin/stats/providers/quota-usage')
+    return cachedRequest(
-    return response.data
+      'admin:stats:providers:quota-usage',
+      async () => {
+        const response = await apiClient.get<QuotaUsageResponse>('/api/admin/stats/providers/quota-usage')
+        return response.data
+      },
+      30 * 1000
+    )
   },
 
   async getPercentiles(params?: {
@@ -831,10 +873,17 @@ export const adminApi = {
     timezone?: string
     tz_offset_minutes?: number
   }): Promise<PercentileItem[]> {
-    const response = await apiClient.get<PercentileItem[]>('/api/admin/stats/performance/percentiles', {
+    const cacheKey = buildCacheKey('admin:stats:performance:percentiles', params)
-      params
+    return cachedRequest(
-    })
+      cacheKey,
-    return response.data
+      async () => {
+        const response = await apiClient.get<PercentileItem[]>('/api/admin/stats/performance/percentiles', {
+          params
+        })
+        return response.data
+      },
+      20 * 1000
+    )
   },
 
   async getErrorDistribution(params?: {
@@ -844,10 +893,17 @@ export const adminApi = {
     timezone?: string
     tz_offset_minutes?: number
   }): Promise<ErrorDistributionResponse> {
-    const response = await apiClient.get<ErrorDistributionResponse>('/api/admin/stats/errors/distribution', {
+    const cacheKey = buildCacheKey('admin:stats:errors:distribution', params)
-      params
+    return cachedRequest(
-    })
+      cacheKey,
-    return response.data
+      async () => {
+        const response = await apiClient.get<ErrorDistributionResponse>('/api/admin/stats/errors/distribution', {
+          params
+        })
+        return response.data
+      },
+      20 * 1000
+    )
   },
 
   async getComparison(params: {
@@ -882,8 +938,15 @@ export const adminApi = {
     model?: string
     provider_name?: string
   }): Promise<Array<Record<string, unknown>>> {
-    const response = await apiClient.get<Array<Record<string, unknown>>>('/api/admin/stats/time-series', { params })
+    const cacheKey = buildCacheKey('admin:stats:time-series', params)
-    return response.data
+    return cachedRequest(
+      cacheKey,
+      async () => {
+        const response = await apiClient.get<Array<Record<string, unknown>>>('/api/admin/stats/time-series', { params })
+        return response.data
+      },
+      20 * 1000
+    )
   },
 
 }

frontend/src/api/cache.ts

@@ -3,6 +3,7 @@
  */
 
 import api from './client'
+import { cachedRequest, buildCacheKey } from '@/utils/cache'
 
 export interface CacheStats {
   scheduler: string
@@ -314,8 +315,15 @@ export const cacheAnalysisApi = {
     user_id?: string
     include_user_info?: boolean
   }): Promise<IntervalTimelineResponse> {
-    const response = await api.get('/api/admin/usage/cache-affinity/interval-timeline', { params })
+    const cacheKey = buildCacheKey('cache-affinity:interval-timeline', params as Record<string, unknown> | undefined)
-    return response.data
+    return cachedRequest(
+      cacheKey,
+      async () => {
+        const response = await api.get('/api/admin/usage/cache-affinity/interval-timeline', { params })
+        return response.data
+      },
+      30000
+    )
   }
 }
 

frontend/src/api/dashboard.ts

@@ -1,4 +1,5 @@
 import apiClient from './client'
+import { cachedRequest, buildCacheKey } from '@/utils/cache'
 
 export interface DashboardStat {
   name: string
@@ -289,8 +290,15 @@ export interface TimeRangeParams {
 export const dashboardApi = {
   // 获取仪表盘统计数据
   async getStats(params?: TimeRangeParams): Promise<DashboardStatsResponse> {
-    const response = await apiClient.get<DashboardStatsResponse>('/api/dashboard/stats', { params })
+    const cacheKey = buildCacheKey('dashboard:stats', params)
-    return response.data
+    return cachedRequest(
+      cacheKey,
+      async () => {
+        const response = await apiClient.get<DashboardStatsResponse>('/api/dashboard/stats', { params })
+        return response.data
+      },
+      10 * 1000
+    )
   },
 
   // 获取最近的请求记录
@@ -303,8 +311,14 @@ export const dashboardApi = {
 
   // 获取提供商状态
   async getProviderStatus(): Promise<ProviderStatus[]> {
-    const response = await apiClient.get<ProviderStatusResponse>('/api/dashboard/provider-status')
+    return cachedRequest(
-    return response.data.providers
+      'dashboard:provider-status',
+      async () => {
+        const response = await apiClient.get<ProviderStatusResponse>('/api/dashboard/provider-status')
+        return response.data.providers
+      },
+      20 * 1000
+    )
   },
 
   // 获取请求详情
@@ -316,10 +330,17 @@ export const dashboardApi = {
 
   // 获取每日统计数据
   async getDailyStats(params?: TimeRangeParams & { days?: number }): Promise<DailyStatsResponse> {
-    const response = await apiClient.get<DailyStatsResponse>('/api/dashboard/daily-stats', {
+    const cacheKey = buildCacheKey('dashboard:daily-stats', params)
-      params
+    return cachedRequest(
-    })
+      cacheKey,
-    return response.data
+      async () => {
+        const response = await apiClient.get<DailyStatsResponse>('/api/dashboard/daily-stats', {
+          params
+        })
+        return response.data
+      },
+      20 * 1000
+    )
   },
 
   // 获取 cURL 命令数据（含明文 API Key）

frontend/src/api/endpoints/global-models.ts

@@ -1,4 +1,5 @@
 import client from '../client'
+import { dedupedRequest, buildCacheKey } from '@/utils/cache'
 import type {
   GlobalModelCreate,
   GlobalModelUpdate,
@@ -27,16 +28,21 @@ export async function getGlobalModels(params?: {
   is_active?: boolean
   search?: string
 }): Promise<GlobalModelListResponse> {
-  const response = await client.get('/api/admin/models/global', { params })
+  const key = buildCacheKey('global-models:list', params as Record<string, unknown> | undefined)
-  return response.data
+  return dedupedRequest(key, async () => {
+    const response = await client.get('/api/admin/models/global', { params })
+    return response.data
+  })
 }
 
 /**
  * 获取单个 GlobalModel 详情
  */
 export async function getGlobalModel(id: string): Promise<GlobalModelWithStats> {
-  const response = await client.get(`/api/admin/models/global/${id}`)
+  return dedupedRequest(`global-models:detail:${id}`, async () => {
-  return response.data
+    const response = await client.get(`/api/admin/models/global/${id}`)
+    return response.data
+  })
 }
 
 /**
@@ -112,10 +118,12 @@ export async function getGlobalModelProviders(globalModelId: string): Promise<{
   providers: ModelCatalogProviderDetail[]
   total: number
 }> {
-  const response = await client.get(
+  return dedupedRequest(`global-models:providers:${globalModelId}`, async () => {
-    `/api/admin/models/global/${globalModelId}/providers`
+    const response = await client.get(
-  )
+      `/api/admin/models/global/${globalModelId}/providers`
-  return response.data
+    )
+    return response.data
+  })
 }
 
 /**

frontend/src/api/endpoints/pool.ts

@@ -1,4 +1,5 @@
 import client from '../client'
+import { dedupedRequest } from '@/utils/cache'
 import type { AllowedModels, ProxyConfig } from './types/provider'
 
 export interface PoolKeyStatus {
@@ -175,16 +176,21 @@ export interface PoolBatchAction {
 }
 
 export async function getPoolOverview(): Promise<PoolOverviewResponse> {
-  const response = await client.get('/api/admin/pool/overview')
+  return dedupedRequest('pool:overview', async () => {
-  return response.data
+    const response = await client.get<PoolOverviewResponse>('/api/admin/pool/overview')
+    return response.data
+  })
 }
 
 export async function listPoolKeys(
   providerId: string,
   params: PoolKeysQuery = {},
 ): Promise<PoolKeysPageResponse> {
-  const response = await client.get(`/api/admin/pool/${providerId}/keys`, { params })
+  const key = `pool:keys:${providerId}|${params.page ?? ''}|${params.page_size ?? ''}|${params.search ?? ''}|${params.status ?? ''}`
-  return response.data
+  return dedupedRequest(key, async () => {
+    const response = await client.get<PoolKeysPageResponse>(`/api/admin/pool/${providerId}/keys`, { params })
+    return response.data
+  })
 }
 
 export async function batchActionPoolKeys(

frontend/src/api/endpoints/providers.ts

@@ -1,4 +1,5 @@
 import client from '../client'
+import { dedupedRequest } from '@/utils/cache'
 import type {
   ClaudeCodeAdvancedConfig,
   FailoverRulesConfig,
@@ -11,16 +12,20 @@ import type {
  * 获取 Providers 摘要（包含 Endpoints 统计）
  */
 export async function getProvidersSummary(): Promise<ProviderWithEndpointsSummary[]> {
-  const response = await client.get('/api/admin/providers/summary')
+  return dedupedRequest('providers:summary', async () => {
-  return response.data
+    const response = await client.get<ProviderWithEndpointsSummary[]>('/api/admin/providers/summary')
+    return response.data
+  })
 }
 
 /**
  * 获取单个 Provider 的详细信息
  */
 export async function getProvider(providerId: string): Promise<ProviderWithEndpointsSummary> {
-  const response = await client.get(`/api/admin/providers/${providerId}/summary`)
+  return dedupedRequest(`providers:detail:${providerId}`, async () => {
-  return response.data
+    const response = await client.get<ProviderWithEndpointsSummary>(`/api/admin/providers/${providerId}/summary`)
+    return response.data
+  })
 }
 
 /**
@@ -214,6 +219,8 @@ export interface ProviderMappingPreviewResponse {
 export async function getProviderMappingPreview(
   providerId: string
 ): Promise<ProviderMappingPreviewResponse> {
-  const response = await client.get(`/api/admin/providers/${providerId}/mapping-preview`)
+  return dedupedRequest(`providers:mapping-preview:${providerId}`, async () => {
-  return response.data
+    const response = await client.get<ProviderMappingPreviewResponse>(`/api/admin/providers/${providerId}/mapping-preview`)
+    return response.data
+  })
 }

frontend/src/api/me.ts

@@ -1,6 +1,7 @@
 import apiClient from './client'
 import type { ActivityHeatmap } from '@/types/activity'
 import type { TieredPricingConfig } from './endpoints/types'
+import { cachedRequest, buildCacheKey } from '@/utils/cache'
 
 export interface Profile {
   id: string // UUID
@@ -328,8 +329,15 @@ export const meApi = {
     points: Array<{ x: string; y: number; model?: string }>
     models?: string[]
   }> {
-    const response = await apiClient.get('/api/users/me/usage/interval-timeline', { params })
+    const cacheKey = buildCacheKey('me:interval-timeline', params as Record<string, unknown> | undefined)
-    return response.data
+    return cachedRequest(
+      cacheKey,
+      async () => {
+        const response = await apiClient.get('/api/users/me/usage/interval-timeline', { params })
+        return response.data
+      },
+      30000
+    )
   },
 
   /**
@@ -337,7 +345,13 @@ export const meApi = {
    * 后端已缓存5分钟
    */
   async getActivityHeatmap(): Promise<ActivityHeatmap> {
-    const response = await apiClient.get<ActivityHeatmap>('/api/users/me/usage/heatmap')
+    return cachedRequest(
-    return response.data
+      'me-activity-heatmap',
+      async () => {
+        const response = await apiClient.get<ActivityHeatmap>('/api/users/me/usage/heatmap')
+        return response.data
+      },
+      60000
+    )
   }
 }

frontend/src/api/usage.ts

@@ -1,5 +1,5 @@
 import apiClient from './client'
-import { cachedRequest } from '@/utils/cache'
+import { cachedRequest, dedupedRequest, buildCacheKey } from '@/utils/cache'
 import type { ActivityHeatmap } from '@/types/activity'
 
 export interface UsageRecord {
@@ -187,8 +187,11 @@ export const usageApi = {
     limit: number
     offset: number
   }> {
-    const response = await apiClient.get('/api/admin/usage/records', { params })
+    const key = buildCacheKey('usage:records', params as Record<string, unknown> | undefined)
-    return response.data
+    return dedupedRequest(key, async () => {
+      const response = await apiClient.get('/api/admin/usage/records', { params })
+      return response.data
+    })
   },
 
   /**
@@ -226,7 +229,13 @@ export const usageApi = {
    * 后端已缓存5分钟
    */
   async getActivityHeatmap(): Promise<ActivityHeatmap> {
-    const response = await apiClient.get<ActivityHeatmap>('/api/admin/usage/heatmap')
+    return cachedRequest(
-    return response.data
+      'admin-usage-activity-heatmap',
+      async () => {
+        const response = await apiClient.get<ActivityHeatmap>('/api/admin/usage/heatmap')
+        return response.data
+      },
+      60000
+    )
   }
 }

frontend/src/features/providers/components/ProviderDetailDrawer.vue

@@ -1132,6 +1132,8 @@ const endpoints = ref<ProviderEndpointWithKeys[]>([])
 const providerKeys = ref<EndpointAPIKey[]>([])  // Provider 级别的 keys
 const providerModels = ref<Model[]>([])  // Provider 级别的 models
 const providerMappingPreview = ref<ProviderMappingPreviewResponse | null>(null)  // 映射预览
+let providerLoadRequestId = 0
+let endpointsLoadRequestId = 0
 
 // 系统级格式转换配置
 const systemFormatConversionEnabled = ref(false)
@@ -1281,12 +1283,19 @@ watch(
       }
       void autoRefreshQuotaInBackground()
     } else if (!newOpen && oldOpen) {
+      // 使在途请求失效，避免关闭后旧响应回写
+      providerLoadRequestId += 1
+      endpointsLoadRequestId += 1
+
       // 停止倒计时定时器
       stopCountdownTimer()
       // 重置所有状态
+      loading.value = false
       provider.value = null
       endpoints.value = []
       providerKeys.value = []  // 清空 Provider 级别的 keys
+      providerModels.value = []
+      providerMappingPreview.value = null
 
       // 重置分页状态
       resetKeysPagination()
@@ -2581,6 +2590,7 @@ async function loadSystemFormatConversionConfig() {
 // 加载 Provider 信息
 async function loadProvider() {
   if (!props.providerId) return
+  const requestId = ++providerLoadRequestId
 
   try {
     loading.value = true
@@ -2589,21 +2599,26 @@ async function loadProvider() {
       getProvider(props.providerId),
       loadSystemFormatConversionConfig(),
     ])
+    if (requestId !== providerLoadRequestId) return
     provider.value = providerData
 
     if (!provider.value) {
       throw new Error('Provider 不存在')
     }
   } catch (err: unknown) {
+    if (requestId !== providerLoadRequestId) return
     showError(parseApiError(err, '加载失败'), '错误')
   } finally {
-    loading.value = false
+    if (requestId === providerLoadRequestId) {
+      loading.value = false
+    }
   }
 }
 
 // 加载端点列表
 async function loadEndpoints() {
   if (!props.providerId) return
+  const requestId = ++endpointsLoadRequestId
 
   try {
     // 并行加载端点列表、Provider 级别的 keys、models 和映射预览
@@ -2613,6 +2628,7 @@ async function loadEndpoints() {
       getProviderModels(props.providerId).catch(() => []),
       getProviderMappingPreview(props.providerId).catch(() => null),
     ])
+    if (requestId !== endpointsLoadRequestId) return
 
     providerKeys.value = providerKeysResult
     providerModels.value = modelsResult
@@ -2627,6 +2643,7 @@ async function loadEndpoints() {
       return aIdx - bIdx
     })
   } catch (err: unknown) {
+    if (requestId !== endpointsLoadRequestId) return
     showError(parseApiError(err, '加载端点失败'), '错误')
   }
 }

frontend/src/features/usage/components/IntervalTimelineCard.vue

@@ -54,7 +54,7 @@
 </template>
 
 <script setup lang="ts">
-import { computed, ref, onMounted, watch } from 'vue'
+import { computed, ref, onMounted, onBeforeUnmount, watch } from 'vue'
 import Card from '@/components/ui/card.vue'
 import ScatterChart from '@/components/charts/ScatterChart.vue'
 import { cacheAnalysisApi, type IntervalTimelineResponse } from '@/api/cache'
@@ -73,6 +73,10 @@ const props = withDefaults(defineProps<{
 const loading = ref(false)
 const timelineData = ref<IntervalTimelineResponse | null>(null)
 const primaryColor = ref('201, 100, 66')  // 默认主题色
+let loadRequestId = 0
+
+const ADMIN_TIMELINE_LIMIT = 1500
+const USER_TIMELINE_LIMIT = 1200
 
 // 获取主题色
 function getPrimaryColor(): string {
@@ -86,7 +90,7 @@ function getPrimaryColor(): string {
 
 onMounted(() => {
   primaryColor.value = getPrimaryColor()
-  loadData()
+  void loadData()
 })
 
 // 预定义的颜色列表（用于区分不同用户/模型）
@@ -278,35 +282,44 @@ const chartOptions = computed<ChartOptions<'scatter'>>(() => ({
 }))
 
 async function loadData() {
+  const requestId = ++loadRequestId
   loading.value = true
   try {
+    const limit = props.isAdmin ? ADMIN_TIMELINE_LIMIT : USER_TIMELINE_LIMIT
     if (props.isAdmin) {
       // 管理员：获取所有用户数据（按比例采样）
-      timelineData.value = await cacheAnalysisApi.getIntervalTimeline({
+      const data = await cacheAnalysisApi.getIntervalTimeline({
         hours: props.hours,
         include_user_info: true,
-        limit: 10000,
+        limit,
       })
+      if (requestId !== loadRequestId) return
+      timelineData.value = data
     } else {
       // 普通用户：获取自己的数据
-      timelineData.value = await meApi.getIntervalTimeline({
+      const data = await meApi.getIntervalTimeline({
         hours: props.hours,
-        limit: 5000,
+        limit,
       })
+      if (requestId !== loadRequestId) return
+      timelineData.value = data
     }
   } catch (error) {
+    if (requestId !== loadRequestId) return
     log.error('加载请求间隔时间线失败:', error)
     timelineData.value = null
   } finally {
-    loading.value = false
+    if (requestId === loadRequestId) {
+      loading.value = false
+    }
   }
 }
 
-watch(() => props.hours, () => {
+watch([() => props.hours, () => props.isAdmin], () => {
-  loadData()
+  void loadData()
 })
 
-watch(() => props.isAdmin, () => {
+onBeforeUnmount(() => {
-  loadData()
+  loadRequestId++
 })
 </script>

frontend/src/features/usage/components/RequestDetailDrawer.vue

@@ -657,7 +657,7 @@
 </template>
 
 <script setup lang="ts">
-import { ref, watch, computed, onBeforeUnmount } from 'vue'
+import { ref, watch, computed, onMounted, onBeforeUnmount } from 'vue'
 import Button from '@/components/ui/button.vue'
 import { useEscapeKey } from '@/composables/useEscapeKey'
 import { useClipboard } from '@/composables/useClipboard'
@@ -716,10 +716,13 @@ const historicalPricing = ref<{
 } | null>(null)
 const autoRefreshTimer = ref<ReturnType<typeof setInterval> | null>(null)
 const autoRefreshing = ref(false)
+const isPageVisible = ref(typeof document === 'undefined' ? true : !document.hidden)
 const curlCopying = ref(false)
 const curlCopied = ref(false)
 const replayDialogOpen = ref(false)
 const AUTO_REFRESH_INTERVAL_MS = 1000
+let loadDetailRequestId = 0
+let loadDetailInFlight = false
 
 // 监听标签页切换
 watch(activeTab, (newTab) => {
@@ -1097,13 +1100,20 @@ watch(() => props.isOpen, async (isOpen) => {
 })
 
 async function loadDetail(id: string, silent = false) {
+  if (silent && loadDetailInFlight) {
+    return
+  }
+  const requestId = ++loadDetailRequestId
+  loadDetailInFlight = true
   if (!silent) {
     loading.value = true
     historicalPricing.value = null
   }
   error.value = null
   try {
-    detail.value = await dashboardApi.getRequestDetail(id)
+    const response = await dashboardApi.getRequestDetail(id)
+    if (requestId !== loadDetailRequestId) return
+    detail.value = response
 
     // 首次加载时选择默认 tab
     if (!silent) {
@@ -1145,14 +1155,18 @@ async function loadDetail(id: string, silent = false) {
       }
     }
   } catch (err) {
+    if (requestId !== loadDetailRequestId) return
     log.error('Failed to load request detail:', err)
     if (!silent) {
       error.value = '加载请求详情失败'
     }
   } finally {
-    if (!silent) {
+    if (!silent && requestId === loadDetailRequestId) {
       loading.value = false
     }
+    if (requestId === loadDetailRequestId) {
+      loadDetailInFlight = false
+    }
   }
 }
 
@@ -1175,12 +1189,17 @@ function stopAutoRefresh() {
 }
 
 function startAutoRefresh() {
-  if (autoRefreshTimer.value || !props.requestId || !props.isOpen) {
+  if (autoRefreshTimer.value) {
+    autoRefreshing.value = true
+    return
+  }
+  if (!isPageVisible.value || !props.requestId || !props.isOpen) {
+    autoRefreshing.value = false
     return
   }
   autoRefreshing.value = true
   autoRefreshTimer.value = setInterval(async () => {
-    if (!props.requestId || !props.isOpen) {
+    if (!isPageVisible.value || !props.requestId || !props.isOpen) {
       stopAutoRefresh()
       return
     }
@@ -1218,8 +1237,26 @@ async function refreshDetail() {
   startAutoRefresh()
 }
 
+function handleVisibilityChange() {
+  isPageVisible.value = !document.hidden
+  if (!isPageVisible.value) {
+    stopAutoRefresh()
+    return
+  }
+  if (props.isOpen && props.requestId && !isRequestCompleted()) {
+    startAutoRefresh()
+  }
+}
+
+onMounted(() => {
+  document.addEventListener('visibilitychange', handleVisibilityChange)
+})
+
 onBeforeUnmount(() => {
+  document.removeEventListener('visibilitychange', handleVisibilityChange)
   stopAutoRefresh()
+  loadDetailRequestId += 1
+  loadDetailInFlight = false
 })
 
 function formatDateTime(dateStr: string | null | undefined): string {

frontend/src/features/usage/components/UsageRecordsTable.vue

@@ -767,7 +767,7 @@ const hasActiveRecords = computed(() => {
 // 使用 VueUse 的 useIntervalFn 管理计时器（自动清理）
 const { pause: stopTimer, resume: startTimer } = useIntervalFn(
   () => { now.value = Date.now() },
-  100,
+  500,
   { immediate: false }
 )
 

frontend/src/features/usage/composables/useUsageData.ts

@@ -52,6 +52,8 @@ export function useUsageData(options: UseUsageDataOptions) {
 
   // 当前的日期范围（用于分页请求）
   const currentDateRange = ref<DateRangeParams | undefined>(undefined)
+  let loadStatsRequestId = 0
+  let loadRecordsRequestId = 0
 
   // 可用的筛选选项（从统计数据获取，而不是从记录中）
   const availableModels = ref<string[]>([])
@@ -69,6 +71,7 @@ export function useUsageData(options: UseUsageDataOptions) {
 
   // 加载统计数据（不加载记录）
   async function loadStats(dateRange?: DateRangeParams) {
+    const requestId = ++loadStatsRequestId
     isLoadingStats.value = true
     currentDateRange.value = dateRange
 
@@ -82,6 +85,10 @@ export function useUsageData(options: UseUsageDataOptions) {
           usageApi.getUsageByApiFormat(dateRange)
         ])
 
+        if (requestId !== loadStatsRequestId) {
+          return
+        }
+
         // statsData may contain additional fields not declared in UsageStats
         const statsRaw = statsData as Record<string, unknown>
         stats.value = {
@@ -138,6 +145,9 @@ export function useUsageData(options: UseUsageDataOptions) {
       } else {
         // 用户页面
         const userData = await meApi.getUsage(dateRange)
+        if (requestId !== loadStatsRequestId) {
+          return
+        }
 
         stats.value = {
           total_requests: userData.total_requests || 0,
@@ -227,6 +237,9 @@ export function useUsageData(options: UseUsageDataOptions) {
           .sort((a, b) => b.request_count - a.request_count)
       }
     } catch (error: unknown) {
+      if (requestId !== loadStatsRequestId) {
+        return
+      }
       if (getErrorStatus(error) !== 403) {
         log.error('加载统计数据失败:', error)
       }
@@ -234,7 +247,9 @@ export function useUsageData(options: UseUsageDataOptions) {
       modelStats.value = []
       currentRecords.value = []
     } finally {
-      isLoadingStats.value = false
+      if (requestId === loadStatsRequestId) {
+        isLoadingStats.value = false
+      }
     }
   }
 
@@ -243,6 +258,7 @@ export function useUsageData(options: UseUsageDataOptions) {
     pagination: PaginationParams,
     filters?: FilterParams
   ): Promise<void> {
+    const requestId = ++loadRecordsRequestId
     isLoadingRecords.value = true
 
     try {
@@ -279,22 +295,33 @@ export function useUsageData(options: UseUsageDataOptions) {
         }
 
         const response = await usageApi.getAllUsageRecords(params)
+        if (requestId !== loadRecordsRequestId) {
+          return
+        }
         const nextRecords = (response.records || []) as UsageRecord[]
         currentRecords.value = mergeRecordStatus(currentRecords.value, nextRecords)
         totalRecords.value = response.total || 0
       } else {
         // 用户页面：使用用户 API
         const userData = await meApi.getUsage(params)
+        if (requestId !== loadRecordsRequestId) {
+          return
+        }
         const nextRecords = (userData.records || []) as UsageRecord[]
         currentRecords.value = mergeRecordStatus(currentRecords.value, nextRecords)
         totalRecords.value = userData.pagination?.total || currentRecords.value.length
       }
     } catch (error) {
+      if (requestId !== loadRecordsRequestId) {
+        return
+      }
       log.error('加载记录失败:', error)
       currentRecords.value = []
       totalRecords.value = 0
     } finally {
-      isLoadingRecords.value = false
+      if (requestId === loadRecordsRequestId) {
+        isLoadingRecords.value = false
+      }
     }
   }
 

frontend/src/layouts/MainLayout.vue

@@ -385,7 +385,6 @@ const isDemo = computed(() => isDemoMode())
 
 const showAuthError = ref(false)
 const mobileMenuOpen = ref(false)
-let authCheckInterval: number | null = null
 
 // 更新检查相关
 const showUpdateDialog = ref(false)
@@ -437,12 +436,35 @@ async function checkForUpdate() {
   }
 }
 
+function syncAuthNotice() {
+  authStore.syncToken()
+  showAuthError.value = !!authStore.user && !authStore.token
+}
+
+function handleStorageChange(event: StorageEvent) {
+  if (event.key === null || event.key === 'access_token') {
+    syncAuthNotice()
+  }
+}
+
+function handleVisibilityChange() {
+  if (!document.hidden) {
+    syncAuthNotice()
+  }
+}
+
+watch(
+  () => [authStore.user, authStore.token] as const,
+  () => {
+    showAuthError.value = !!authStore.user && !authStore.token
+  },
+  { immediate: true }
+)
+
 onMounted(() => {
-  authCheckInterval = setInterval(() => {
+  window.addEventListener('storage', handleStorageChange)
-    if (authStore.user && !authStore.token) {
+  document.addEventListener('visibilitychange', handleVisibilityChange)
-      showAuthError.value = true
+  syncAuthNotice()
-    }
-  }, 5000)
 
   // 管理员预加载模块状态（路由守卫会按需加载，这里提前加载以避免菜单闪烁）
   if (authStore.user?.role === 'admin' && !moduleStore.loaded && !moduleStore.loading) {
@@ -456,10 +478,8 @@ onMounted(() => {
 })
 
 onUnmounted(() => {
-  if (authCheckInterval) {
+  window.removeEventListener('storage', handleStorageChange)
-    clearInterval(authCheckInterval)
+  document.removeEventListener('visibilitychange', handleVisibilityChange)
-    authCheckInterval = null
-  }
 })
 
 function handleRelogin() {

frontend/src/utils/cache.ts

@@ -11,6 +11,7 @@ interface CacheItem<T> {
 
 class MemoryCache {
   private cache: Map<string, CacheItem<unknown>> = new Map()
+  private inFlight: Map<string, Promise<unknown>> = new Map()
   private defaultTTL = 60000 // 默认缓存60秒
 
   /**
@@ -61,6 +62,7 @@ class MemoryCache {
    */
   clear(): void {
     this.cache.clear()
+    this.inFlight.clear()
   }
 
   /**
@@ -81,6 +83,27 @@ class MemoryCache {
   size(): number {
     return this.cache.size
   }
+
+  /**
+   * 获取进行中的请求
+   */
+  getInFlight<T>(key: string): Promise<T> | null {
+    return (this.inFlight.get(key) as Promise<T> | undefined) ?? null
+  }
+
+  /**
+   * 标记进行中的请求
+   */
+  setInFlight<T>(key: string, promise: Promise<T>): void {
+    this.inFlight.set(key, promise as Promise<unknown>)
+  }
+
+  /**
+   * 清除进行中的请求
+   */
+  deleteInFlight(key: string): void {
+    this.inFlight.delete(key)
+  }
 }
 
 // 创建全局缓存实例
@@ -103,18 +126,62 @@ export async function cachedRequest<T>(
   ttl?: number
 ): Promise<T> {
   // 尝试从缓存获取
-  const cached = cache.get<T>(key)
+  if (ttl !== 0) {
-  if (cached !== null) {
+    const cached = cache.get<T>(key)
-    return cached
+    if (cached !== null) {
+      return cached
+    }
   }
 
-  // 缓存未命中，执行请求
+  // 命中进行中的同 key 请求，直接复用
-  const data = await fetcher()
+  const inFlight = cache.getInFlight<T>(key)
+  if (inFlight) {
+    return inFlight
+  }
 
-  // 存入缓存
+  // 缓存未命中，执行请求并登记为 in-flight
-  cache.set(key, data, ttl)
+  const request = (async () => {
+    try {
+      const data = await fetcher()
+      if (ttl !== 0) {
+        cache.set(key, data, ttl)
+      }
+      return data
+    } finally {
+      cache.deleteInFlight(key)
+    }
+  })()
 
-  return data
+  cache.setInFlight(key, request)
+  return request
 }
 
-export default cache
+/**
+ * 仅做请求去重（不缓存结果）
+ * 相同 key 的并发请求会复用同一个 Promise
+ */
+export function dedupedRequest<T>(
+  key: string,
+  fetcher: () => Promise<T>,
+): Promise<T> {
+  return cachedRequest(key, fetcher, 0)
+}
+
+/**
+ * 构建归一化的缓存 key
+ * 将 params 的 key 排序并过滤 undefined 值，确保相同参数生成相同 key
+ */
+export function buildCacheKey(prefix: string, params?: Record<string, unknown>): string {
+  if (!params) {
+    return prefix
+  }
+  const normalizedParams = Object.entries(params)
+    .filter(([, value]) => value !== undefined)
+    .sort(([a], [b]) => a.localeCompare(b))
+  if (normalizedParams.length === 0) {
+    return prefix
+  }
+  return `${prefix}:${JSON.stringify(normalizedParams)}`
+}
+
+export default cache

frontend/src/views/admin/AsyncTasks.vue

@@ -911,6 +911,8 @@ const showDetail = ref(false)
 const selectedTask = ref<AsyncTaskDetail | null>(null)
 const detailAutoRefresh = ref(false)
 let detailRefreshInterval: ReturnType<typeof setInterval> | null = null
+const isPageVisible = ref(typeof document === 'undefined' ? true : !document.hidden)
+let overviewRefreshInFlight = false
 
 // 使用记录详情抽屉状态
 const usageDetailOpen = ref(false)
@@ -953,6 +955,16 @@ async function fetchStats() {
   }
 }
 
+async function refreshOverview() {
+  if (overviewRefreshInFlight) return
+  overviewRefreshInFlight = true
+  try {
+    await Promise.all([fetchTasks(), fetchStats()])
+  } finally {
+    overviewRefreshInFlight = false
+  }
+}
+
 // 打开任务详情
 async function openTaskDetail(task: AsyncTaskItem) {
   try {
@@ -993,22 +1005,27 @@ function toggleDetailAutoRefresh() {
 
 // 开始详情自动刷新
 function startDetailAutoRefresh() {
+  if (!isPageVisible.value) return
   if (detailRefreshInterval) return
   // 立即刷新一次
   refreshTaskDetail()
   detailRefreshInterval = setInterval(() => {
-    if (selectedTask.value && showDetail.value) {
+    if (isPageVisible.value && selectedTask.value && showDetail.value) {
       refreshTaskDetail()
     }
   }, 5000)
 }
 
-// 停止详情自动刷新
+function pauseDetailAutoRefresh() {
-function stopDetailAutoRefresh() {
   if (detailRefreshInterval) {
     clearInterval(detailRefreshInterval)
     detailRefreshInterval = null
   }
+}
+
+// 停止详情自动刷新
+function stopDetailAutoRefresh() {
+  pauseDetailAutoRefresh()
   detailAutoRefresh.value = false
 }
 
@@ -1052,8 +1069,7 @@ async function cancelTask(task: AsyncTaskItem | AsyncTaskDetail) {
     toast({
       title: '任务已取消',
     })
-    fetchTasks()
+    await refreshOverview()
-    fetchStats()
     if (showDetail.value) {
       closeDetail()
     }
@@ -1222,11 +1238,11 @@ let autoRefreshInterval: ReturnType<typeof setInterval> | null = null
 const AUTO_REFRESH_INTERVAL = 5000 // 5秒
 
 function startAutoRefresh() {
+  if (!isPageVisible.value) return
   if (autoRefreshInterval) return
   autoRefreshInterval = setInterval(() => {
-    if (hasProcessingTasks.value && !loading.value) {
+    if (isPageVisible.value && hasProcessingTasks.value && !loading.value) {
-      fetchTasks()
+      refreshOverview()
-      fetchStats()
     }
   }, AUTO_REFRESH_INTERVAL)
 }
@@ -1240,19 +1256,35 @@ function stopAutoRefresh() {
 
 // 监听是否有进行中的任务，动态启停自动刷新
 watch(hasProcessingTasks, (has) => {
-  if (has) {
+  if (has && isPageVisible.value) {
     startAutoRefresh()
   } else {
     stopAutoRefresh()
   }
 }, { immediate: true })
 
+function handleVisibilityChange() {
+  isPageVisible.value = !document.hidden
+  if (!isPageVisible.value) {
+    stopAutoRefresh()
+    pauseDetailAutoRefresh()
+    return
+  }
+  if (hasProcessingTasks.value) {
+    startAutoRefresh()
+  }
+  if (detailAutoRefresh.value && selectedTask.value && showDetail.value) {
+    startDetailAutoRefresh()
+  }
+}
+
 onMounted(() => {
-  fetchTasks()
+  document.addEventListener('visibilitychange', handleVisibilityChange)
-  fetchStats()
+  refreshOverview()
 })
 
 onUnmounted(() => {
+  document.removeEventListener('visibilitychange', handleVisibilityChange)
   stopAutoRefresh()
   stopDetailAutoRefresh()
   clearTimeout(filterTimeout)

frontend/src/views/admin/AuditLogs.vue

@@ -406,7 +406,7 @@
 </template>
 
 <script setup lang="ts">
-import { ref, onMounted, computed } from 'vue'
+import { ref, onMounted, onBeforeUnmount, computed } from 'vue'
 import {
   Card,
   Button,
@@ -462,6 +462,7 @@ interface AuditLog {
 const loading = ref(false)
 const logs = ref<AuditLog[]>([])
 const selectedLog = ref<AuditLog | null>(null)
+let logsRequestId = 0
 
 // 搜索查询
 const searchQuery = ref('')
@@ -480,9 +481,9 @@ const currentPage = ref(1)
 const pageSize = ref(20)
 const totalRecords = ref(0)
 
-let loadTimeout: number
+let loadTimeout: number | null = null
 const debouncedLoadLogs = () => {
-  clearTimeout(loadTimeout)
+  if (loadTimeout !== null) clearTimeout(loadTimeout)
   loadTimeout = window.setTimeout(resetAndLoad, 500)
 }
 
@@ -493,6 +494,7 @@ const hasActiveFilters = computed(() => {
 })
 
 async function loadLogs() {
+  const requestId = ++logsRequestId
   loading.value = true
   try {
     const offset = (currentPage.value - 1) * pageSize.value
@@ -506,14 +508,18 @@ async function loadLogs() {
     }
 
     const data = await auditApi.getAuditLogs(filterParams)
+    if (requestId !== logsRequestId) return
     logs.value = data.items || []
     totalRecords.value = data.meta?.total ?? logs.value.length
   } catch (error) {
+    if (requestId !== logsRequestId) return
     log.error('获取审计日志失败:', error)
     logs.value = []
     totalRecords.value = 0
   } finally {
-    loading.value = false
+    if (requestId === logsRequestId) {
+      loading.value = false
+    }
   }
 }
 
@@ -711,4 +717,12 @@ function formatDateTime(dateStr: string): string {
 onMounted(() => {
   loadLogs()
 })
+
+onBeforeUnmount(() => {
+  if (loadTimeout !== null) {
+    clearTimeout(loadTimeout)
+    loadTimeout = null
+  }
+  logsRequestId += 1
+})
 </script>

frontend/src/views/admin/CacheMonitoring.vue

@@ -47,6 +47,7 @@ const clearingRowAffinityKey = ref<string | null>(null)
 const currentPage = ref(1)
 const pageSize = ref(20)
 const currentTime = ref(Math.floor(Date.now() / 1000))
+const isPageVisible = ref(typeof document === 'undefined' ? true : !document.hidden)
 
 // ==================== 模型映射缓存 ====================
 
@@ -236,6 +237,7 @@ function handlePageChange() {
 // ==================== 定时器管理 ====================
 
 function startCountdown() {
+  if (!isPageVisible.value) return
   if (countdownTimer) clearInterval(countdownTimer)
 
   countdownTimer = setInterval(() => {
@@ -260,6 +262,16 @@ function stopCountdown() {
   }
 }
 
+function handleVisibilityChange() {
+  isPageVisible.value = !document.hidden
+  if (!isPageVisible.value) {
+    stopCountdown()
+    return
+  }
+  currentTime.value = Math.floor(Date.now() / 1000)
+  startCountdown()
+}
+
 // ==================== 模型映射缓存方法 ====================
 
 async function fetchModelMappingStats() {
@@ -431,6 +443,7 @@ watch(tableKeyword, (value) => {
 })
 
 onMounted(() => {
+  document.addEventListener('visibilitychange', handleVisibilityChange)
   fetchCacheStats()
   fetchCacheConfig()
   fetchAffinityList()
@@ -441,6 +454,7 @@ onMounted(() => {
 })
 
 onBeforeUnmount(() => {
+  document.removeEventListener('visibilitychange', handleVisibilityChange)
   if (searchDebounceTimer) clearTimeout(searchDebounceTimer)
   stopCountdown()
 })

frontend/src/views/admin/CostAnalysis.vue

@@ -72,7 +72,7 @@
 </template>
 
 <script setup lang="ts">
-import { ref, computed, onMounted, watch } from 'vue'
+import { ref, computed, onMounted, onUnmounted, watch } from 'vue'
 import Card from '@/components/ui/card.vue'
 import { TimeRangePicker } from '@/components/common'
 import { CostForecastChart, QuotaProgressCard } from '@/components/stats'
@@ -93,6 +93,13 @@ const providerStats = ref<ProviderStatsItem[]>([])
 
 const forecastLoading = ref(false)
 const quotaLoading = ref(false)
+let forecastRequestId = 0
+let savingsRequestId = 0
+let quotaRequestId = 0
+let providerStatsRequestId = 0
+let loadAllPromise: Promise<void> | null = null
+let hasPendingLoadAll = false
+let loadAllDebounceTimer: ReturnType<typeof setTimeout> | null = null
 
 const forecastHistory = computed(() => forecast.value?.history || [])
 const forecastFuture = computed(() => forecast.value?.forecast || [])
@@ -108,40 +115,93 @@ function buildTimeRangeParams() {
 }
 
 async function loadForecast() {
+  const requestId = ++forecastRequestId
   forecastLoading.value = true
   try {
-    forecast.value = await adminApi.getCostForecast(buildTimeRangeParams())
+    const data = await adminApi.getCostForecast(buildTimeRangeParams())
+    if (requestId !== forecastRequestId) return
+    forecast.value = data
   } finally {
-    forecastLoading.value = false
+    if (requestId === forecastRequestId) {
+      forecastLoading.value = false
+    }
   }
 }
 
 async function loadSavings() {
-  costSavings.value = await adminApi.getCostSavings(buildTimeRangeParams())
+  const requestId = ++savingsRequestId
+  const data = await adminApi.getCostSavings(buildTimeRangeParams())
+  if (requestId !== savingsRequestId) return
+  costSavings.value = data
 }
 
 async function loadQuotaUsage() {
+  const requestId = ++quotaRequestId
   quotaLoading.value = true
   try {
     const response = await adminApi.getQuotaUsage()
+    if (requestId !== quotaRequestId) return
     quotaProviders.value = response.providers
   } finally {
-    quotaLoading.value = false
+    if (requestId === quotaRequestId) {
+      quotaLoading.value = false
+    }
   }
 }
 
 async function loadProviderStats() {
-  providerStats.value = await usageApi.getUsageByProvider({
+  const requestId = ++providerStatsRequestId
+  const stats = await usageApi.getUsageByProvider({
     ...buildTimeRangeParams(),
     limit: 8
   })
+  if (requestId !== providerStatsRequestId) return
+  providerStats.value = stats
 }
 
 async function loadAll() {
-  await Promise.all([loadForecast(), loadSavings(), loadQuotaUsage(), loadProviderStats()])
+  if (loadAllPromise) {
+    hasPendingLoadAll = true
+    return loadAllPromise
+  }
+  loadAllPromise = Promise.all([loadForecast(), loadSavings(), loadQuotaUsage(), loadProviderStats()])
+    .then(() => undefined)
+    .finally(() => {
+      loadAllPromise = null
+      if (hasPendingLoadAll) {
+        hasPendingLoadAll = false
+        void loadAll()
+      }
+    })
+  return loadAllPromise
 }
 
-watch(timeRange, loadAll, { deep: true })
+function scheduleLoadAll() {
+  if (loadAllDebounceTimer) {
+    clearTimeout(loadAllDebounceTimer)
+  }
+  loadAllDebounceTimer = setTimeout(() => {
+    loadAllDebounceTimer = null
+    void loadAll()
+  }, 120)
+}
 
-onMounted(loadAll)
+watch(timeRange, scheduleLoadAll, { deep: true })
+
+onMounted(() => {
+  void loadAll()
+})
+
+onUnmounted(() => {
+  if (loadAllDebounceTimer) {
+    clearTimeout(loadAllDebounceTimer)
+    loadAllDebounceTimer = null
+  }
+  hasPendingLoadAll = false
+  loadAllPromise = null
+  forecastRequestId += 1
+  savingsRequestId += 1
+  quotaRequestId += 1
+  providerStatsRequestId += 1
+})
 </script>

frontend/src/views/admin/ModelManagement.vue

@@ -611,7 +611,7 @@
 </template>
 
 <script setup lang="ts">
-import { ref, computed, onMounted, watch } from 'vue'
+import { ref, computed, onMounted, onBeforeUnmount, watch } from 'vue'
 import {
   Plus,
   Edit,
@@ -703,6 +703,11 @@ const editingModel = ref<GlobalModelResponse | null>(null)
 const globalModels = ref<GlobalModelResponse[]>([])
 const providers = ref<ProviderWithEndpointsSummary[]>([])
 const GLOBAL_MODELS_FETCH_PAGE_SIZE = 1000
+let globalModelsRequestId = 0
+let modelSelectionRequestId = 0
+let modelProvidersRequestId = 0
+let providersRequestId = 0
+let providerOptionsRequest: Promise<void> | null = null
 
 // 模型目录分页
 const catalogCurrentPage = ref(1)
@@ -1025,19 +1030,27 @@ watch([searchQuery, capabilityFilters], () => {
 }, { deep: true })
 
 async function loadGlobalModels() {
+  const requestId = ++globalModelsRequestId
   loading.value = true
   try {
     const allModels: GlobalModelResponse[] = []
     let skip = 0
+    let expectedTotal: number | null = null
 
     while (true) {
       const response = await listGlobalModels({
         skip,
         limit: GLOBAL_MODELS_FETCH_PAGE_SIZE,
       })
+      if (expectedTotal === null && typeof response.total === 'number') {
+        expectedTotal = response.total
+      }
       const pageModels = response.models || []
       allModels.push(...pageModels)
 
+      if (expectedTotal !== null && allModels.length >= expectedTotal) {
+        break
+      }
       if (pageModels.length < GLOBAL_MODELS_FETCH_PAGE_SIZE) {
         break
       }
@@ -1045,12 +1058,16 @@ async function loadGlobalModels() {
       skip += pageModels.length
     }
 
+    if (requestId !== globalModelsRequestId) return
     globalModels.value = allModels
   } catch (err: unknown) {
+    if (requestId !== globalModelsRequestId) return
     log.error('加载模型失败:', err)
     showError(parseApiError(err, '加载模型失败'), '加载模型失败')
   } finally {
-    loading.value = false
+    if (requestId === globalModelsRequestId) {
+      loading.value = false
+    }
   }
 }
 
@@ -1064,6 +1081,7 @@ function handleRowClick(event: MouseEvent, model: GlobalModelResponse) {
 }
 
 async function selectModel(model: GlobalModelResponse) {
+  const requestId = ++modelSelectionRequestId
   // 先显示缓存数据，提升响应速度
   selectedModel.value = model
   detailTab.value = 'basic'
@@ -1078,6 +1096,7 @@ async function selectModel(model: GlobalModelResponse) {
   ])
 
   // 更新为最新数据（如果获取成功）
+  if (requestId !== modelSelectionRequestId) return
   if (latestModel) {
     selectedModel.value = latestModel
   }
@@ -1096,10 +1115,12 @@ async function refreshSelectedModel() {
 
 // 加载指定模型的关联提供商
 async function loadModelProviders(_globalModelId: string) {
+  const requestId = ++modelProvidersRequestId
   loadingModelProviders.value = true
   try {
     // 使用新的 API 获取所有关联提供商（包括非活跃的）
     const response = await getGlobalModelProviders(_globalModelId)
+    if (requestId !== modelProvidersRequestId) return
 
     // 转换为展示格式
     selectedModelProviders.value = response.providers.map(p => ({
@@ -1124,11 +1145,14 @@ async function loadModelProviders(_globalModelId: string) {
       supports_streaming: p.supports_streaming
     }))
   } catch (err: unknown) {
+    if (requestId !== modelProvidersRequestId) return
     log.error('加载关联提供商失败:', err)
     showError(parseApiError(err, '加载关联提供商失败'), '错误')
     selectedModelProviders.value = []
   } finally {
-    loadingModelProviders.value = false
+    if (requestId === modelProvidersRequestId) {
+      loadingModelProviders.value = false
+    }
   }
 }
 
@@ -1137,14 +1161,25 @@ async function ensureProviderOptions() {
   if (providerOptions.value.length > 0 || loadingProviderOptions.value) {
     return
   }
+  if (providerOptionsRequest) {
+    await providerOptionsRequest
+    return
+  }
+  providerOptionsRequest = (async () => {
+    try {
+      loadingProviderOptions.value = true
+      providerOptions.value = await getProvidersSummary()
+    } catch (err: unknown) {
+      const message = parseApiError(err, '加载 Provider 列表失败')
+      showError(message, '错误')
+    } finally {
+      loadingProviderOptions.value = false
+    }
+  })()
   try {
-    loadingProviderOptions.value = true
+    await providerOptionsRequest
-    providerOptions.value = await getProvidersSummary()
-  } catch (err: unknown) {
-    const message = parseApiError(err, '加载 Provider 列表失败')
-    showError(message, '错误')
   } finally {
-    loadingProviderOptions.value = false
+    providerOptionsRequest = null
   }
 }
 
@@ -1320,6 +1355,8 @@ async function confirmBatchDeleteModels() {
 // 抽屉控制函数
 function handleDrawerOpenChange(value: boolean) {
   if (!value && !hasBlockingDialogOpen.value) {
+    modelSelectionRequestId += 1
+    modelProvidersRequestId += 1
     selectedModel.value = null
   }
 }
@@ -1455,9 +1492,13 @@ async function refreshData() {
 }
 
 async function loadProviders() {
+  const requestId = ++providersRequestId
   try {
-    providers.value = await getProvidersSummary()
+    const nextProviders = await getProvidersSummary()
+    if (requestId !== providersRequestId) return
+    providers.value = nextProviders
   } catch (err: unknown) {
+    if (requestId !== providersRequestId) return
     showError(parseApiError(err, '加载 Provider 列表失败'), '加载 Provider 列表失败')
   }
 }
@@ -1468,6 +1509,13 @@ onMounted(async () => {
     loadProviders(),
   ])
 })
+
+onBeforeUnmount(() => {
+  globalModelsRequestId += 1
+  modelSelectionRequestId += 1
+  modelProvidersRequestId += 1
+  providersRequestId += 1
+})
 </script>
 
 <style scoped>

frontend/src/views/admin/PerformanceAnalysis.vue

@@ -91,7 +91,7 @@
 </template>
 
 <script setup lang="ts">
-import { ref, computed, onMounted, watch } from 'vue'
+import { ref, computed, onMounted, onUnmounted, watch } from 'vue'
 import Card from '@/components/ui/card.vue'
 import { LoadingState, TimeRangePicker } from '@/components/common'
 import { ErrorDistributionChart, PercentileChart } from '@/components/stats'
@@ -112,6 +112,12 @@ const errorLoading = ref(false)
 
 const providerStatus = ref<ProviderStatus[]>([])
 const providerLoading = ref(false)
+let percentilesRequestId = 0
+let errorsRequestId = 0
+let providersRequestId = 0
+let loadAllPromise: Promise<void> | null = null
+let hasPendingLoadAll = false
+let loadAllDebounceTimer: ReturnType<typeof setTimeout> | null = null
 
 function buildTimeRangeParams() {
   return {
@@ -124,31 +130,45 @@ function buildTimeRangeParams() {
 }
 
 async function loadPercentiles() {
+  const requestId = ++percentilesRequestId
   percentileLoading.value = true
   try {
-    percentiles.value = await adminApi.getPercentiles(buildTimeRangeParams())
+    const data = await adminApi.getPercentiles(buildTimeRangeParams())
+    if (requestId !== percentilesRequestId) return
+    percentiles.value = data
   } finally {
-    percentileLoading.value = false
+    if (requestId === percentilesRequestId) {
+      percentileLoading.value = false
+    }
   }
 }
 
 async function loadErrors() {
+  const requestId = ++errorsRequestId
   errorLoading.value = true
   try {
     const response = await adminApi.getErrorDistribution(buildTimeRangeParams())
+    if (requestId !== errorsRequestId) return
     errorDistribution.value = response.distribution
     errorTrend.value = response.trend
   } finally {
-    errorLoading.value = false
+    if (requestId === errorsRequestId) {
+      errorLoading.value = false
+    }
   }
 }
 
 async function loadProviders() {
+  const requestId = ++providersRequestId
   providerLoading.value = true
   try {
-    providerStatus.value = await dashboardApi.getProviderStatus()
+    const data = await dashboardApi.getProviderStatus()
+    if (requestId !== providersRequestId) return
+    providerStatus.value = data
   } finally {
-    providerLoading.value = false
+    if (requestId === providersRequestId) {
+      providerLoading.value = false
+    }
   }
 }
 
@@ -166,10 +186,47 @@ const errorTrendChartData = computed(() => ({
 }))
 
 async function loadAll() {
-  await Promise.all([loadPercentiles(), loadErrors(), loadProviders()])
+  if (loadAllPromise) {
+    hasPendingLoadAll = true
+    return loadAllPromise
+  }
+  loadAllPromise = Promise.all([loadPercentiles(), loadErrors(), loadProviders()])
+    .then(() => undefined)
+    .finally(() => {
+      loadAllPromise = null
+      if (hasPendingLoadAll) {
+        hasPendingLoadAll = false
+        void loadAll()
+      }
+    })
+  return loadAllPromise
 }
 
-watch(timeRange, loadAll, { deep: true })
+function scheduleLoadAll() {
+  if (loadAllDebounceTimer) {
+    clearTimeout(loadAllDebounceTimer)
+  }
+  loadAllDebounceTimer = setTimeout(() => {
+    loadAllDebounceTimer = null
+    void loadAll()
+  }, 120)
+}
 
-onMounted(loadAll)
+watch(timeRange, scheduleLoadAll, { deep: true })
+
+onMounted(() => {
+  void loadAll()
+})
+
+onUnmounted(() => {
+  if (loadAllDebounceTimer) {
+    clearTimeout(loadAllDebounceTimer)
+    loadAllDebounceTimer = null
+  }
+  hasPendingLoadAll = false
+  loadAllPromise = null
+  percentilesRequestId += 1
+  errorsRequestId += 1
+  providersRequestId += 1
+})
 </script>

frontend/src/views/admin/PoolManagement.vue

@@ -1158,7 +1158,7 @@
 </template>
 
 <script setup lang="ts">
-import { ref, computed, watch, onMounted } from 'vue'
+import { ref, computed, watch, onMounted, onBeforeUnmount } from 'vue'
 import {
   Search,
   Upload,
@@ -1243,11 +1243,19 @@ const proxyNodesStore = useProxyNodesStore()
 // --- Overview ---
 const poolProviders = ref<PoolOverviewItem[]>([])
 const overviewLoading = ref(true)
+let overviewRequestId = 0
+let selectProviderRequestId = 0
+let providerDataRequestId = 0
+let keysRequestId = 0
+let keysSearchDebounceTimer: number | null = null
+let suppressFiltersWatch = false
 
 async function loadOverview() {
+  const requestId = ++overviewRequestId
   overviewLoading.value = true
   try {
     const res = await getPoolOverview()
+    if (requestId !== overviewRequestId) return
     const enabledProviders = res.items.filter(item => item.pool_enabled)
     poolProviders.value = enabledProviders
 
@@ -1266,9 +1274,12 @@ async function loadOverview() {
       }
     }
   } catch (err) {
+    if (requestId !== overviewRequestId) return
     showError(parseApiError(err))
   } finally {
-    overviewLoading.value = false
+    if (requestId === overviewRequestId) {
+      overviewLoading.value = false
+    }
   }
 }
 
@@ -1306,6 +1317,7 @@ const showAccountQuotaColumn = computed(() => {
 })
 
 async function selectProvider(id: string) {
+  const requestId = ++selectProviderRequestId
   selectedProviderId.value = id
   editingKeyDetail.value = null
   keyPermissionsDialogOpen.value = false
@@ -1315,16 +1327,27 @@ async function selectProvider(id: string) {
   proxyMobilePopoverOpenKeyId.value = null
   schedulingDetailDesktopPopoverOpenKeyId.value = null
   schedulingDetailMobilePopoverOpenKeyId.value = null
+  suppressFiltersWatch = true
   currentPage.value = 1
   searchQuery.value = ''
   statusFilter.value = 'all'
+  suppressFiltersWatch = false
+  if (keysSearchDebounceTimer !== null) {
+    clearTimeout(keysSearchDebounceTimer)
+    keysSearchDebounceTimer = null
+  }
   await Promise.all([loadKeys(), loadProviderData(id)])
+  if (requestId !== selectProviderRequestId) return
 }
 
 async function loadProviderData(id: string) {
+  const requestId = ++providerDataRequestId
   try {
-    selectedProviderData.value = await getProvider(id)
+    const providerData = await getProvider(id)
+    if (requestId !== providerDataRequestId || selectedProviderId.value !== id) return
+    selectedProviderData.value = providerData
   } catch {
+    if (requestId !== providerDataRequestId || selectedProviderId.value !== id) return
     selectedProviderData.value = null
   }
 }
@@ -1434,25 +1457,52 @@ async function refreshCurrentPage() {
 
 async function loadKeys() {
   if (!selectedProviderId.value) return
+  const requestId = ++keysRequestId
+  const providerId = selectedProviderId.value
+  const page = currentPage.value
+  const pageSizeValue = pageSize.value
+  const search = searchQuery.value || undefined
+  const status = statusFilter.value as 'all' | 'active' | 'cooldown' | 'inactive'
   keysLoading.value = true
   try {
-    keyPage.value = await listPoolKeys(selectedProviderId.value, {
+    const nextPage = await listPoolKeys(providerId, {
-      page: currentPage.value,
+      page,
-      page_size: pageSize.value,
+      page_size: pageSizeValue,
-      search: searchQuery.value || undefined,
+      search,
-      status: statusFilter.value as 'all' | 'active' | 'cooldown' | 'inactive',
+      status,
     })
+    if (requestId !== keysRequestId || selectedProviderId.value !== providerId) return
+    keyPage.value = nextPage
   } catch (err) {
+    if (requestId !== keysRequestId || selectedProviderId.value !== providerId) return
     showError(parseApiError(err))
   } finally {
-    keysLoading.value = false
+    if (requestId === keysRequestId) {
+      keysLoading.value = false
+    }
   }
 }
 
-watch([currentPage, pageSize], () => loadKeys())
+watch([currentPage, pageSize], () => {
-watch([searchQuery, statusFilter], () => {
+  void loadKeys()
+})
+
+watch(statusFilter, () => {
+  if (suppressFiltersWatch) return
   currentPage.value = 1
-  loadKeys()
+  void loadKeys()
+})
+
+watch(searchQuery, () => {
+  if (suppressFiltersWatch) return
+  currentPage.value = 1
+  if (keysSearchDebounceTimer !== null) {
+    clearTimeout(keysSearchDebounceTimer)
+  }
+  keysSearchDebounceTimer = window.setTimeout(() => {
+    keysSearchDebounceTimer = null
+    void loadKeys()
+  }, 300)
 })
 
 function normalizeAuthTypeForEdit(authType: string): EndpointAPIKey['auth_type'] {
@@ -2224,4 +2274,15 @@ onMounted(async () => {
   await loadOverview()
   void refreshCurrentPageQuotaInBackground({ silent: true })
 })
+
+onBeforeUnmount(() => {
+  if (keysSearchDebounceTimer !== null) {
+    clearTimeout(keysSearchDebounceTimer)
+    keysSearchDebounceTimer = null
+  }
+  overviewRequestId += 1
+  selectProviderRequestId += 1
+  providerDataRequestId += 1
+  keysRequestId += 1
+})
 </script>

frontend/src/views/admin/ProviderManagement.vue

@@ -226,6 +226,7 @@ const { confirmDanger } = useConfirm()
 // 状态
 const loading = ref(false)
 const providers = ref<ProviderWithEndpointsSummary[]>([])
+let providersRequestId = 0
 const providerDialogOpen = ref(false)
 const providerToEdit = ref<ProviderWithEndpointsSummary | null>(null)
 const priorityDialogOpen = ref(false)
@@ -350,15 +351,21 @@ async function loadGlobalModelList() {
 
 // 加载提供商列表
 async function loadProviders() {
+  const requestId = ++providersRequestId
   loading.value = true
   try {
-    providers.value = await getProvidersSummary()
+    const nextProviders = await getProvidersSummary()
+    if (requestId !== providersRequestId) return
+    providers.value = nextProviders
     // 异步加载配置了 ops 的 provider 的余额数据
     loadBalances(providers.value)
   } catch (err: unknown) {
+    if (requestId !== providersRequestId) return
     showError(parseApiError(err, '加载提供商列表失败'), '错误')
   } finally {
-    loading.value = false
+    if (requestId === providersRequestId) {
+      loading.value = false
+    }
   }
 }
 

frontend/src/views/admin/UserStats.vue

@@ -144,7 +144,7 @@
 </template>
 
 <script setup lang="ts">
-import { ref, computed, onMounted, watch } from 'vue'
+import { ref, computed, onMounted, onUnmounted, watch } from 'vue'
 import { Card, Select, SelectContent, SelectItem, SelectTrigger, SelectValue } from '@/components/ui'
 import LineChart from '@/components/charts/LineChart.vue'
 import { LoadingState, TimeRangePicker } from '@/components/common'
@@ -184,6 +184,15 @@ const summaryLoading = ref(false)
 const series = ref<TimeSeriesItem[]>([])
 const comparisonSeries = ref<TimeSeriesItem[]>([])
 const seriesLoading = ref(false)
+let leaderboardRequestId = 0
+let summaryRequestId = 0
+let seriesRequestId = 0
+let leaderboardLoadPromise: Promise<void> | null = null
+let hasPendingLeaderboardLoad = false
+let leaderboardDebounceTimer: ReturnType<typeof setTimeout> | null = null
+let userPanelsLoadPromise: Promise<void> | null = null
+let hasPendingUserPanelsLoad = false
+let userPanelsDebounceTimer: ReturnType<typeof setTimeout> | null = null
 
 function buildTimeRangeParams() {
   return {
@@ -204,6 +213,12 @@ async function loadUsers() {
 }
 
 async function loadLeaderboard() {
+  if (leaderboardLoadPromise) {
+    hasPendingLeaderboardLoad = true
+    return leaderboardLoadPromise
+  }
+  leaderboardLoadPromise = (async () => {
+  const requestId = ++leaderboardRequestId
   leaderboardLoading.value = true
   try {
     const response = await adminApi.getLeaderboardUsers({
@@ -211,46 +226,90 @@ async function loadLeaderboard() {
       metric: metric.value,
       limit: 10
     })
+    if (requestId !== leaderboardRequestId) return
     leaderboard.value = response.items
   } finally {
-    leaderboardLoading.value = false
+    if (requestId === leaderboardRequestId) {
+      leaderboardLoading.value = false
+    }
   }
+  })().finally(() => {
+    leaderboardLoadPromise = null
+    if (hasPendingLeaderboardLoad) {
+      hasPendingLeaderboardLoad = false
+      void loadLeaderboard()
+    }
+  })
+  return leaderboardLoadPromise
 }
 
 async function loadSummary() {
   if (!selectedUserId.value) return
+  const requestId = ++summaryRequestId
   summaryLoading.value = true
   try {
-    userSummary.value = await usageApi.getUsageStats({
+    const summary = await usageApi.getUsageStats({
       ...buildTimeRangeParams(),
       user_id: selectedUserId.value
     })
+    if (requestId !== summaryRequestId) return
+    userSummary.value = summary
   } finally {
-    summaryLoading.value = false
+    if (requestId === summaryRequestId) {
+      summaryLoading.value = false
+    }
   }
 }
 
 async function loadSeries() {
   if (!selectedUserId.value) return
+  const requestId = ++seriesRequestId
   seriesLoading.value = true
   try {
-    series.value = await adminApi.getTimeSeries({
+    const baseParams = {
       ...buildTimeRangeParams(),
       user_id: selectedUserId.value
-    })
+    }
-
+    const shouldCompare = Boolean(compareUserId.value && compareUserId.value !== '__none__')
-    comparisonSeries.value = []
+    const comparePromise: Promise<TimeSeriesItem[]> = shouldCompare
-    if (compareUserId.value && compareUserId.value !== '__none__') {
+      ? adminApi.getTimeSeries({
-      comparisonSeries.value = await adminApi.getTimeSeries({
         ...buildTimeRangeParams(),
         user_id: compareUserId.value
       })
-    }
+      : Promise.resolve([])
+
+    const [primarySeries, compareSeries] = await Promise.all([
+      adminApi.getTimeSeries(baseParams),
+      comparePromise
+    ])
+
+    if (requestId !== seriesRequestId) return
+    series.value = primarySeries
+    comparisonSeries.value = compareSeries
   } finally {
-    seriesLoading.value = false
+    if (requestId === seriesRequestId) {
+      seriesLoading.value = false
+    }
   }
 }
 
+async function loadUserPanels() {
+  if (userPanelsLoadPromise) {
+    hasPendingUserPanelsLoad = true
+    return userPanelsLoadPromise
+  }
+  userPanelsLoadPromise = Promise.all([loadSummary(), loadSeries()])
+    .then(() => undefined)
+    .finally(() => {
+      userPanelsLoadPromise = null
+      if (hasPendingUserPanelsLoad) {
+        hasPendingUserPanelsLoad = false
+        void loadUserPanels()
+      }
+    })
+  return userPanelsLoadPromise
+}
+
 const seriesChartData = computed(() => ({
   labels: series.value.map(item => item.date),
   datasets: [
@@ -284,16 +343,52 @@ const comparisonChartData = computed(() => ({
   ]
 }))
 
-watch([timeRange, metric], loadLeaderboard, { deep: true })
+function scheduleLeaderboardLoad() {
-watch([timeRange, selectedUserId, compareUserId], () => {
+  if (leaderboardDebounceTimer) {
-  loadSummary()
+    clearTimeout(leaderboardDebounceTimer)
-  loadSeries()
+  }
-}, { deep: true })
+  leaderboardDebounceTimer = setTimeout(() => {
+    leaderboardDebounceTimer = null
+    void loadLeaderboard()
+  }, 120)
+}
+
+function scheduleUserPanelsLoad() {
+  if (userPanelsDebounceTimer) {
+    clearTimeout(userPanelsDebounceTimer)
+  }
+  userPanelsDebounceTimer = setTimeout(() => {
+    userPanelsDebounceTimer = null
+    void loadUserPanels()
+  }, 120)
+}
+
+watch([timeRange, metric], scheduleLeaderboardLoad, { deep: true })
+watch([timeRange, selectedUserId, compareUserId], scheduleUserPanelsLoad, { deep: true })
 
 onMounted(async () => {
   await loadUsers()
-  await loadLeaderboard()
+  await Promise.all([
-  await loadSummary()
+    loadLeaderboard(),
-  await loadSeries()
+    loadUserPanels()
+  ])
+})
+
+onUnmounted(() => {
+  if (leaderboardDebounceTimer) {
+    clearTimeout(leaderboardDebounceTimer)
+    leaderboardDebounceTimer = null
+  }
+  if (userPanelsDebounceTimer) {
+    clearTimeout(userPanelsDebounceTimer)
+    userPanelsDebounceTimer = null
+  }
+  hasPendingLeaderboardLoad = false
+  hasPendingUserPanelsLoad = false
+  leaderboardLoadPromise = null
+  userPanelsLoadPromise = null
+  leaderboardRequestId += 1
+  summaryRequestId += 1
+  seriesRequestId += 1
 })
 </script>

frontend/src/views/admin/Users.vue

@@ -793,6 +793,7 @@ const apiKeyInput = ref<HTMLInputElement>()
 // 用户统计
 const userStats = ref<Record<string, UsageByUser>>({})
 const loadingStats = ref(false)
+let userStatsRequestId = 0
 
 const searchQuery = ref('')
 const filterRole = ref('all')
@@ -846,13 +847,17 @@ watch([searchQuery, filterRole, filterStatus], () => {
 })
 
 onMounted(async () => {
-  await usersStore.fetchUsers()
+  await Promise.all([
-  await loadUserStats()
+    usersStore.fetchUsers(),
+    loadUserStats()
+  ])
 })
 
 async function refreshUsers() {
-  await usersStore.fetchUsers()
+  await Promise.all([
-  await loadUserStats()
+    usersStore.fetchUsers(),
+    loadUserStats()
+  ])
 }
 
 function formatDate(dateString: string) {
@@ -860,9 +865,11 @@ function formatDate(dateString: string) {
 }
 
 async function loadUserStats() {
+  const requestId = ++userStatsRequestId
   loadingStats.value = true
   try {
     const data = await usageApi.getUsageByUser()
+    if (requestId !== userStatsRequestId) return
     userStats.value = data.reduce((acc: Record<string, UsageByUser>, stat: UsageByUser) => {
       acc[stat.user_id] = stat
       return acc
@@ -870,7 +877,9 @@ async function loadUserStats() {
   } catch (err) {
     log.error('加载用户统计失败:', err)
   } finally {
-    loadingStats.value = false
+    if (requestId === userStatsRequestId) {
+      loadingStats.value = false
+    }
   }
 }
 

frontend/src/views/admin/system-settings/CleanupPolicySection.vue

@@ -64,7 +64,7 @@
           id="compressed-log-retention-days"
           :model-value="compressedLogRetentionDays"
           type="number"
-          placeholder="90"
+          placeholder="30"
           class="mt-1"
           @update:model-value="$emit('update:compressedLogRetentionDays', Number($event))"
         />
@@ -98,7 +98,7 @@
           for="log-retention-days"
           class="block text-sm font-medium"
         >
-          完整记录保留天数
+          请求记录保存天数
         </Label>
         <Input
           id="log-retention-days"

frontend/src/views/admin/system-settings/composables/useSystemConfig.ts

@@ -111,7 +111,7 @@ function createDefaultConfig(): SystemConfig {
     // 请求记录清理
     enable_auto_cleanup: true,
     detail_log_retention_days: 7,
-    compressed_log_retention_days: 90,
+    compressed_log_retention_days: 30,
     header_retention_days: 90,
     log_retention_days: 365,
     cleanup_batch_size: 1000,
@@ -179,7 +179,7 @@ export function useSystemConfig() {
       systemConfig.value.max_request_body_size !== originalConfig.value.max_request_body_size ||
       systemConfig.value.max_response_body_size !== originalConfig.value.max_response_body_size ||
       JSON.stringify(systemConfig.value.sensitive_headers) !==
-        JSON.stringify(originalConfig.value.sensitive_headers)
+      JSON.stringify(originalConfig.value.sensitive_headers)
     )
   })
 
@@ -187,14 +187,14 @@ export function useSystemConfig() {
     if (!originalConfig.value) return false
     return (
       systemConfig.value.detail_log_retention_days !==
-        originalConfig.value.detail_log_retention_days ||
+      originalConfig.value.detail_log_retention_days ||
       systemConfig.value.compressed_log_retention_days !==
-        originalConfig.value.compressed_log_retention_days ||
+      originalConfig.value.compressed_log_retention_days ||
       systemConfig.value.header_retention_days !== originalConfig.value.header_retention_days ||
       systemConfig.value.log_retention_days !== originalConfig.value.log_retention_days ||
       systemConfig.value.cleanup_batch_size !== originalConfig.value.cleanup_batch_size ||
       systemConfig.value.audit_log_retention_days !==
-        originalConfig.value.audit_log_retention_days
+      originalConfig.value.audit_log_retention_days
     )
   })
 
@@ -231,7 +231,7 @@ export function useSystemConfig() {
         try {
           const response = await adminApi.getSystemConfig(key)
           if (response.value !== null && response.value !== undefined) {
-            ;(systemConfig.value as Record<string, unknown>)[key] = response.value
+            ; (systemConfig.value as Record<string, unknown>)[key] = response.value
           }
         } catch {
           // 配置不存在时使用默认值，无需处理

frontend/src/views/shared/Dashboard.vue

@@ -987,6 +987,10 @@ const dailyTimeRange = ref<DateRangeParams>(getDateRangeFromPeriod('last7days'))
 // 统计周期
 const loadingDaily = ref(false)
 const loading = ref(false)
+let dailyStatsRequestId = 0
+let dailyStatsLoadPromise: Promise<void> | null = null
+let hasPendingDailyStatsLoad = false
+let dailyStatsDebounceTimer: ReturnType<typeof setTimeout> | null = null
 
 
 // 公告
@@ -1279,7 +1283,8 @@ onMounted(async () => {
   }
   await Promise.all([
     loadDashboardData(),
-    loadAnnouncements()
+    loadAnnouncements(),
+    loadDailyStats()
   ])
   await nextTick()
   setupTimelineResizeObserver()
@@ -1298,6 +1303,13 @@ onBeforeUnmount(() => {
   statsPanelObserver = null
   announcementsTimelineObserver?.disconnect()
   announcementsTimelineObserver = null
+  if (dailyStatsDebounceTimer) {
+    clearTimeout(dailyStatsDebounceTimer)
+    dailyStatsDebounceTimer = null
+  }
+  hasPendingDailyStatsLoad = false
+  dailyStatsLoadPromise = null
+  dailyStatsRequestId += 1
 })
 
 async function loadDashboardData() {
@@ -1326,22 +1338,48 @@ async function loadDashboardData() {
 }
 
 async function loadDailyStats() {
-  loadingDaily.value = true
+  if (dailyStatsLoadPromise) {
-  try {
+    hasPendingDailyStatsLoad = true
-    const response = await dashboardApi.getDailyStats(dailyTimeRange.value)
+    return dailyStatsLoadPromise
-    dailyStats.value = response.daily_stats
-    providerSummary.value = response.provider_summary || []
-  } catch {
-    dailyStats.value = []
-    providerSummary.value = []
-  } finally {
-    loadingDaily.value = false
   }
+  const requestId = ++dailyStatsRequestId
+  loadingDaily.value = true
+  dailyStatsLoadPromise = (async () => {
+    try {
+      const response = await dashboardApi.getDailyStats(dailyTimeRange.value)
+      if (requestId !== dailyStatsRequestId) return
+      dailyStats.value = response.daily_stats
+      providerSummary.value = response.provider_summary || []
+    } catch {
+      if (requestId !== dailyStatsRequestId) return
+      dailyStats.value = []
+      providerSummary.value = []
+    } finally {
+      if (requestId === dailyStatsRequestId) {
+        loadingDaily.value = false
+      }
+    }
+  })().finally(() => {
+    dailyStatsLoadPromise = null
+    if (hasPendingDailyStatsLoad) {
+      hasPendingDailyStatsLoad = false
+      void loadDailyStats()
+    }
+  })
+  return dailyStatsLoadPromise
 }
 
-watch(dailyTimeRange, async () => {
+function scheduleDailyStatsLoad() {
-  await loadDailyStats()
+  if (dailyStatsDebounceTimer) {
-}, { deep: true })
+    clearTimeout(dailyStatsDebounceTimer)
+  }
+  dailyStatsDebounceTimer = setTimeout(() => {
+    dailyStatsDebounceTimer = null
+    void loadDailyStats()
+  }, 120)
+}
+
+watch(dailyTimeRange, scheduleDailyStatsLoad, { deep: true })
 
 function formatDate(dateString: string): string {
   const date = new Date(dateString)

frontend/src/views/shared/Usage.vue

@@ -247,14 +247,17 @@ const hasActiveRequests = computed(() => activeRequestIds.value.length > 0)
 // 自动刷新定时器
 let autoRefreshTimer: ReturnType<typeof setInterval> | null = null
 let globalAutoRefreshTimer: ReturnType<typeof setInterval> | null = null
+let refreshInFlight: Promise<void> | null = null
 const AUTO_REFRESH_INTERVAL = 1000 // 1秒刷新一次（用于活跃请求）
 const GLOBAL_AUTO_REFRESH_INTERVAL = 5000 // 5秒刷新一次（全局自动刷新）
 const globalAutoRefresh = ref(false) // 全局自动刷新开关
+const isPageVisible = ref(typeof document === 'undefined' ? true : !document.hidden)
 
 // 轮询活跃请求状态（轻量级，只更新状态变化的记录）
 
 let pollInFlight = false
 async function pollActiveRequests() {
+  if (!isPageVisible.value) return
   if (!hasActiveRequests.value) return
   if (pollInFlight) return
   pollInFlight = true
@@ -338,6 +341,7 @@ async function pollActiveRequests() {
 
 // 启动自动刷新
 function startAutoRefresh() {
+  if (!isPageVisible.value) return
   if (autoRefreshTimer) return
   autoRefreshTimer = setInterval(pollActiveRequests, AUTO_REFRESH_INTERVAL)
 }
@@ -353,7 +357,7 @@ function stopAutoRefresh() {
 // 监听活跃请求状态，自动启动/停止刷新
 // 1秒轮询始终用于活跃请求的实时更新，不受全局刷新影响
 watch(hasActiveRequests, (hasActive) => {
-  if (hasActive) {
+  if (hasActive && isPageVisible.value) {
     startAutoRefresh()
   } else {
     stopAutoRefresh()
@@ -362,6 +366,7 @@ watch(hasActiveRequests, (hasActive) => {
 
 // 启动全局自动刷新
 function startGlobalAutoRefresh() {
+  if (!isPageVisible.value) return
   if (globalAutoRefreshTimer) return
   globalAutoRefreshTimer = setInterval(refreshData, GLOBAL_AUTO_REFRESH_INTERVAL)
 }
@@ -378,15 +383,34 @@ function stopGlobalAutoRefresh() {
 function handleAutoRefreshChange(value: boolean) {
   globalAutoRefresh.value = value
   if (value) {
-    refreshData() // 立即刷新一次
+    if (isPageVisible.value) {
+      refreshData() // 立即刷新一次
+    }
     startGlobalAutoRefresh()
   } else {
     stopGlobalAutoRefresh()
   }
 }
 
+function handleVisibilityChange() {
+  isPageVisible.value = !document.hidden
+  if (!isPageVisible.value) {
+    stopAutoRefresh()
+    stopGlobalAutoRefresh()
+    return
+  }
+  if (hasActiveRequests.value) {
+    startAutoRefresh()
+  }
+  if (globalAutoRefresh.value) {
+    refreshData()
+    startGlobalAutoRefresh()
+  }
+}
+
 // 组件卸载时清理定时器
 onUnmounted(() => {
+  document.removeEventListener('visibilitychange', handleVisibilityChange)
   stopAutoRefresh()
   stopGlobalAutoRefresh()
 })
@@ -419,6 +443,8 @@ const selectedRequestId = ref<string | null>(null)
 
 // 初始化加载
 onMounted(async () => {
+  document.addEventListener('visibilitychange', handleVisibilityChange)
+
   // 所有数据源并行加载（stats/heatmap/records/users 之间没有数据依赖）
   const statsTask = loadStats(timeRange.value).catch(err => {
     log.error('加载统计数据失败:', err)
@@ -548,11 +574,28 @@ async function handleFilterStatusChange(value: string) {
 
 // 刷新数据
 async function refreshData() {
-  await loadStats(timeRange.value)
+  if (!isPageVisible.value) return
-  if (isAdminPage.value) {
+  if (refreshInFlight) return refreshInFlight
-    await loadRecords({ page: currentPage.value, pageSize: pageSize.value }, getCurrentFilters())
+
+  refreshInFlight = (async () => {
+    if (isAdminPage.value) {
+      // loadStats 会同步更新 currentDateRange，随后 loadRecords 复用同一时间范围
+      await Promise.all([
+        loadStats(timeRange.value),
+        loadRecords({ page: currentPage.value, pageSize: pageSize.value }, getCurrentFilters())
+      ])
+      return
+    }
+
+    await loadStats(timeRange.value)
+    // 用户页面：loadStats 已包含记录加载
+  })()
+
+  try {
+    await refreshInFlight
+  } finally {
+    refreshInFlight = null
   }
-  // 用户页面：loadStats 已包含记录加载
 }
 
 // 显示请求详情

src/api/admin/api_keys/routes.py

@@ -11,6 +11,7 @@ from typing import Any
 from zoneinfo import ZoneInfo
 
 from fastapi import APIRouter, Depends, HTTPException, Query, Request
+from sqlalchemy import func
 from sqlalchemy.orm import Session
 
 from src.api.base.admin_adapter import AdminApiAdapter
@@ -366,7 +367,7 @@ class AdminListStandaloneKeysAdapter(AdminApiAdapter):
         if self.is_active is not None:
             query = query.filter(ApiKey.is_active == self.is_active)
 
-        total = query.count()
+        total = int(query.with_entities(func.count(ApiKey.id)).scalar() or 0)
         api_keys = (
             query.order_by(ApiKey.created_at.desc()).offset(self.skip).limit(self.limit).all()
         )

src/api/admin/billing/routes.py

@@ -13,6 +13,7 @@ from typing import Any, Literal
 
 from fastapi import APIRouter, Depends, Query, Request
 from pydantic import BaseModel, Field
+from sqlalchemy import func
 from sqlalchemy.exc import IntegrityError
 from sqlalchemy.orm import Session
 
@@ -251,7 +252,7 @@ class BillingRuleListAdapter(AdminApiAdapter):
         if self.is_enabled is not None:
             q = q.filter(BillingRule.is_enabled == self.is_enabled)
 
-        total = q.count()
+        total = int(q.with_entities(func.count(BillingRule.id)).scalar() or 0)
         items = (
             q.order_by(BillingRule.updated_at.desc())
             .offset((self.page - 1) * self.page_size)
@@ -365,7 +366,7 @@ class DimensionCollectorListAdapter(AdminApiAdapter):
         if self.is_enabled is not None:
             q = q.filter(DimensionCollector.is_enabled == self.is_enabled)
 
-        total = q.count()
+        total = int(q.with_entities(func.count(DimensionCollector.id)).scalar() or 0)
         items = (
             q.order_by(DimensionCollector.updated_at.desc())
             .offset((self.page - 1) * self.page_size)

src/api/admin/gemini_files.py

@@ -21,7 +21,7 @@ from typing import Any
 from fastapi import APIRouter, Depends, File, HTTPException, Query, UploadFile
 from pydantic import BaseModel
 from sqlalchemy import delete, func
-from sqlalchemy.orm import Session
+from sqlalchemy.orm import Session, load_only
 
 from src.clients.http_client import HTTPClientPool
 from src.core.crypto import crypto_service
@@ -106,26 +106,45 @@ async def list_file_mappings(
     now = datetime.now(timezone.utc)
 
     query = db.query(GeminiFileMapping)
+    count_query = db.query(func.count(GeminiFileMapping.id))
 
     # 过滤过期
     if not include_expired:
-        query = query.filter(GeminiFileMapping.expires_at > now)
+        active_filter = GeminiFileMapping.expires_at > now
+        query = query.filter(active_filter)
+        count_query = count_query.filter(active_filter)
 
     # 搜索
     if search:
         search_pattern = f"%{search}%"
-        query = query.filter(
+        search_filter = (GeminiFileMapping.file_name.ilike(search_pattern)) | (
-            (GeminiFileMapping.file_name.ilike(search_pattern))
+            GeminiFileMapping.display_name.ilike(search_pattern)
-            | (GeminiFileMapping.display_name.ilike(search_pattern))
         )
+        query = query.filter(search_filter)
+        count_query = count_query.filter(search_filter)
 
     # 总数
-    total = query.count()
+    total = int(count_query.scalar() or 0)
 
     # 分页
     offset = (page - 1) * page_size
     mappings = (
-        query.order_by(GeminiFileMapping.created_at.desc()).offset(offset).limit(page_size).all()
+        query.options(
+            load_only(
+                GeminiFileMapping.id,
+                GeminiFileMapping.file_name,
+                GeminiFileMapping.key_id,
+                GeminiFileMapping.user_id,
+                GeminiFileMapping.display_name,
+                GeminiFileMapping.mime_type,
+                GeminiFileMapping.created_at,
+                GeminiFileMapping.expires_at,
+            )
+        )
+        .order_by(GeminiFileMapping.created_at.desc())
+        .offset(offset)
+        .limit(page_size)
+        .all()
     )
 
     # 获取关联的 Key 和 User 信息
@@ -134,12 +153,22 @@ async def list_file_mappings(
 
     keys_map = {}
     if key_ids:
-        keys = db.query(ProviderAPIKey).filter(ProviderAPIKey.id.in_(key_ids)).all()
+        keys = (
+            db.query(ProviderAPIKey)
+            .options(load_only(ProviderAPIKey.id, ProviderAPIKey.name))
+            .filter(ProviderAPIKey.id.in_(key_ids))
+            .all()
+        )
         keys_map = {str(k.id): k.name for k in keys}
 
     users_map = {}
     if user_ids:
-        users = db.query(User).filter(User.id.in_(user_ids)).all()
+        users = (
+            db.query(User)
+            .options(load_only(User.id, User.username))
+            .filter(User.id.in_(user_ids))
+            .all()
+        )
         users_map = {str(u.id): u.username for u in users}
 
     items = []
@@ -199,9 +228,11 @@ async def get_file_mapping_stats(
     by_mime_type = {(mt or "unknown"): count for mt, count in mime_stats}
 
     # 有 gemini_files 能力的 Key 数量
-    keys = db.query(ProviderAPIKey).filter(ProviderAPIKey.is_active.is_(True)).all()
+    keys = db.query(ProviderAPIKey.capabilities).filter(ProviderAPIKey.is_active.is_(True)).all()
     capable_keys_count = sum(
-        1 for key in keys if key.capabilities and key.capabilities.get("gemini_files", False)
+        1
+        for (capabilities,) in keys
+        if isinstance(capabilities, dict) and capabilities.get("gemini_files", False)
     )
 
     return FileMappingStatsResponse(
@@ -294,15 +325,28 @@ async def list_capable_keys(
     """获取所有具有 gemini_files 能力的 Key 列表"""
     from src.models.database import Provider
 
-    keys = db.query(ProviderAPIKey).filter(ProviderAPIKey.is_active.is_(True)).all()
+    key_rows = (
+        db.query(
+            ProviderAPIKey.id,
+            ProviderAPIKey.name,
+            ProviderAPIKey.provider_id,
+            ProviderAPIKey.capabilities,
+        )
+        .filter(ProviderAPIKey.is_active.is_(True))
+        .all()
+    )
     capable_keys = [
-        key for key in keys if key.capabilities and key.capabilities.get("gemini_files", False)
+        key
+        for key in key_rows
+        if isinstance(key.capabilities, dict) and key.capabilities.get("gemini_files", False)
     ]
 
     # 获取 Provider 名称
-    provider_ids = {key.provider_id for key in capable_keys}
+    provider_ids = {key.provider_id for key in capable_keys if key.provider_id}
-    providers = db.query(Provider).filter(Provider.id.in_(provider_ids)).all()
+    provider_map: dict[str, str] = {}
-    provider_map = {str(p.id): p.name for p in providers}
+    if provider_ids:
+        providers = db.query(Provider.id, Provider.name).filter(Provider.id.in_(provider_ids)).all()
+        provider_map = {str(provider_id): provider_name for provider_id, provider_name in providers}
 
     return [
         CapableKeyResponse(
@@ -454,6 +498,15 @@ async def upload_file(
     with create_session() as db:
         keys = (
             db.query(ProviderAPIKey)
+            .options(
+                load_only(
+                    ProviderAPIKey.id,
+                    ProviderAPIKey.name,
+                    ProviderAPIKey.api_key,
+                    ProviderAPIKey.capabilities,
+                    ProviderAPIKey.is_active,
+                )
+            )
             .filter(
                 ProviderAPIKey.id.in_(key_id_list),
                 ProviderAPIKey.is_active.is_(True),
@@ -483,6 +536,7 @@ async def upload_file(
             now = datetime.now(timezone.utc)
             existing = (
                 db.query(GeminiFileMapping)
+                .options(load_only(GeminiFileMapping.key_id, GeminiFileMapping.file_name))
                 .filter(
                     GeminiFileMapping.source_hash == source_hash,
                     GeminiFileMapping.key_id.in_(capable_key_ids),

src/api/admin/management_tokens/routes.py

@@ -7,6 +7,7 @@ from typing import Any
 
 from fastapi import APIRouter, Depends, HTTPException, Query, Request
 from fastapi.responses import JSONResponse
+from sqlalchemy import func
 from sqlalchemy.orm import Session
 
 from src.api.base.admin_adapter import AdminApiAdapter
@@ -190,7 +191,7 @@ class AdminListManagementTokensAdapter(AdminManagementTokenApiAdapter):
         if self.is_active is not None:
             query = query.filter(ManagementToken.is_active == self.is_active)
 
-        total = query.count()
+        total = int(query.with_entities(func.count(ManagementToken.id)).scalar() or 0)
         tokens = (
             query.order_by(ManagementToken.created_at.desc())
             .offset(self.skip)

src/api/admin/models/global_models.py

@@ -276,17 +276,24 @@ class AdminListGlobalModelsAdapter(AdminApiAdapter):
     search: str | None
 
     async def handle(self, context: ApiRequestContext) -> Any:  # type: ignore[override]
-        from sqlalchemy import and_, case, func
+        from sqlalchemy import and_, case, func, or_
 
-        from src.models.database import Model, Provider
+        from src.models.database import GlobalModel, Model, Provider
 
-        models = GlobalModelService.list_global_models(
+        query = context.db.query(GlobalModel)
-            db=context.db,
+        if self.is_active is not None:
-            skip=self.skip,
+            query = query.filter(GlobalModel.is_active == self.is_active)
-            limit=self.limit,
+        if self.search:
-            is_active=self.is_active,
+            search_pattern = f"%{self.search}%"
-            search=self.search,
+            query = query.filter(
-        )
+                or_(
+                    GlobalModel.name.ilike(search_pattern),
+                    GlobalModel.display_name.ilike(search_pattern),
+                )
+            )
+
+        total = int(query.with_entities(func.count(GlobalModel.id)).scalar() or 0)
+        models = query.order_by(GlobalModel.name).offset(self.skip).limit(self.limit).all()
 
         # 一次性查询所有 GlobalModel 的 provider_count（优化 N+1 问题）
         # 用条件聚合同时获取总数和活跃数，减少一次 DB 往返
@@ -331,7 +338,7 @@ class AdminListGlobalModelsAdapter(AdminApiAdapter):
 
         return GlobalModelListResponse(
             models=model_responses,
-            total=len(models),
+            total=total,
         )
 
 
@@ -349,10 +356,9 @@ class AdminGetGlobalModelAdapter(AdminApiAdapter):
         global_model = GlobalModelService.get_global_model(context.db, self.global_model_id)
         stats = GlobalModelService.get_global_model_stats(context.db, self.global_model_id)
 
-        # 查询 provider_count 和 active_provider_count（与列表 API 一致）
+        # total_providers 已由 stats 提供，这里只查询活跃 provider 数量
-        count_row = (
+        active_count = (
             context.db.query(
-                func.count(func.distinct(Model.provider_id)),
                 func.count(
                     func.distinct(
                         case(
@@ -366,17 +372,17 @@ class AdminGetGlobalModelAdapter(AdminApiAdapter):
                             else_=None,
                         )
                     )
-                ),
+                )
             )
             .join(Provider, Model.provider_id == Provider.id)
             .filter(Model.global_model_id == global_model.id)
-            .first()
+            .scalar()
+            or 0
         )
-        total_count, active_count = count_row if count_row else (0, 0)
 
         response = GlobalModelResponse.model_validate(global_model)
-        response.provider_count = total_count
+        response.provider_count = stats["total_providers"]
-        response.active_provider_count = active_count
+        response.active_provider_count = int(active_count)
 
         return GlobalModelWithStats(
             **response.model_dump(),

src/api/admin/monitoring/audit.py

@@ -7,13 +7,14 @@ from datetime import datetime, timedelta, timezone
 from typing import Any
 
 from fastapi import APIRouter, Depends, HTTPException, Query, Request
-from sqlalchemy import func
+from sqlalchemy import case, func
 from sqlalchemy.orm import Session
 
 from src.api.base.admin_adapter import AdminApiAdapter
 from src.api.base.context import ApiRequestContext
-from src.api.base.pagination import PaginationMeta, build_pagination_payload, paginate_query
+from src.api.base.pagination import PaginationMeta, build_pagination_payload
 from src.api.base.pipeline import ApiRequestPipeline
+from src.config.constants import CacheTTL
 from src.core.logger import logger
 from src.database import get_db
 from src.models.database import (
@@ -26,6 +27,7 @@ from src.models.database import (
 from src.models.database import User as DBUser
 from src.services.health.monitor import HealthMonitor
 from src.services.system.audit import audit_service
+from src.utils.cache_decorator import cache_result
 from src.utils.database_helpers import escape_like_pattern
 
 router = APIRouter(prefix="/api/admin/monitoring", tags=["Admin - Monitoring"])
@@ -223,10 +225,26 @@ class AdminGetAuditLogsAdapter(AdminApiAdapter):
     # 查看审计日志本身不应该产生审计记录，避免刷新页面时产生大量无意义的日志
     audit_log_enabled: bool = False
 
+    @cache_result(
+        key_prefix="admin:monitoring:audit-logs",
+        ttl=CacheTTL.ADMIN_USAGE_RECORDS,
+        user_specific=False,
+        vary_by=["username", "event_type", "days", "limit", "offset"],
+    )
     async def handle(self, context: ApiRequestContext) -> Any:  # type: ignore[override]
         db = context.db
         cutoff_time = datetime.now(timezone.utc) - timedelta(days=self.days)
 
+        count_query = db.query(func.count(AuditLog.id)).filter(AuditLog.created_at >= cutoff_time)
+        if self.username:
+            escaped = escape_like_pattern(self.username)
+            count_query = count_query.outerjoin(DBUser, AuditLog.user_id == DBUser.id).filter(
+                DBUser.username.ilike(f"%{escaped}%", escape="\\")
+            )
+        if self.event_type:
+            count_query = count_query.filter(AuditLog.event_type == self.event_type)
+        total = int(count_query.scalar() or 0)
+
         base_query = (
             db.query(AuditLog, DBUser)
             .outerjoin(DBUser, AuditLog.user_id == DBUser.id)
@@ -238,8 +256,12 @@ class AdminGetAuditLogsAdapter(AdminApiAdapter):
         if self.event_type:
             base_query = base_query.filter(AuditLog.event_type == self.event_type)
 
-        ordered_query = base_query.order_by(AuditLog.created_at.desc())
+        logs_with_users = (
-        total, logs_with_users = paginate_query(ordered_query, self.limit, self.offset)
+            base_query.order_by(AuditLog.created_at.desc())
+            .offset(self.offset)
+            .limit(self.limit)
+            .all()
+        )
 
         items = [
             {
@@ -287,39 +309,51 @@ class AdminGetAuditLogsAdapter(AdminApiAdapter):
 
 
 class AdminSystemStatusAdapter(AdminApiAdapter):
+    @cache_result(
+        key_prefix="admin:monitoring:system-status",
+        ttl=CacheTTL.ADMIN_USAGE_RECORDS,
+        user_specific=False,
+    )
     async def handle(self, context: ApiRequestContext) -> Any:  # type: ignore[override]
         db = context.db
 
-        total_users = db.query(func.count(DBUser.id)).scalar()
+        user_stats = db.query(
-        active_users = db.query(func.count(DBUser.id)).filter(DBUser.is_active.is_(True)).scalar()
+            func.count(DBUser.id).label("total"),
+            func.sum(case((DBUser.is_active.is_(True), 1), else_=0)).label("active"),
+        ).first()
+        total_users = int((user_stats.total if user_stats else 0) or 0)
+        active_users = int((user_stats.active if user_stats else 0) or 0)
 
-        total_providers = db.query(func.count(Provider.id)).scalar()
+        provider_stats = db.query(
-        active_providers = (
+            func.count(Provider.id).label("total"),
-            db.query(func.count(Provider.id)).filter(Provider.is_active.is_(True)).scalar()
+            func.sum(case((Provider.is_active.is_(True), 1), else_=0)).label("active"),
-        )
+        ).first()
+        total_providers = int((provider_stats.total if provider_stats else 0) or 0)
+        active_providers = int((provider_stats.active if provider_stats else 0) or 0)
 
-        total_api_keys = db.query(func.count(ApiKey.id)).scalar()
+        api_key_stats = db.query(
-        active_api_keys = (
+            func.count(ApiKey.id).label("total"),
-            db.query(func.count(ApiKey.id)).filter(ApiKey.is_active.is_(True)).scalar()
+            func.sum(case((ApiKey.is_active.is_(True), 1), else_=0)).label("active"),
-        )
+        ).first()
+        total_api_keys = int((api_key_stats.total if api_key_stats else 0) or 0)
+        active_api_keys = int((api_key_stats.active if api_key_stats else 0) or 0)
 
         today_start = datetime.now(timezone.utc).replace(hour=0, minute=0, second=0, microsecond=0)
-        today_requests = (
+        today_stats = (
-            db.query(func.count(Usage.id)).filter(Usage.created_at >= today_start).scalar()
+            db.query(
-        )
+                func.count(Usage.id).label("requests"),
-        today_tokens = (
+                func.coalesce(func.sum(Usage.total_tokens), 0).label("tokens"),
-            db.query(func.sum(Usage.total_tokens)).filter(Usage.created_at >= today_start).scalar()
+                func.coalesce(func.sum(Usage.total_cost_usd), 0.0).label("cost"),
-            or 0
+            )
-        )
-        today_cost = (
-            db.query(func.sum(Usage.total_cost_usd))
             .filter(Usage.created_at >= today_start)
-            .scalar()
+            .first()
-            or 0
         )
+        today_requests = int((today_stats.requests if today_stats else 0) or 0)
+        today_tokens = int((today_stats.tokens if today_stats else 0) or 0)
+        today_cost = float((today_stats.cost if today_stats else 0.0) or 0.0)
 
         recent_errors = (
-            db.query(AuditLog)
+            db.query(func.count(AuditLog.id))
             .filter(
                 AuditLog.event_type.in_(
                     [
@@ -329,20 +363,21 @@ class AdminSystemStatusAdapter(AdminApiAdapter):
                 ),
                 AuditLog.created_at >= datetime.now(timezone.utc) - timedelta(hours=1),
             )
-            .count()
+            .scalar()
+            or 0
         )
 
         context.add_audit_metadata(
             action="system_status_snapshot",
-            total_users=int(total_users or 0),
+            total_users=total_users,
-            active_users=int(active_users or 0),
+            active_users=active_users,
-            total_providers=int(total_providers or 0),
+            total_providers=total_providers,
-            active_providers=int(active_providers or 0),
+            active_providers=active_providers,
-            total_api_keys=int(total_api_keys or 0),
+            total_api_keys=total_api_keys,
-            active_api_keys=int(active_api_keys or 0),
+            active_api_keys=active_api_keys,
-            today_requests=int(today_requests or 0),
+            today_requests=today_requests,
-            today_tokens=int(today_tokens or 0),
+            today_tokens=today_tokens,
-            today_cost=float(today_cost or 0.0),
+            today_cost=today_cost,
             recent_errors=int(recent_errors or 0),
         )
 

src/api/admin/pool/routes.py

@@ -617,13 +617,68 @@ class AdminPoolOverviewAdapter(AdminApiAdapter):
             .all()
         )
 
+        # 先批量计算池化 Provider 的 Key 总数/启用数，避免每个 Provider 单独查询（N+1）。
+        pool_provider_ids: list[str] = []
+        pool_enabled_map: dict[str, bool] = {}
+        for p in providers:
+            pid = str(p.id)
+            enabled = parse_pool_config(getattr(p, "config", None)) is not None
+            pool_enabled_map[pid] = enabled
+            if enabled:
+                pool_provider_ids.append(pid)
+
+        key_ids_by_provider: dict[str, list[str]] = {pid: [] for pid in pool_provider_ids}
+        key_stats_by_provider: dict[str, dict[str, int]] = {
+            pid: {"total": 0, "active": 0} for pid in pool_provider_ids
+        }
+        if pool_provider_ids:
+            key_rows = (
+                db.query(
+                    ProviderAPIKey.provider_id,
+                    ProviderAPIKey.id,
+                    ProviderAPIKey.is_active,
+                )
+                .filter(ProviderAPIKey.provider_id.in_(pool_provider_ids))
+                .all()
+            )
+            for provider_id, key_id, is_active in key_rows:
+                pid = str(provider_id)
+                kid = str(key_id)
+                key_ids_by_provider.setdefault(pid, []).append(kid)
+                stats = key_stats_by_provider.setdefault(pid, {"total": 0, "active": 0})
+                stats["total"] += 1
+                if is_active:
+                    stats["active"] += 1
+
+        # Redis 冷却状态并发获取，避免逐 Provider 串行等待。
+        cooldown_count_by_provider: dict[str, int] = {}
+        cooldown_targets = [
+            (pid, key_ids) for pid, key_ids in key_ids_by_provider.items() if key_ids
+        ]
+        if cooldown_targets:
+            cooldown_results = await asyncio.gather(
+                *[
+                    pool_redis.batch_get_cooldowns(pid, key_ids)
+                    for pid, key_ids in cooldown_targets
+                ],
+                return_exceptions=True,
+            )
+            for (pid, _key_ids), result in zip(cooldown_targets, cooldown_results, strict=False):
+                if isinstance(result, Exception):
+                    logger.warning(
+                        "池管理概览读取冷却状态失败",
+                        extra={"provider_id": pid, "error": str(result)},
+                    )
+                    cooldown_count_by_provider[pid] = 0
+                else:
+                    cooldown_count_by_provider[pid] = sum(
+                        1 for value in result.values() if value is not None
+                    )
+
         items: list[PoolOverviewItem] = []
         for p in providers:
             pid = str(p.id)
-            pcfg = parse_pool_config(getattr(p, "config", None))
+            if not pool_enabled_map.get(pid, False):
-
-            # Non-pool providers: skip Redis + key queries entirely.
-            if pcfg is None:
                 items.append(
                     PoolOverviewItem(
                         provider_id=pid,
@@ -634,22 +689,16 @@ class AdminPoolOverviewAdapter(AdminApiAdapter):
                 )
                 continue
 
-            keys = db.query(ProviderAPIKey).filter(ProviderAPIKey.provider_id == pid).all()
+            key_stats = key_stats_by_provider.get(pid, {"total": 0, "active": 0})
-            key_ids = [str(k.id) for k in keys]
-
-            cooldown_count = 0
-            if key_ids:
-                cooldowns = await pool_redis.batch_get_cooldowns(pid, key_ids)
-                cooldown_count = sum(1 for v in cooldowns.values() if v is not None)
 
             items.append(
                 PoolOverviewItem(
                     provider_id=pid,
                     provider_name=p.name,
                     provider_type=str(getattr(p, "provider_type", "custom") or "custom"),
-                    total_keys=len(keys),
+                    total_keys=key_stats["total"],
-                    active_keys=sum(1 for k in keys if k.is_active),
+                    active_keys=key_stats["active"],
-                    cooldown_count=cooldown_count,
+                    cooldown_count=cooldown_count_by_provider.get(pid, 0),
                     pool_enabled=True,
                 )
             )
@@ -688,7 +737,7 @@ class AdminListPoolKeysAdapter(AdminApiAdapter):
             q = q.filter(ProviderAPIKey.is_active.is_(False))
         # "cooldown" filtering is done post-query (Redis state)
 
-        total = q.count()
+        total = 0
 
         # For cooldown filtering we need to fetch all, then filter, then paginate.
         # Limit scan range to avoid loading the entire table into memory.
@@ -702,6 +751,7 @@ class AdminListPoolKeysAdapter(AdminApiAdapter):
             offset = (self.page - 1) * self.page_size
             keys = all_keys[offset : offset + self.page_size]
         else:
+            total = int(q.with_entities(func.count(ProviderAPIKey.id)).scalar() or 0)
             offset = (self.page - 1) * self.page_size
             keys = (
                 q.order_by(ProviderAPIKey.created_at.desc())

src/api/admin/providers/routes.py

@@ -9,12 +9,14 @@ from typing import Any
 
 from fastapi import APIRouter, Depends, Query, Request
 from pydantic import BaseModel, ConfigDict, Field, ValidationError
-from sqlalchemy.orm import Session
+from sqlalchemy import func
+from sqlalchemy.orm import Session, load_only
 
 from src.api.base.admin_adapter import AdminApiAdapter
 from src.api.base.context import ApiRequestContext
 from src.api.base.models_service import invalidate_models_list_cache
 from src.api.base.pipeline import ApiRequestPipeline
+from src.config.constants import CacheTTL
 from src.core.enums import ProviderBillingType
 from src.core.exceptions import InvalidRequestException, NotFoundException
 from src.core.logger import logger
@@ -27,6 +29,7 @@ from src.models.database import GlobalModel, Provider, ProviderAPIKey, ProviderE
 from src.models.endpoint_models import ProviderWithEndpointsSummary
 from src.services.cache.model_cache import ModelCacheService
 from src.services.cache.provider_cache import ProviderCacheService
+from src.utils.cache_decorator import cache_result
 
 from .summary import _build_provider_summary
 
@@ -341,9 +344,24 @@ class AdminListProvidersAdapter(AdminApiAdapter):
         self.limit = limit
         self.is_active = is_active
 
+    @cache_result(
+        key_prefix="admin:providers:list",
+        ttl=CacheTTL.ADMIN_USAGE_RECORDS,
+        user_specific=False,
+        vary_by=["skip", "limit", "is_active"],
+    )
     async def handle(self, context: ApiRequestContext) -> Any:  # type: ignore[override]
         db = context.db
-        query = db.query(Provider)
+        query = db.query(Provider).options(
+            load_only(
+                Provider.id,
+                Provider.name,
+                Provider.provider_priority,
+                Provider.is_active,
+                Provider.created_at,
+                Provider.updated_at,
+            )
+        )
         if self.is_active is not None:
             query = query.filter(Provider.is_active == self.is_active)
         providers = query.offset(self.skip).limit(self.limit).all()
@@ -723,11 +741,22 @@ class AdminGetProviderMappingPreviewAdapter(AdminApiAdapter):
     def __init__(self, provider_id: str):
         self.provider_id = provider_id
 
-    async def handle(self, context: ApiRequestContext) -> ProviderMappingPreviewResponse:  # type: ignore[override]
+    @cache_result(
+        key_prefix="admin:providers:mapping-preview",
+        ttl=CacheTTL.ADMIN_USAGE_RECORDS,
+        user_specific=False,
+        vary_by=["provider_id"],
+    )
+    async def handle(self, context: ApiRequestContext) -> Any:  # type: ignore[override]
         db = context.db
 
         # 获取 Provider
-        provider = db.query(Provider).filter(Provider.id == self.provider_id).first()
+        provider = (
+            db.query(Provider)
+            .options(load_only(Provider.id, Provider.name))
+            .filter(Provider.id == self.provider_id)
+            .first()
+        )
         if not provider:
             raise NotFoundException("提供商不存在", "provider")
 
@@ -735,19 +764,6 @@ class AdminGetProviderMappingPreviewAdapter(AdminApiAdapter):
         truncated_keys = 0
         truncated_models = 0
 
-        # 获取该 Provider 有白名单配置的 Key 总数（用于截断统计）
-        from sqlalchemy import func
-
-        total_keys_with_allowed_models = (
-            db.query(func.count(ProviderAPIKey.id))
-            .filter(
-                ProviderAPIKey.provider_id == self.provider_id,
-                ProviderAPIKey.allowed_models.isnot(None),
-            )
-            .scalar()
-            or 0
-        )
-
         # 获取该 Provider 有白名单配置的 Key（只查询需要的字段）
         keys = (
             db.query(
@@ -761,25 +777,22 @@ class AdminGetProviderMappingPreviewAdapter(AdminApiAdapter):
                 ProviderAPIKey.provider_id == self.provider_id,
                 ProviderAPIKey.allowed_models.isnot(None),
             )
-            .limit(MAPPING_PREVIEW_MAX_KEYS)
+            .limit(MAPPING_PREVIEW_MAX_KEYS + 1)
             .all()
         )
 
-        # 计算被截断的 Key 数量
+        if len(keys) > MAPPING_PREVIEW_MAX_KEYS:
-        if total_keys_with_allowed_models > MAPPING_PREVIEW_MAX_KEYS:
+            keys = keys[:MAPPING_PREVIEW_MAX_KEYS]
-            truncated_keys = total_keys_with_allowed_models - MAPPING_PREVIEW_MAX_KEYS
+            total_keys_with_allowed_models = (
-
+                db.query(func.count(ProviderAPIKey.id))
-        # 获取有 model_mappings 配置的 GlobalModel 总数（用于截断统计）
+                .filter(
-        total_models_with_mappings = (
+                    ProviderAPIKey.provider_id == self.provider_id,
-            db.query(func.count(GlobalModel.id))
+                    ProviderAPIKey.allowed_models.isnot(None),
-            .filter(
+                )
-                GlobalModel.config.isnot(None),
+                .scalar()
-                GlobalModel.config["model_mappings"].isnot(None),
+                or 0
-                func.jsonb_array_length(GlobalModel.config["model_mappings"]) > 0,
             )
-            .scalar()
+            truncated_keys = total_keys_with_allowed_models - MAPPING_PREVIEW_MAX_KEYS
-            or 0
-        )
 
         # 只查询有 model_mappings 配置的 GlobalModel（使用 SQLAlchemy JSONB 操作符）
         global_models = (
@@ -795,12 +808,22 @@ class AdminGetProviderMappingPreviewAdapter(AdminApiAdapter):
                 GlobalModel.config["model_mappings"].isnot(None),
                 func.jsonb_array_length(GlobalModel.config["model_mappings"]) > 0,
             )
-            .limit(MAPPING_PREVIEW_MAX_MODELS)
+            .limit(MAPPING_PREVIEW_MAX_MODELS + 1)
             .all()
         )
 
-        # 计算被截断的 GlobalModel 数量
+        if len(global_models) > MAPPING_PREVIEW_MAX_MODELS:
-        if total_models_with_mappings > MAPPING_PREVIEW_MAX_MODELS:
+            global_models = global_models[:MAPPING_PREVIEW_MAX_MODELS]
+            total_models_with_mappings = (
+                db.query(func.count(GlobalModel.id))
+                .filter(
+                    GlobalModel.config.isnot(None),
+                    GlobalModel.config["model_mappings"].isnot(None),
+                    func.jsonb_array_length(GlobalModel.config["model_mappings"]) > 0,
+                )
+                .scalar()
+                or 0
+            )
             truncated_models = total_models_with_mappings - MAPPING_PREVIEW_MAX_MODELS
 
         # 构建有映射配置的 GlobalModel 映射
@@ -819,10 +842,10 @@ class AdminGetProviderMappingPreviewAdapter(AdminApiAdapter):
                 keys=[],
                 total_keys=0,
                 total_matches=0,
-                truncated=False,
+                truncated=truncated_keys > 0 or truncated_models > 0,
-                truncated_keys=0,
+                truncated_keys=truncated_keys,
-                truncated_models=0,
+                truncated_models=truncated_models,
-            )
+            ).model_dump()
 
         key_infos: list[MappingMatchingKey] = []
         total_matches = 0
@@ -837,18 +860,6 @@ class AdminGetProviderMappingPreviewAdapter(AdminApiAdapter):
             if not allowed_models_list:
                 continue
 
-            # 生成脱敏 Key
-            masked_key = "***"
-            if key.api_key:
-                try:
-                    decrypted_key = crypto.decrypt(key.api_key, silent=True)
-                    if len(decrypted_key) > 8:
-                        masked_key = f"{decrypted_key[:4]}***{decrypted_key[-4:]}"
-                    else:
-                        masked_key = f"{decrypted_key[:2]}***"
-                except Exception:
-                    pass
-
             # 查找匹配的 GlobalModel
             matching_global_models: list[MappingMatchingGlobalModel] = []
 
@@ -879,6 +890,18 @@ class AdminGetProviderMappingPreviewAdapter(AdminApiAdapter):
                     total_matches += 1
 
             if matching_global_models:
+                # 只有有匹配结果的 key 才做解密脱敏，减少 CPU 开销
+                masked_key = "***"
+                if key.api_key:
+                    try:
+                        decrypted_key = crypto.decrypt(key.api_key, silent=True)
+                        if len(decrypted_key) > 8:
+                            masked_key = f"{decrypted_key[:4]}***{decrypted_key[-4:]}"
+                        else:
+                            masked_key = f"{decrypted_key[:2]}***"
+                    except Exception:
+                        pass
+
                 key_infos.append(
                     MappingMatchingKey(
                         key_id=key.id or "",
@@ -901,7 +924,7 @@ class AdminGetProviderMappingPreviewAdapter(AdminApiAdapter):
             truncated=is_truncated,
             truncated_keys=truncated_keys,
             truncated_models=truncated_models,
-        )
+        ).model_dump()
 
 
 # ========== Claude Code Pool Management ==========

src/api/admin/providers/summary.py

@@ -8,12 +8,13 @@ from typing import Any
 
 from fastapi import APIRouter, Depends, Query, Request
 from sqlalchemy import case, func
-from sqlalchemy.orm import Session
+from sqlalchemy.orm import Session, load_only
 
 from src.api.base.admin_adapter import AdminApiAdapter
 from src.api.base.context import ApiRequestContext
 from src.api.base.models_service import invalidate_models_list_cache
 from src.api.base.pipeline import ApiRequestPipeline
+from src.config.constants import CacheTTL
 from src.core.enums import ProviderBillingType
 from src.core.exceptions import InvalidRequestException, NotFoundException
 from src.core.logger import logger
@@ -39,6 +40,7 @@ from src.models.endpoint_models import (
 )
 from src.services.cache.model_cache import ModelCacheService
 from src.services.cache.provider_cache import ProviderCacheService
+from src.utils.cache_decorator import cache_result
 
 router = APIRouter(tags=["Provider Summary"])
 pipeline = ApiRequestPipeline()
@@ -130,11 +132,8 @@ async def get_provider_summary(
     - `created_at`: 创建时间
     - `updated_at`: 更新时间
     """
-    provider = db.query(Provider).filter(Provider.id == provider_id).first()
+    adapter = AdminProviderDetailAdapter(provider_id=provider_id)
-    if not provider:
+    return await pipeline.run(adapter=adapter, http_request=request, db=db, mode=adapter.mode)
-        raise NotFoundException(f"Provider {provider_id} not found")
-
-    return _build_provider_summary(db, provider)
 
 
 @router.get("/{provider_id}/health-monitor", response_model=ProviderEndpointHealthMonitorResponse)
@@ -319,12 +318,20 @@ def _extract_failover_rules_from_config(
 
 
 def _build_provider_summary(db: Session, provider: Provider) -> ProviderWithEndpointsSummary:
-    endpoints = db.query(ProviderEndpoint).filter(ProviderEndpoint.provider_id == provider.id).all()
+    endpoints = (
+        db.query(ProviderEndpoint)
+        .options(
+            load_only(
+                ProviderEndpoint.id,
+                ProviderEndpoint.provider_id,
+                ProviderEndpoint.api_format,
+                ProviderEndpoint.is_active,
+            )
+        )
+        .filter(ProviderEndpoint.provider_id == provider.id)
+        .all()
+    )
 
-    total_endpoints = len(endpoints)
-    active_endpoints = sum(1 for e in endpoints if e.is_active)
-
-    # Key 统计（合并为单个查询）
     key_stats = (
         db.query(
             func.count(ProviderAPIKey.id).label("total"),
@@ -333,10 +340,9 @@ def _build_provider_summary(db: Session, provider: Provider) -> ProviderWithEndp
         .filter(ProviderAPIKey.provider_id == provider.id)
         .first()
     )
-    total_keys = key_stats.total or 0
+    total_keys = int(key_stats.total or 0)
     active_keys = int(key_stats.active or 0)
 
-    # Model 统计（合并为单个查询）
     model_stats = (
         db.query(
             func.count(Model.id).label("total"),
@@ -345,25 +351,62 @@ def _build_provider_summary(db: Session, provider: Provider) -> ProviderWithEndp
         .filter(Model.provider_id == provider.id)
         .first()
     )
-    total_models = model_stats.total or 0
+    total_models = int(model_stats.total or 0)
     active_models = int(model_stats.active or 0)
 
-    # 活跃模型关联的全局模型 ID 列表
     global_model_ids = [
         row[0]
         for row in db.query(Model.global_model_id)
         .filter(
             Model.provider_id == provider.id,
             Model.is_active == True,
+            Model.global_model_id.isnot(None),
         )
         .distinct()
         .all()
     ]
 
-    api_formats = [e.api_format for e in endpoints]
+    all_keys = (
+        db.query(ProviderAPIKey)
+        .options(
+            load_only(
+                ProviderAPIKey.id,
+                ProviderAPIKey.provider_id,
+                ProviderAPIKey.is_active,
+                ProviderAPIKey.api_formats,
+                ProviderAPIKey.health_by_format,
+            )
+        )
+        .filter(ProviderAPIKey.provider_id == provider.id)
+        .all()
+    )
 
-    # 优化: 一次性加载 Provider 的 keys，避免 N+1 查询
+    return _compose_provider_summary(
-    all_keys = db.query(ProviderAPIKey).filter(ProviderAPIKey.provider_id == provider.id).all()
+        provider=provider,
+        endpoints=endpoints,
+        all_keys=all_keys,
+        total_keys=total_keys,
+        active_keys=active_keys,
+        total_models=total_models,
+        active_models=active_models,
+        global_model_ids=global_model_ids,
+    )
+
+
+def _compose_provider_summary(
+    *,
+    provider: Provider,
+    endpoints: list[ProviderEndpoint],
+    all_keys: list[ProviderAPIKey],
+    total_keys: int,
+    active_keys: int,
+    total_models: int,
+    active_models: int,
+    global_model_ids: list[Any],
+) -> ProviderWithEndpointsSummary:
+    total_endpoints = len(endpoints)
+    active_endpoints = sum(1 for e in endpoints if e.is_active)
+    api_formats = [e.api_format for e in endpoints]
 
     # 按 api_formats 分组 keys（通过 api_formats 关联）
     format_to_endpoint_id: dict[str, str] = {e.api_format: e.id for e in endpoints}
@@ -379,9 +422,8 @@ def _build_provider_summary(db: Session, provider: Provider) -> ProviderWithEndp
     for endpoint in endpoints:
         keys = keys_by_endpoint.get(endpoint.id, [])
         if keys:
-            # 从 health_by_format 获取对应格式的健康度
             api_fmt = endpoint.api_format
-            health_scores = []
+            health_scores: list[float] = []
             for k in keys:
                 health_by_format = k.health_by_format or {}
                 if api_fmt in health_by_format:
@@ -389,7 +431,7 @@ def _build_provider_summary(db: Session, provider: Provider) -> ProviderWithEndp
                     if score is not None:
                         health_scores.append(float(score))
                 else:
-                    health_scores.append(1.0)  # 默认健康度
+                    health_scores.append(1.0)
             avg_health = sum(health_scores) / len(health_scores) if health_scores else 1.0
             endpoint_health_map[endpoint.id] = avg_health
         else:
@@ -399,7 +441,6 @@ def _build_provider_summary(db: Session, provider: Provider) -> ProviderWithEndp
     avg_health_score = sum(all_health_scores) / len(all_health_scores) if all_health_scores else 1.0
     unhealthy_endpoints = sum(1 for score in all_health_scores if score < 0.5)
 
-    # 计算每个端点的活跃密钥数量
     active_keys_by_endpoint: dict[str, int] = {}
     for endpoint_id, keys in keys_by_endpoint.items():
         active_keys_by_endpoint[endpoint_id] = sum(1 for k in keys if k.is_active)
@@ -484,6 +525,119 @@ def _build_provider_summary(db: Session, provider: Provider) -> ProviderWithEndp
     )
 
 
+def _build_provider_summaries_batch(
+    db: Session, providers: list[Provider]
+) -> list[ProviderWithEndpointsSummary]:
+    if not providers:
+        return []
+
+    provider_ids = [provider.id for provider in providers]
+
+    endpoint_rows = (
+        db.query(ProviderEndpoint)
+        .options(
+            load_only(
+                ProviderEndpoint.id,
+                ProviderEndpoint.provider_id,
+                ProviderEndpoint.api_format,
+                ProviderEndpoint.is_active,
+            )
+        )
+        .filter(ProviderEndpoint.provider_id.in_(provider_ids))
+        .all()
+    )
+    endpoints_by_provider: dict[str, list[ProviderEndpoint]] = {}
+    for endpoint in endpoint_rows:
+        endpoints_by_provider.setdefault(str(endpoint.provider_id), []).append(endpoint)
+
+    key_rows = (
+        db.query(ProviderAPIKey)
+        .options(
+            load_only(
+                ProviderAPIKey.id,
+                ProviderAPIKey.provider_id,
+                ProviderAPIKey.is_active,
+                ProviderAPIKey.api_formats,
+                ProviderAPIKey.health_by_format,
+            )
+        )
+        .filter(ProviderAPIKey.provider_id.in_(provider_ids))
+        .all()
+    )
+    keys_by_provider: dict[str, list[ProviderAPIKey]] = {}
+    for key in key_rows:
+        keys_by_provider.setdefault(str(key.provider_id), []).append(key)
+
+    key_stats_rows = (
+        db.query(
+            ProviderAPIKey.provider_id.label("provider_id"),
+            func.count(ProviderAPIKey.id).label("total"),
+            func.sum(case((ProviderAPIKey.is_active == True, 1), else_=0)).label("active"),
+        )
+        .filter(ProviderAPIKey.provider_id.in_(provider_ids))
+        .group_by(ProviderAPIKey.provider_id)
+        .all()
+    )
+    key_stats_by_provider: dict[str, dict[str, int]] = {
+        str(row.provider_id): {
+            "total": int(row.total or 0),
+            "active": int(row.active or 0),
+        }
+        for row in key_stats_rows
+    }
+
+    model_stats_rows = (
+        db.query(
+            Model.provider_id.label("provider_id"),
+            func.count(Model.id).label("total"),
+            func.sum(case((Model.is_active == True, 1), else_=0)).label("active"),
+        )
+        .filter(Model.provider_id.in_(provider_ids))
+        .group_by(Model.provider_id)
+        .all()
+    )
+    model_stats_by_provider: dict[str, dict[str, int]] = {
+        str(row.provider_id): {
+            "total": int(row.total or 0),
+            "active": int(row.active or 0),
+        }
+        for row in model_stats_rows
+    }
+
+    global_model_rows = (
+        db.query(Model.provider_id, Model.global_model_id)
+        .filter(
+            Model.provider_id.in_(provider_ids),
+            Model.is_active == True,
+            Model.global_model_id.isnot(None),
+        )
+        .distinct()
+        .all()
+    )
+    global_model_ids_by_provider: dict[str, list[Any]] = {}
+    for provider_id, global_model_id in global_model_rows:
+        global_model_ids_by_provider.setdefault(str(provider_id), []).append(global_model_id)
+
+    summaries: list[ProviderWithEndpointsSummary] = []
+    for provider in providers:
+        pid = str(provider.id)
+        key_stats = key_stats_by_provider.get(pid, {"total": 0, "active": 0})
+        model_stats = model_stats_by_provider.get(pid, {"total": 0, "active": 0})
+        summaries.append(
+            _compose_provider_summary(
+                provider=provider,
+                endpoints=endpoints_by_provider.get(pid, []),
+                all_keys=keys_by_provider.get(pid, []),
+                total_keys=key_stats["total"],
+                active_keys=key_stats["active"],
+                total_models=model_stats["total"],
+                active_models=model_stats["active"],
+                global_model_ids=global_model_ids_by_provider.get(pid, []),
+            )
+        )
+    return summaries
+
+
 # -------- Adapters --------
 
 
@@ -493,6 +647,12 @@ class AdminProviderHealthMonitorAdapter(AdminApiAdapter):
     lookback_hours: int
     per_endpoint_limit: int
 
+    @cache_result(
+        key_prefix="admin:providers:health-monitor",
+        ttl=CacheTTL.ADMIN_USAGE_RECORDS,
+        user_specific=False,
+        vary_by=["provider_id", "lookback_hours", "per_endpoint_limit"],
+    )
     async def handle(self, context: ApiRequestContext) -> Any:  # type: ignore[override]
         db = context.db
         provider = db.query(Provider).filter(Provider.id == self.provider_id).first()
@@ -501,6 +661,14 @@ class AdminProviderHealthMonitorAdapter(AdminApiAdapter):
 
         endpoints = (
             db.query(ProviderEndpoint)
+            .options(
+                load_only(
+                    ProviderEndpoint.id,
+                    ProviderEndpoint.provider_id,
+                    ProviderEndpoint.api_format,
+                    ProviderEndpoint.is_active,
+                )
+            )
             .filter(ProviderEndpoint.provider_id == self.provider_id)
             .all()
         )
@@ -508,7 +676,7 @@ class AdminProviderHealthMonitorAdapter(AdminApiAdapter):
         now = datetime.now(timezone.utc)
         since = now - timedelta(hours=self.lookback_hours)
 
-        endpoint_ids = [endpoint.id for endpoint in endpoints]
+        endpoint_ids = [str(endpoint.id) for endpoint in endpoints]
         if not endpoint_ids:
             response = ProviderEndpointHealthMonitorResponse(
                 provider_id=provider.id,
@@ -522,46 +690,65 @@ class AdminProviderHealthMonitorAdapter(AdminApiAdapter):
                 endpoint_count=0,
                 lookback_hours=self.lookback_hours,
             )
-            return response
+            return response.model_dump()
 
-        limit_rows = max(200, self.per_endpoint_limit * max(1, len(endpoint_ids)) * 2)
+        ranked_attempts_subq = (
-        attempts_query = (
             db.query(RequestCandidate)
+            .with_entities(
+                RequestCandidate.endpoint_id.label("endpoint_id"),
+                RequestCandidate.status.label("status"),
+                RequestCandidate.status_code.label("status_code"),
+                RequestCandidate.latency_ms.label("latency_ms"),
+                RequestCandidate.error_type.label("error_type"),
+                RequestCandidate.error_message.label("error_message"),
+                func.coalesce(
+                    RequestCandidate.finished_at,
+                    RequestCandidate.started_at,
+                    RequestCandidate.created_at,
+                ).label("event_timestamp"),
+                func.row_number()
+                .over(
+                    partition_by=RequestCandidate.endpoint_id,
+                    order_by=RequestCandidate.created_at.desc(),
+                )
+                .label("rn"),
+            )
             .filter(
                 RequestCandidate.endpoint_id.in_(endpoint_ids),
                 RequestCandidate.created_at >= since,
             )
-            .order_by(RequestCandidate.created_at.desc())
+            .subquery()
+        )
+        attempt_rows = (
+            db.query(ranked_attempts_subq)
+            .filter(ranked_attempts_subq.c.rn <= self.per_endpoint_limit)
+            .order_by(
+                ranked_attempts_subq.c.endpoint_id.asc(),
+                ranked_attempts_subq.c.event_timestamp.asc(),
+            )
+            .all()
         )
-        attempts = attempts_query.limit(limit_rows).all()
 
-        buffered_attempts: dict[str, list[RequestCandidate]] = {eid: [] for eid in endpoint_ids}
+        events_by_endpoint: dict[str, list[EndpointHealthEvent]] = {eid: [] for eid in endpoint_ids}
-        counters: dict[str, int] = {eid: 0 for eid in endpoint_ids}
+        for row in attempt_rows:
-
+            endpoint_id = str(row.endpoint_id) if row.endpoint_id is not None else ""
-        for attempt in attempts:
+            if not endpoint_id or endpoint_id not in events_by_endpoint:
-            if not attempt.endpoint_id or attempt.endpoint_id not in buffered_attempts:
                 continue
-            if counters[attempt.endpoint_id] >= self.per_endpoint_limit:
+            events_by_endpoint[endpoint_id].append(
-                continue
+                EndpointHealthEvent(
-            buffered_attempts[attempt.endpoint_id].append(attempt)
+                    timestamp=row.event_timestamp,
-            counters[attempt.endpoint_id] += 1
+                    status=row.status,
+                    status_code=row.status_code,
+                    latency_ms=row.latency_ms,
+                    error_type=row.error_type,
+                    error_message=row.error_message,
+                )
+            )
 
         endpoint_monitors: list[EndpointHealthMonitor] = []
         for endpoint in endpoints:
-            attempt_list = list(reversed(buffered_attempts.get(endpoint.id, [])))
+            endpoint_id = str(endpoint.id)
-            events: list[EndpointHealthEvent] = []
+            events = events_by_endpoint.get(endpoint_id, [])
-            for attempt in attempt_list:
-                event_timestamp = attempt.finished_at or attempt.started_at or attempt.created_at
-                events.append(
-                    EndpointHealthEvent(
-                        timestamp=event_timestamp,
-                        status=attempt.status,
-                        status_code=attempt.status_code,
-                        latency_ms=attempt.latency_ms,
-                        error_type=attempt.error_type,
-                        error_message=attempt.error_message,
-                    )
-                )
 
             success_count = sum(1 for event in events if event.status == "success")
             failed_count = sum(1 for event in events if event.status == "failed")
@@ -598,10 +785,15 @@ class AdminProviderHealthMonitorAdapter(AdminApiAdapter):
             lookback_hours=self.lookback_hours,
             per_endpoint_limit=self.per_endpoint_limit,
         )
-        return response
+        return response.model_dump()
 
 
 class AdminProviderSummaryAdapter(AdminApiAdapter):
+    @cache_result(
+        key_prefix="admin:providers:summary",
+        ttl=CacheTTL.ADMIN_USAGE_RECORDS,
+        user_specific=False,
+    )
     async def handle(self, context: ApiRequestContext) -> Any:  # type: ignore[override]
         db = context.db
         providers = (
@@ -609,7 +801,25 @@ class AdminProviderSummaryAdapter(AdminApiAdapter):
             .order_by(Provider.provider_priority.asc(), Provider.created_at.asc())
             .all()
         )
-        return [_build_provider_summary(db, provider) for provider in providers]
+        return [item.model_dump() for item in _build_provider_summaries_batch(db, providers)]
+
+
+@dataclass
+class AdminProviderDetailAdapter(AdminApiAdapter):
+    provider_id: str
+
+    @cache_result(
+        key_prefix="admin:providers:summary:detail",
+        ttl=CacheTTL.ADMIN_USAGE_RECORDS,
+        user_specific=False,
+        vary_by=["provider_id"],
+    )
+    async def handle(self, context: ApiRequestContext) -> Any:  # type: ignore[override]
+        db = context.db
+        provider = db.query(Provider).filter(Provider.id == self.provider_id).first()
+        if not provider:
+            raise NotFoundException(f"Provider {self.provider_id} not found")
+        return _build_provider_summary(db, provider).model_dump()
 
 
 @dataclass

src/api/admin/stats/comparison.py

@@ -10,9 +10,11 @@ from sqlalchemy.orm import Session
 
 from src.api.base.admin_adapter import AdminApiAdapter
 from src.api.base.context import ApiRequestContext
+from src.config.constants import CacheTTL
 from src.database import get_db
 from src.services.system.stats_aggregator import AggregatedStats, StatsFilter, query_stats_hybrid
 from src.services.system.time_range import TimeRangeParams
+from src.utils.cache_decorator import cache_result
 
 from .common import pipeline
 
@@ -34,6 +36,18 @@ class AdminComparisonAdapter(AdminApiAdapter):
         self.timezone_name = timezone_name
         self.tz_offset_minutes = tz_offset_minutes
 
+    @cache_result(
+        key_prefix="admin:stats:comparison",
+        ttl=CacheTTL.ADMIN_USAGE_AGGREGATION,
+        user_specific=False,
+        vary_by=[
+            "current_start",
+            "current_end",
+            "comparison_type",
+            "timezone_name",
+            "tz_offset_minutes",
+        ],
+    )
     async def handle(self, context: ApiRequestContext) -> Any:  # type: ignore[override]
         if self.current_start > self.current_end:
             raise HTTPException(status_code=400, detail="current_start must be <= current_end")

src/api/admin/stats/cost.py

@@ -11,10 +11,12 @@ from sqlalchemy.orm import Session
 
 from src.api.base.admin_adapter import AdminApiAdapter
 from src.api.base.context import ApiRequestContext
+from src.config.constants import CacheTTL
 from src.database import get_db
 from src.models.database import Usage
 from src.services.system.stats_aggregator import query_time_series
 from src.services.system.time_range import TimeRangeParams
+from src.utils.cache_decorator import cache_result
 
 from .common import (
     _apply_admin_default_range,
@@ -40,11 +42,27 @@ class AdminCostForecastAdapter(AdminApiAdapter):
         self.days = days
         self.forecast_days = forecast_days
         self.timezone_name = timezone_name
-        self.tz_offset_minutes = tz_offset_minutes
+        self.fallback_tz_offset_minutes = tz_offset_minutes
 
+    @cache_result(
+        key_prefix="admin:stats:cost:forecast",
+        ttl=CacheTTL.ADMIN_USAGE_AGGREGATION,
+        user_specific=False,
+        vary_by=[
+            "time_range.start_date",
+            "time_range.end_date",
+            "time_range.preset",
+            "time_range.timezone",
+            "time_range.tz_offset_minutes",
+            "days",
+            "forecast_days",
+            "timezone_name",
+            "fallback_tz_offset_minutes",
+        ],
+    )
     async def handle(self, context: ApiRequestContext) -> Any:  # type: ignore[override]
         time_range = self.time_range or _build_time_range_from_days(
-            self.days, self.timezone_name, self.tz_offset_minutes
+            self.days, self.timezone_name, self.fallback_tz_offset_minutes
         )
         time_range.granularity = "day"
         try:
@@ -120,6 +138,20 @@ class AdminCostSavingsAdapter(AdminApiAdapter):
         self.provider_name = provider_name
         self.model = model
 
+    @cache_result(
+        key_prefix="admin:stats:cost:savings",
+        ttl=CacheTTL.ADMIN_USAGE_AGGREGATION,
+        user_specific=False,
+        vary_by=[
+            "time_range.start_date",
+            "time_range.end_date",
+            "time_range.preset",
+            "time_range.timezone",
+            "time_range.tz_offset_minutes",
+            "provider_name",
+            "model",
+        ],
+    )
     async def handle(self, context: ApiRequestContext) -> Any:  # type: ignore[override]
         if not self.time_range:
             return {

src/api/admin/stats/errors.py

@@ -11,9 +11,11 @@ from sqlalchemy.orm import Session
 
 from src.api.base.admin_adapter import AdminApiAdapter
 from src.api.base.context import ApiRequestContext
+from src.config.constants import CacheTTL
 from src.database import get_db
 from src.models.database import StatsDailyError, Usage
 from src.services.system.time_range import TimeRangeParams
+from src.utils.cache_decorator import cache_result
 
 from .common import _apply_admin_default_range, _build_time_range_params, pipeline
 
@@ -24,6 +26,18 @@ class AdminErrorDistributionAdapter(AdminApiAdapter):
     def __init__(self, time_range: TimeRangeParams | None) -> None:
         self.time_range = _apply_admin_default_range(time_range)
 
+    @cache_result(
+        key_prefix="admin:stats:errors:distribution",
+        ttl=CacheTTL.ADMIN_USAGE_AGGREGATION,
+        user_specific=False,
+        vary_by=[
+            "time_range.start_date",
+            "time_range.end_date",
+            "time_range.preset",
+            "time_range.timezone",
+            "time_range.tz_offset_minutes",
+        ],
+    )
     async def handle(self, context: ApiRequestContext) -> Any:  # type: ignore[override]
         if not self.time_range:
             return {"distribution": [], "trend": []}

src/api/admin/stats/performance.py

@@ -10,10 +10,12 @@ from sqlalchemy.orm import Session
 
 from src.api.base.admin_adapter import AdminApiAdapter
 from src.api.base.context import ApiRequestContext
+from src.config.constants import CacheTTL
 from src.database import get_db
 from src.models.database import StatsDaily
 from src.services.system.stats_aggregator import StatsAggregatorService
 from src.services.system.time_range import TimeRangeParams
+from src.utils.cache_decorator import cache_result
 
 from .common import _apply_admin_default_range, _build_time_range_params, pipeline
 
@@ -24,6 +26,18 @@ class AdminPercentilesAdapter(AdminApiAdapter):
     def __init__(self, time_range: TimeRangeParams | None) -> None:
         self.time_range = _apply_admin_default_range(time_range)
 
+    @cache_result(
+        key_prefix="admin:stats:performance:percentiles",
+        ttl=CacheTTL.ADMIN_USAGE_AGGREGATION,
+        user_specific=False,
+        vary_by=[
+            "time_range.start_date",
+            "time_range.end_date",
+            "time_range.preset",
+            "time_range.timezone",
+            "time_range.tz_offset_minutes",
+        ],
+    )
     async def handle(self, context: ApiRequestContext) -> Any:  # type: ignore[override]
         if not self.time_range:
             return []

src/api/admin/stats/quota.py

@@ -10,9 +10,11 @@ from sqlalchemy.orm import Session
 
 from src.api.base.admin_adapter import AdminApiAdapter
 from src.api.base.context import ApiRequestContext
+from src.config.constants import CacheTTL
 from src.core.enums import ProviderBillingType
 from src.database import get_db
 from src.models.database import Provider
+from src.utils.cache_decorator import cache_result
 
 from .common import pipeline
 
@@ -20,6 +22,11 @@ router = APIRouter()
 
 
 class AdminQuotaUsageAdapter(AdminApiAdapter):
+    @cache_result(
+        key_prefix="admin:stats:providers:quota_usage",
+        ttl=CacheTTL.ADMIN_USAGE_AGGREGATION,
+        user_specific=False,
+    )
     async def handle(self, context: ApiRequestContext) -> Any:  # type: ignore[override]
         db = context.db
         providers = (

src/api/admin/stats/time_series.py

@@ -10,9 +10,11 @@ from sqlalchemy.orm import Session
 
 from src.api.base.admin_adapter import AdminApiAdapter
 from src.api.base.context import ApiRequestContext
+from src.config.constants import CacheTTL
 from src.database import get_db
 from src.services.system.stats_aggregator import TimeSeriesFilter, query_time_series
 from src.services.system.time_range import TimeRangeParams
+from src.utils.cache_decorator import cache_result
 
 from .common import _apply_admin_default_range, _build_time_range_params, pipeline
 
@@ -32,6 +34,22 @@ class AdminTimeSeriesAdapter(AdminApiAdapter):
         self.model = model
         self.provider_name = provider_name
 
+    @cache_result(
+        key_prefix="admin:stats:time_series",
+        ttl=CacheTTL.ADMIN_USAGE_AGGREGATION,
+        user_specific=False,
+        vary_by=[
+            "time_range.start_date",
+            "time_range.end_date",
+            "time_range.preset",
+            "time_range.timezone",
+            "time_range.tz_offset_minutes",
+            "time_range.granularity",
+            "user_id",
+            "model",
+            "provider_name",
+        ],
+    )
     async def handle(self, context: ApiRequestContext) -> Any:  # type: ignore[override]
         if not self.time_range:
             return []

src/api/admin/system.py

@@ -9,11 +9,13 @@ from typing import Any
 
 from fastapi import APIRouter, Depends, HTTPException, Request, status
 from pydantic import ValidationError
+from sqlalchemy import case, func
 from sqlalchemy.orm import Session
 
 from src.api.base.admin_adapter import AdminApiAdapter
 from src.api.base.context import ApiRequestContext
 from src.api.base.pipeline import ApiRequestPipeline
+from src.config.constants import CacheTTL
 from src.core.exceptions import InvalidRequestException, NotFoundException, translate_pydantic_error
 from src.core.logger import logger
 from src.database import get_db
@@ -21,6 +23,7 @@ from src.models.api import SystemSettingsRequest, SystemSettingsResponse
 from src.models.database import ApiKey, Provider, Usage, User
 from src.services.email.email_template import EmailTemplate
 from src.services.system.config import SystemConfigService
+from src.utils.cache_decorator import cache_result
 
 router = APIRouter(prefix="/api/admin/system", tags=["Admin - System"])
 
@@ -749,14 +752,27 @@ class AdminDeleteSystemConfigAdapter(AdminApiAdapter):
 
 
 class AdminSystemStatsAdapter(AdminApiAdapter):
+    @cache_result(
+        key_prefix="admin:system:stats",
+        ttl=CacheTTL.DASHBOARD_STATS,
+        user_specific=False,
+    )
     async def handle(self, context: ApiRequestContext) -> Any:  # type: ignore[override]
         db = context.db
-        total_users = db.query(User).count()
+        user_stats = db.query(
-        active_users = db.query(User).filter(User.is_active.is_(True)).count()
+            func.count(User.id).label("total"),
-        total_providers = db.query(Provider).count()
+            func.sum(case((User.is_active.is_(True), 1), else_=0)).label("active"),
-        active_providers = db.query(Provider).filter(Provider.is_active.is_(True)).count()
+        ).first()
-        total_api_keys = db.query(ApiKey).count()
+        provider_stats = db.query(
-        total_requests = db.query(Usage).count()
+            func.count(Provider.id).label("total"),
+            func.sum(case((Provider.is_active.is_(True), 1), else_=0)).label("active"),
+        ).first()
+        total_api_keys = int(db.query(func.count(ApiKey.id)).scalar() or 0)
+        total_requests = int(db.query(func.count(Usage.id)).scalar() or 0)
+        total_users = int(user_stats.total or 0) if user_stats else 0
+        active_users = int(user_stats.active or 0) if user_stats else 0
+        total_providers = int(provider_stats.total or 0) if provider_stats else 0
+        active_providers = int(provider_stats.active or 0) if provider_stats else 0
 
         return {
             "users": {"total": total_users, "active": active_users},
@@ -776,16 +792,18 @@ class AdminTriggerCleanupAdapter(AdminApiAdapter):
         db = context.db
 
         # 获取清理前的统计信息
-        total_before = db.query(Usage).count()
+        total_before = int(db.query(func.count(Usage.id)).scalar() or 0)
         with_body_before = (
-            db.query(Usage)
+            db.query(func.count(Usage.id))
             .filter((Usage.request_body.isnot(None)) | (Usage.response_body.isnot(None)))
-            .count()
+            .scalar()
+            or 0
         )
         with_headers_before = (
-            db.query(Usage)
+            db.query(func.count(Usage.id))
             .filter((Usage.request_headers.isnot(None)) | (Usage.response_headers.isnot(None)))
-            .count()
+            .scalar()
+            or 0
         )
 
         # 触发清理
@@ -793,16 +811,18 @@ class AdminTriggerCleanupAdapter(AdminApiAdapter):
         await maintenance_scheduler._perform_cleanup()
 
         # 获取清理后的统计信息
-        total_after = db.query(Usage).count()
+        total_after = int(db.query(func.count(Usage.id)).scalar() or 0)
         with_body_after = (
-            db.query(Usage)
+            db.query(func.count(Usage.id))
             .filter((Usage.request_body.isnot(None)) | (Usage.response_body.isnot(None)))
-            .count()
+            .scalar()
+            or 0
         )
         with_headers_after = (
-            db.query(Usage)
+            db.query(func.count(Usage.id))
             .filter((Usage.request_headers.isnot(None)) | (Usage.response_headers.isnot(None)))
-            .count()
+            .scalar()
+            or 0
         )
 
         return {
@@ -2402,11 +2422,11 @@ class AdminPurgeConfigAdapter(AdminApiAdapter):
         db = context.db
 
         # 统计
-        providers_count = db.query(Provider).count()
+        providers_count = int(db.query(func.count(Provider.id)).scalar() or 0)
-        endpoints_count = db.query(ProviderEndpoint).count()
+        endpoints_count = int(db.query(func.count(ProviderEndpoint.id)).scalar() or 0)
-        keys_count = db.query(ProviderAPIKey).count()
+        keys_count = int(db.query(func.count(ProviderAPIKey.id)).scalar() or 0)
-        models_count = db.query(Model).count()
+        models_count = int(db.query(func.count(Model.id)).scalar() or 0)
-        global_models_count = db.query(GlobalModel).count()
+        global_models_count = int(db.query(func.count(GlobalModel.id)).scalar() or 0)
 
         # VideoTask 的 provider_id/endpoint_id/key_id 无 ondelete，置 NULL 保留任务记录
         db.query(VideoTask).filter(
@@ -2471,7 +2491,9 @@ class AdminPurgeUsersAdapter(AdminApiAdapter):
             )
 
             # 统计关联 API Keys 数量（DB 级别 CASCADE 会随 User 自动删除）
-            keys_count = db.query(ApiKey).filter(ApiKey.user_id.in_(user_ids)).count()
+            keys_count = int(
+                db.query(func.count(ApiKey.id)).filter(ApiKey.user_id.in_(user_ids)).scalar() or 0
+            )
 
             # 将使用记录的 user_id 置空（保留记录）
             db.query(Usage).filter(Usage.user_id.in_(user_ids)).update(
@@ -2565,9 +2587,9 @@ class AdminPurgeUsageAdapter(AdminApiAdapter):
 
         db = context.db
 
-        usage_count = db.query(Usage).count()
+        usage_count = int(db.query(func.count(Usage.id)).scalar() or 0)
-        candidates_count = db.query(RequestCandidate).count()
+        candidates_count = int(db.query(func.count(RequestCandidate.id)).scalar() or 0)
-        usage_counts_count = db.query(UserModelUsageCount).count()
+        usage_counts_count = int(db.query(func.count(UserModelUsageCount.id)).scalar() or 0)
 
         # 清空使用记录
         db.query(RequestCandidate).delete()
@@ -2594,7 +2616,7 @@ class AdminPurgeAuditLogsAdapter(AdminApiAdapter):
 
         db = context.db
 
-        count = db.query(AuditLog).count()
+        count = int(db.query(func.count(AuditLog.id)).scalar() or 0)
         db.query(AuditLog).delete()
         db.commit()
 
@@ -2612,8 +2634,8 @@ class AdminPurgeRequestBodiesAdapter(AdminApiAdapter):
         db = context.db
 
         # 统计有 body 的记录数
-        with_body = (
+        with_body = int(
-            db.query(Usage)
+            db.query(func.count(Usage.id))
             .filter(
                 (Usage.request_body.isnot(None))
                 | (Usage.response_body.isnot(None))
@@ -2624,7 +2646,8 @@ class AdminPurgeRequestBodiesAdapter(AdminApiAdapter):
                 | (Usage.provider_request_body_compressed.isnot(None))
                 | (Usage.client_response_body_compressed.isnot(None))
             )
-            .count()
+            .scalar()
+            or 0
         )
 
         # 批量清空所有 body 字段

src/api/admin/usage/routes.py

@@ -2117,7 +2117,7 @@ class CacheHitAnalysisAdapter(AdminApiAdapter):
 async def get_interval_timeline(
     request: Request,
     hours: int = Query(24, ge=1, le=720, description="分析最近多少小时的数据"),
-    limit: int = Query(10000, ge=100, le=50000, description="最大返回数据点数量"),
+    limit: int = Query(3000, ge=100, le=50000, description="最大返回数据点数量"),
     user_id: str | None = Query(None, description="指定用户 ID"),
     include_user_info: bool = Query(False, description="是否包含用户信息（用于管理员多用户视图）"),
     db: Session = Depends(get_db),
@@ -2156,6 +2156,12 @@ class IntervalTimelineAdapter(AdminApiAdapter):
         self.user_id = user_id
         self.include_user_info = include_user_info
 
+    @cache_result(
+        key_prefix="admin:usage:interval_timeline",
+        ttl=CacheTTL.ADMIN_USAGE_AGGREGATION,
+        user_specific=False,
+        vary_by=["hours", "limit", "user_id", "include_user_info"],
+    )
     async def handle(self, context: ApiRequestContext) -> Any:  # type: ignore[override]
         db = context.db
 

src/api/admin/users/routes.py

@@ -7,11 +7,13 @@ from typing import Any
 
 from fastapi import APIRouter, Depends, HTTPException, Query, Request, status
 from pydantic import ValidationError
-from sqlalchemy.orm import Session
+from sqlalchemy import func
+from sqlalchemy.orm import Session, load_only
 
 from src.api.base.admin_adapter import AdminApiAdapter
 from src.api.base.context import ApiRequestContext
 from src.api.base.pipeline import ApiRequestPipeline
+from src.config.constants import CacheTTL
 from src.core.exceptions import InvalidRequestException, NotFoundException, translate_pydantic_error
 from src.core.logger import logger
 from src.database import get_db
@@ -21,6 +23,7 @@ from src.models.database import ApiKey, User, UserRole
 from src.services.system.config import SystemConfigService
 from src.services.user.apikey import ApiKeyService
 from src.services.user.service import UserService
+from src.utils.cache_decorator import cache_result
 
 router = APIRouter(prefix="/api/admin/users", tags=["Admin - Users"])
 pipeline = ApiRequestPipeline()
@@ -282,10 +285,43 @@ class AdminListUsersAdapter(AdminApiAdapter):
         self.role = role
         self.is_active = is_active
 
+    @cache_result(
+        key_prefix="admin:users:list",
+        ttl=CacheTTL.USER,
+        user_specific=False,
+        vary_by=["skip", "limit", "role", "is_active"],
+    )
     async def handle(self, context: ApiRequestContext) -> Any:  # type: ignore[override]
         db = context.db
-        role_enum = UserRole[self.role.upper()] if self.role else None
+        role_enum = None
-        users = UserService.list_users(db, self.skip, self.limit, role_enum, self.is_active)
+        if self.role:
+            try:
+                role_enum = UserRole[self.role.upper()]
+            except KeyError as exc:
+                raise InvalidRequestException("角色参数不合法") from exc
+
+        query = db.query(User).options(
+            load_only(
+                User.id,
+                User.email,
+                User.username,
+                User.role,
+                User.allowed_providers,
+                User.allowed_api_formats,
+                User.allowed_models,
+                User.quota_usd,
+                User.used_usd,
+                User.total_usd,
+                User.is_active,
+                User.created_at,
+            )
+        )
+        if role_enum:
+            query = query.filter(User.role == role_enum)
+        if self.is_active is not None:
+            query = query.filter(User.is_active == self.is_active)
+
+        users = query.order_by(User.created_at.desc()).offset(self.skip).limit(self.limit).all()
         return [
             {
                 "id": u.id,
@@ -416,7 +452,9 @@ class AdminDeleteUserAdapter(AdminApiAdapter):
             raise HTTPException(status_code=status.HTTP_404_NOT_FOUND, detail="用户不存在")
 
         if user.role == UserRole.ADMIN:
-            admin_count = db.query(User).filter(User.role == UserRole.ADMIN).count()
+            admin_count = int(
+                db.query(func.count(User.id)).filter(User.role == UserRole.ADMIN).scalar() or 0
+            )
             if admin_count <= 1:
                 raise InvalidRequestException("不能删除最后一个管理员账户")
 

src/api/admin/video_tasks/routes.py

@@ -18,11 +18,13 @@ from src.api.base.context import ApiRequestContext
 from src.api.base.pipeline import ApiRequestPipeline
 from src.api.dashboard.routes import DashboardAdapter
 from src.clients.http_client import HTTPClientPool
+from src.config.constants import CacheTTL
 from src.core.crypto import crypto_service
 from src.core.enums import UserRole
 from src.core.logger import logger
 from src.database import get_db
 from src.models.database import Provider, ProviderAPIKey, ProviderEndpoint, User, VideoTask
+from src.utils.cache_decorator import cache_result
 
 router = APIRouter(prefix="/api/admin/video-tasks", tags=["Admin - Video Tasks"])
 pipeline = ApiRequestPipeline()
@@ -247,6 +249,12 @@ class VideoTaskListAdapter(DashboardAdapter):
     page: int
     page_size: int
 
+    @cache_result(
+        key_prefix="admin:video_tasks:list",
+        ttl=min(5, CacheTTL.ADMIN_USAGE_RECORDS),
+        user_specific=True,
+        vary_by=["status", "user_id", "model", "page", "page_size"],
+    )
     async def handle(self, context: ApiRequestContext) -> Any:
         db = context.db
         user = context.user
@@ -270,8 +278,8 @@ class VideoTaskListAdapter(DashboardAdapter):
             escaped = self.model.replace("%", "\\%").replace("_", "\\_")
             query = query.filter(VideoTask.model.ilike(f"%{escaped}%"))
 
-        # 统计总数
+        # 统计总数（避免 Query.count() 生成大子查询）
-        total = query.count()
+        total = int(query.with_entities(func.count(VideoTask.id)).scalar() or 0)
 
         # 分页
         offset = (self.page - 1) * self.page_size
@@ -341,6 +349,11 @@ class VideoTaskListAdapter(DashboardAdapter):
 class VideoTaskStatsAdapter(DashboardAdapter):
     """视频任务统计适配器"""
 
+    @cache_result(
+        key_prefix="admin:video_tasks:stats",
+        ttl=min(5, CacheTTL.ADMIN_USAGE_RECORDS),
+        user_specific=True,
+    )
     async def handle(self, context: ApiRequestContext) -> Any:
         db = context.db
         user = context.user
@@ -350,8 +363,8 @@ class VideoTaskStatsAdapter(DashboardAdapter):
         if not is_admin:
             base_query = base_query.filter(VideoTask.user_id == user.id)
 
-        # 总数
+        # 总数（避免 Query.count() 生成大子查询）
-        total = base_query.count()
+        total = int(base_query.with_entities(func.count(VideoTask.id)).scalar() or 0)
 
         # 按状态分组
         status_stats = (
@@ -379,7 +392,12 @@ class VideoTaskStatsAdapter(DashboardAdapter):
 
         # 今日任务数
         today = datetime.now(timezone.utc).replace(hour=0, minute=0, second=0, microsecond=0)
-        today_count = base_query.filter(VideoTask.created_at >= today).count()
+        today_count = int(
+            base_query.filter(VideoTask.created_at >= today)
+            .with_entities(func.count(VideoTask.id))
+            .scalar()
+            or 0
+        )
 
         # 管理员额外统计
         result = {

src/api/base/pagination.py

@@ -2,6 +2,7 @@ from collections.abc import Sequence
 from dataclasses import asdict, dataclass
 from typing import Any, TypeVar
 
+from sqlalchemy import func
 from sqlalchemy.orm import Query
 
 T = TypeVar("T")
@@ -22,7 +23,7 @@ def paginate_query(query: Query, limit: int, offset: int) -> tuple[int, list[T]]
     """
     对 SQLAlchemy 查询应用 limit/offset，并返回总数与结果列表。
     """
-    total = query.order_by(None).count()
+    total = int(query.order_by(None).with_entities(func.count()).scalar() or 0)
     records = query.offset(offset).limit(limit).all()
     return total, records
 

src/api/dashboard/routes.py

@@ -7,7 +7,7 @@ from datetime import date, datetime, timedelta, timezone
 from typing import Any
 
 from fastapi import APIRouter, Depends, HTTPException, Query, Request
-from sqlalchemy import and_, func
+from sqlalchemy import and_, case, func
 from sqlalchemy.orm import Session
 
 from src.api.base.adapter import ApiAdapter, ApiMode
@@ -661,95 +661,96 @@ class UserDashboardStatsAdapter(DashboardAdapter):
         month_start_local = today_local.replace(day=1)
         month_start = month_start_local.astimezone(timezone.utc)
 
-        user_api_keys = db.query(func.count(ApiKey.id)).filter(ApiKey.user_id == user.id).scalar()
+        api_key_stats = (
-        active_keys = (
-            db.query(func.count(ApiKey.id))
-            .filter(and_(ApiKey.user_id == user.id, ApiKey.is_active.is_(True)))
-            .scalar()
-        )
-
-        # 全局 Token 统计
-        all_time_token_stats = (
             db.query(
-                func.sum(Usage.input_tokens).label("input_tokens"),
+                func.count(ApiKey.id).label("total"),
-                func.sum(Usage.output_tokens).label("output_tokens"),
+                func.sum(case((ApiKey.is_active.is_(True), 1), else_=0)).label("active"),
-                func.sum(Usage.cache_creation_input_tokens).label("cache_creation_tokens"),
+            )
-                func.sum(Usage.cache_read_input_tokens).label("cache_read_tokens"),
+            .filter(ApiKey.user_id == user.id)
+            .first()
+        )
+        user_api_keys = int(api_key_stats.total or 0) if api_key_stats else 0
+        active_keys = int(api_key_stats.active or 0) if api_key_stats else 0
+
+        # 使用单次聚合查询返回全量 + 本月 + 今日 + 昨日统计
+        usage_stats = (
+            db.query(
+                # 全量 Token 统计
+                func.sum(Usage.input_tokens).label("all_time_input_tokens"),
+                func.sum(Usage.output_tokens).label("all_time_output_tokens"),
+                func.sum(Usage.cache_creation_input_tokens).label("all_time_cache_creation_tokens"),
+                func.sum(Usage.cache_read_input_tokens).label("all_time_cache_read_tokens"),
+                # 本月
+                func.sum(case((Usage.created_at >= month_start, 1), else_=0)).label(
+                    "monthly_requests"
+                ),
+                func.sum(
+                    case((Usage.created_at >= month_start, Usage.total_cost_usd), else_=0.0)
+                ).label("monthly_cost"),
+                func.sum(
+                    case(
+                        (Usage.created_at >= month_start, Usage.cache_creation_input_tokens),
+                        else_=0,
+                    )
+                ).label("monthly_cache_creation_tokens"),
+                func.sum(
+                    case((Usage.created_at >= month_start, Usage.cache_read_input_tokens), else_=0)
+                ).label("monthly_cache_read_tokens"),
+                func.sum(
+                    case((Usage.created_at >= month_start, Usage.input_tokens), else_=0)
+                ).label("monthly_input_tokens"),
+                # 今日
+                func.sum(case((Usage.created_at >= today, 1), else_=0)).label("today_requests"),
+                func.sum(case((Usage.created_at >= today, Usage.total_cost_usd), else_=0.0)).label(
+                    "today_cost"
+                ),
+                func.sum(case((Usage.created_at >= today, Usage.total_tokens), else_=0)).label(
+                    "today_tokens"
+                ),
+                func.sum(
+                    case((Usage.created_at >= today, Usage.cache_creation_input_tokens), else_=0)
+                ).label("today_cache_creation_tokens"),
+                func.sum(
+                    case((Usage.created_at >= today, Usage.cache_read_input_tokens), else_=0)
+                ).label("today_cache_read_tokens"),
+                # 昨日（用于变化趋势）
+                func.sum(
+                    case(
+                        (
+                            and_(
+                                Usage.created_at >= yesterday,
+                                Usage.created_at < today,
+                            ),
+                            1,
+                        ),
+                        else_=0,
+                    )
+                ).label("yesterday_requests"),
             )
             .filter(Usage.user_id == user.id)
             .first()
         )
-        all_time_input_tokens = (
+
-            int(all_time_token_stats.input_tokens or 0) if all_time_token_stats else 0
+        all_time_input_tokens = int(usage_stats.all_time_input_tokens or 0) if usage_stats else 0
-        )
+        all_time_output_tokens = int(usage_stats.all_time_output_tokens or 0) if usage_stats else 0
-        all_time_output_tokens = (
-            int(all_time_token_stats.output_tokens or 0) if all_time_token_stats else 0
-        )
         all_time_cache_creation = (
-            int(all_time_token_stats.cache_creation_tokens or 0) if all_time_token_stats else 0
+            int(usage_stats.all_time_cache_creation_tokens or 0) if usage_stats else 0
-        )
-        all_time_cache_read = (
-            int(all_time_token_stats.cache_read_tokens or 0) if all_time_token_stats else 0
         )
+        all_time_cache_read = int(usage_stats.all_time_cache_read_tokens or 0) if usage_stats else 0
 
-        # 本月请求统计
+        user_requests = int(usage_stats.monthly_requests or 0) if usage_stats else 0
-        user_requests = (
+        user_cost = float(usage_stats.monthly_cost or 0.0) if usage_stats else 0.0
-            db.query(func.count(Usage.id))
-            .filter(and_(Usage.user_id == user.id, Usage.created_at >= month_start))
-            .scalar()
-        )
-        user_cost = (
-            db.query(func.sum(Usage.total_cost_usd))
-            .filter(and_(Usage.user_id == user.id, Usage.created_at >= month_start))
-            .scalar()
-            or 0
-        )
 
-        # 今日统计
+        requests_today = int(usage_stats.today_requests or 0) if usage_stats else 0
-        requests_today = (
+        cost_today = float(usage_stats.today_cost or 0.0) if usage_stats else 0.0
-            db.query(func.count(Usage.id))
+        tokens_today = int(usage_stats.today_tokens or 0) if usage_stats else 0
-            .filter(and_(Usage.user_id == user.id, Usage.created_at >= today))
+        requests_yesterday = int(usage_stats.yesterday_requests or 0) if usage_stats else 0
-            .scalar()
-        )
-        cost_today = (
-            db.query(func.sum(Usage.total_cost_usd))
-            .filter(and_(Usage.user_id == user.id, Usage.created_at >= today))
-            .scalar()
-            or 0
-        )
-        tokens_today = (
-            db.query(func.sum(Usage.total_tokens))
-            .filter(and_(Usage.user_id == user.id, Usage.created_at >= today))
-            .scalar()
-            or 0
-        )
 
-        # 昨日统计（用于计算变化）
+        cache_creation_tokens = (
-        requests_yesterday = (
+            int(usage_stats.monthly_cache_creation_tokens or 0) if usage_stats else 0
-            db.query(func.count(Usage.id))
-            .filter(
-                and_(
-                    Usage.user_id == user.id,
-                    Usage.created_at >= yesterday,
-                    Usage.created_at < today,
-                )
-            )
-            .scalar()
         )
-
+        cache_read_tokens = int(usage_stats.monthly_cache_read_tokens or 0) if usage_stats else 0
-        # 缓存统计（本月）
+        monthly_input_tokens = int(usage_stats.monthly_input_tokens or 0) if usage_stats else 0
-        cache_stats = (
-            db.query(
-                func.sum(Usage.cache_creation_input_tokens).label("cache_creation_tokens"),
-                func.sum(Usage.cache_read_input_tokens).label("cache_read_tokens"),
-                func.sum(Usage.input_tokens).label("total_input_tokens"),
-            )
-            .filter(and_(Usage.user_id == user.id, Usage.created_at >= month_start))
-            .first()
-        )
-        cache_creation_tokens = int(cache_stats.cache_creation_tokens or 0) if cache_stats else 0
-        cache_read_tokens = int(cache_stats.cache_read_tokens or 0) if cache_stats else 0
-        monthly_input_tokens = int(cache_stats.total_input_tokens or 0) if cache_stats else 0
 
         # 计算本月缓存命中率：cache_read / (input_tokens + cache_read)
         # input_tokens 是实际发送给模型的输入（不含缓存读取），cache_read 是从缓存读取的
@@ -762,19 +763,11 @@ class UserDashboardStatsAdapter(DashboardAdapter):
         )
 
         # 今日缓存统计
-        cache_stats_today = (
-            db.query(
-                func.sum(Usage.cache_creation_input_tokens).label("cache_creation_tokens"),
-                func.sum(Usage.cache_read_input_tokens).label("cache_read_tokens"),
-            )
-            .filter(and_(Usage.user_id == user.id, Usage.created_at >= today))
-            .first()
-        )
         cache_creation_tokens_today = (
-            int(cache_stats_today.cache_creation_tokens or 0) if cache_stats_today else 0
+            int(usage_stats.today_cache_creation_tokens or 0) if usage_stats else 0
         )
         cache_read_tokens_today = (
-            int(cache_stats_today.cache_read_tokens or 0) if cache_stats_today else 0
+            int(usage_stats.today_cache_read_tokens or 0) if usage_stats else 0
         )
 
         # 配额状态
@@ -859,6 +852,12 @@ class UserDashboardStatsAdapter(DashboardAdapter):
 class DashboardRecentRequestsAdapter(DashboardAdapter):
     limit: int
 
+    @cache_result(
+        key_prefix="dashboard:recent:requests",
+        ttl=CacheTTL.ADMIN_USAGE_RECORDS,
+        user_specific=True,
+        vary_by=["limit"],
+    )
     async def handle(self, context: ApiRequestContext) -> Any:  # type: ignore[override]
         db = context.db
         user = context.user
@@ -1003,28 +1002,53 @@ class DashboardDailyStatsAdapter(DashboardAdapter):
 
             # 补充 unique_models / unique_providers
             # query_time_series 使用小时粒度数据，不含这些维度统计
-            # 直接从 Usage 表按本地日 UTC 范围查询，避免 StatsDaily 历史数据未回填的问题
+            # 使用 CASE 一次性分桶，避免按天循环查询造成 N 次 SQL。
             granularity = (self.time_range.granularity or "day").lower()
-            if formatted and granularity == "day":
+            local_days = self.time_range.get_local_day_hours()
-                local_days = self.time_range.get_local_day_hours()
+            range_start = local_days[0][1] if local_days else None
-                enrichment: dict[str, dict] = {}
+            range_end = local_days[-1][2] if local_days else None
+            day_bucket = (
+                case(
+                    *[
+                        (
+                            and_(Usage.created_at >= day_start, Usage.created_at < day_end),
+                            local_date.isoformat(),
+                        )
+                        for local_date, day_start, day_end in local_days
+                    ],
+                    else_=None,
+                ).label("local_day")
+                if local_days
+                else None
+            )
 
-                for local_date, day_start_utc, day_end_utc in local_days:
+            if (
-                    q = db.query(
+                formatted
-                        func.count(func.distinct(Usage.model)).label("um"),
+                and granularity == "day"
-                        func.count(func.distinct(Usage.provider_name)).label("up"),
+                and day_bucket is not None
-                    ).filter(
+                and range_start
-                        Usage.created_at >= day_start_utc,
+                and range_end
-                        Usage.created_at < day_end_utc,
+            ):
-                    )
+                enrichment: dict[str, dict[str, int]] = {}
-                    if not is_admin:
+                enrich_query = db.query(
-                        q = q.filter(Usage.user_id == user.id)
+                    day_bucket,
-                    row = q.first()
+                    func.count(func.distinct(Usage.model)).label("um"),
-                    if row:
+                    func.count(func.distinct(Usage.provider_name)).label("up"),
-                        enrichment[local_date.isoformat()] = {
+                ).filter(
-                            "unique_models": row.um or 0,
+                    Usage.created_at >= range_start,
-                            "unique_providers": row.up or 0,
+                    Usage.created_at < range_end,
-                        }
+                )
+                if not is_admin:
+                    enrich_query = enrich_query.filter(Usage.user_id == user.id)
+                enrich_rows = enrich_query.group_by(day_bucket).all()
+
+                for local_day, unique_models, unique_providers in enrich_rows:
+                    if not local_day:
+                        continue
+                    enrichment[str(local_day)] = {
+                        "unique_models": int(unique_models or 0),
+                        "unique_providers": int(unique_providers or 0),
+                    }
 
                 for item in formatted:
                     date_key = item["date"][:10]  # YYYY-MM-DD
@@ -1062,26 +1086,35 @@ class DashboardDailyStatsAdapter(DashboardAdapter):
 
             # Daily model breakdown (aligned to local days)
             breakdown_map: dict[str, list[dict]] = {}
-            for local_date, day_start, day_end in self.time_range.get_local_day_hours():
+            if granularity == "day" and day_bucket is not None and range_start and range_end:
-                day_query = db.query(
+                breakdown_query = db.query(
+                    day_bucket,
                     Usage.model,
                     func.count(Usage.id).label("requests"),
                     func.sum(Usage.total_tokens).label("tokens"),
                     func.sum(Usage.total_cost_usd).label("cost"),
-                ).filter(Usage.created_at >= day_start, Usage.created_at < day_end)
+                ).filter(
+                    Usage.created_at >= range_start,
+                    Usage.created_at < range_end,
+                )
                 if not is_admin:
-                    day_query = day_query.filter(Usage.user_id == user.id)
+                    breakdown_query = breakdown_query.filter(Usage.user_id == user.id)
-                day_stats = day_query.group_by(Usage.model).all()
+                breakdown_rows = (
-                breakdown_map[local_date.isoformat()] = [
+                    breakdown_query.group_by(day_bucket, Usage.model)
-                    {
+                    .order_by(day_bucket.asc(), func.sum(Usage.total_cost_usd).desc())
-                        "model": stat.model,
+                    .all()
-                        "requests": stat.requests or 0,
+                )
-                        "tokens": int(stat.tokens or 0),
+                for local_day, model_name, requests, tokens, cost in breakdown_rows:
-                        "cost": float(stat.cost or 0),
+                    if not local_day or not model_name:
-                    }
+                        continue
-                    for stat in day_stats
+                    breakdown_map.setdefault(str(local_day), []).append(
-                    if stat.model
+                        {
-                ]
+                            "model": model_name,
+                            "requests": int(requests or 0),
+                            "tokens": int(tokens or 0),
+                            "cost": float(cost or 0.0),
+                        }
+                    )
 
             for item in formatted:
                 item["model_breakdown"] = breakdown_map.get(item["date"], [])

src/api/public/catalog.py

@@ -11,12 +11,13 @@ from datetime import datetime, timedelta, timezone
 from typing import Any
 
 from fastapi import APIRouter, Depends, Query, Request
-from sqlalchemy import and_, or_
+from sqlalchemy import and_, func, or_
-from sqlalchemy.orm import Session, joinedload
+from sqlalchemy.orm import Session, joinedload, load_only
 
 from src.api.base.adapter import ApiAdapter, ApiMode
 from src.api.base.context import ApiRequestContext
 from src.api.base.pipeline import ApiRequestPipeline
+from src.config.constants import CacheTTL
 from src.core.logger import logger
 from src.database import get_db
 from src.models.api import (
@@ -40,6 +41,7 @@ from src.models.endpoint_models import (
 )
 from src.services.health.endpoint import EndpointHealthService
 from src.services.system.config import SystemConfigService
+from src.utils.cache_decorator import cache_result
 
 router = APIRouter(prefix="/api/public", tags=["System Catalog"])
 pipeline = ApiRequestPipeline()
@@ -300,33 +302,84 @@ class PublicProvidersAdapter(PublicApiAdapter):
     skip: int
     limit: int
 
+    @cache_result(
+        key_prefix="public:catalog:providers",
+        ttl=CacheTTL.PROVIDER,
+        user_specific=False,
+        vary_by=["is_active", "skip", "limit"],
+    )
     async def handle(self, context: ApiRequestContext) -> Any:  # type: ignore[override]
         db = context.db
         logger.debug("公共API请求提供商列表")
-        query = db.query(Provider)
+        query = db.query(Provider).options(
+            load_only(
+                Provider.id,
+                Provider.name,
+                Provider.description,
+                Provider.is_active,
+                Provider.provider_priority,
+            )
+        )
         if self.is_active is not None:
             query = query.filter(Provider.is_active == self.is_active)
         else:
             query = query.filter(Provider.is_active.is_(True))
 
         providers = query.offset(self.skip).limit(self.limit).all()
+        provider_ids = [provider.id for provider in providers]
+
+        models_count_map: dict[str, int] = {}
+        active_models_count_map: dict[str, int] = {}
+        endpoints_count_map: dict[str, int] = {}
+        active_endpoints_count_map: dict[str, int] = {}
+        if provider_ids:
+            model_counts = (
+                db.query(Model.provider_id, func.count(Model.id))
+                .filter(Model.provider_id.in_(provider_ids))
+                .group_by(Model.provider_id)
+                .all()
+            )
+            models_count_map = {provider_id: int(count) for provider_id, count in model_counts}
+
+            active_model_counts = (
+                db.query(Model.provider_id, func.count(Model.id))
+                .filter(Model.provider_id.in_(provider_ids), Model.is_active.is_(True))
+                .group_by(Model.provider_id)
+                .all()
+            )
+            active_models_count_map = {
+                provider_id: int(count) for provider_id, count in active_model_counts
+            }
+
+            endpoint_counts = (
+                db.query(ProviderEndpoint.provider_id, func.count(ProviderEndpoint.id))
+                .filter(ProviderEndpoint.provider_id.in_(provider_ids))
+                .group_by(ProviderEndpoint.provider_id)
+                .all()
+            )
+            endpoints_count_map = {
+                provider_id: int(count) for provider_id, count in endpoint_counts
+            }
+
+            active_endpoint_counts = (
+                db.query(ProviderEndpoint.provider_id, func.count(ProviderEndpoint.id))
+                .filter(
+                    ProviderEndpoint.provider_id.in_(provider_ids),
+                    ProviderEndpoint.is_active.is_(True),
+                )
+                .group_by(ProviderEndpoint.provider_id)
+                .all()
+            )
+            active_endpoints_count_map = {
+                provider_id: int(count) for provider_id, count in active_endpoint_counts
+            }
+
         result = []
         for provider in providers:
-            models_count = db.query(Model).filter(Model.provider_id == provider.id).count()
+            models_count = models_count_map.get(provider.id, 0)
-            active_models_count = (
+            active_models_count = active_models_count_map.get(provider.id, 0)
-                db.query(Model)
+            endpoints_count = endpoints_count_map.get(provider.id, 0)
-                .filter(
+            active_endpoints_count = active_endpoints_count_map.get(provider.id, 0)
-                    and_(
-                        Model.provider_id == provider.id,
-                        Model.is_active.is_(True),
-                    )
-                )
-                .count()
-            )
-            endpoints_count = len(provider.endpoints) if provider.endpoints else 0
-            active_endpoints_count = (
-                sum(1 for ep in provider.endpoints if ep.is_active) if provider.endpoints else 0
-            )
             provider_data = PublicProviderResponse(
                 id=provider.id,
                 name=provider.name,
@@ -351,6 +404,12 @@ class PublicModelsAdapter(PublicApiAdapter):
     skip: int
     limit: int
 
+    @cache_result(
+        key_prefix="public:catalog:models",
+        ttl=CacheTTL.ADMIN_USAGE_AGGREGATION,
+        user_specific=False,
+        vary_by=["provider_id", "is_active", "skip", "limit"],
+    )
     async def handle(self, context: ApiRequestContext) -> Any:  # type: ignore[override]
         db = context.db
         logger.debug("公共API请求模型列表")
@@ -407,12 +466,19 @@ class PublicModelsAdapter(PublicApiAdapter):
 
 
 class PublicStatsAdapter(PublicApiAdapter):
+    @cache_result(
+        key_prefix="public:catalog:stats",
+        ttl=CacheTTL.ADMIN_USAGE_AGGREGATION,
+        user_specific=False,
+    )
     async def handle(self, context: ApiRequestContext) -> Any:  # type: ignore[override]
         db = context.db
         logger.debug("公共API请求系统统计信息")
-        active_providers = db.query(Provider).filter(Provider.is_active.is_(True)).count()
+        active_providers = int(
-        active_models = (
+            db.query(func.count(Provider.id)).filter(Provider.is_active.is_(True)).scalar() or 0
-            db.query(Model)
+        )
+        active_models = int(
+            db.query(func.count(Model.id))
             .join(Provider)
             .filter(
                 and_(
@@ -420,12 +486,21 @@ class PublicStatsAdapter(PublicApiAdapter):
                     Provider.is_active.is_(True),
                 )
             )
-            .count()
+            .scalar()
+            or 0
         )
         formats = (
-            db.query(Provider.api_format).filter(Provider.is_active.is_(True)).distinct().all()
+            db.query(ProviderEndpoint.api_format)
+            .join(Provider, ProviderEndpoint.provider_id == Provider.id)
+            .filter(
+                ProviderEndpoint.is_active.is_(True),
+                Provider.is_active.is_(True),
+                ProviderEndpoint.api_format.isnot(None),
+            )
+            .distinct()
+            .all()
         )
-        supported_formats = [f.api_format for f in formats if f.api_format]
+        supported_formats = [row[0] for row in formats if row[0]]
         stats = ProviderStatsResponse(
             total_providers=active_providers,
             active_providers=active_providers,
@@ -443,6 +518,12 @@ class PublicSearchModelsAdapter(PublicApiAdapter):
     provider_id: int | None
     limit: int
 
+    @cache_result(
+        key_prefix="public:catalog:search_models",
+        ttl=CacheTTL.ADMIN_USAGE_AGGREGATION,
+        user_specific=False,
+        vary_by=["query", "provider_id", "limit"],
+    )
     async def handle(self, context: ApiRequestContext) -> Any:  # type: ignore[override]
         db = context.db
         logger.debug(f"公共API搜索模型: {self.query}")
@@ -512,6 +593,12 @@ class PublicApiFormatHealthMonitorAdapter(PublicApiAdapter):
     lookback_hours: int
     per_format_limit: int
 
+    @cache_result(
+        key_prefix="public:catalog:health_api_formats",
+        ttl=CacheTTL.ADMIN_USAGE_RECORDS,
+        user_specific=False,
+        vary_by=["lookback_hours", "per_format_limit"],
+    )
     async def handle(self, context: ApiRequestContext) -> Any:  # type: ignore[override]
         db = context.db
         now = datetime.now(timezone.utc)
@@ -657,7 +744,7 @@ class PublicApiFormatHealthMonitorAdapter(PublicApiAdapter):
         )
 
         logger.debug(f"公开健康监控: 返回 {len(monitors)} 个 API 格式的健康数据")
-        return response
+        return response.model_dump()
 
 
 @dataclass
@@ -669,6 +756,12 @@ class PublicGlobalModelsAdapter(PublicApiAdapter):
     is_active: bool | None
     search: str | None
 
+    @cache_result(
+        key_prefix="public:catalog:global_models",
+        ttl=CacheTTL.MODEL,
+        user_specific=False,
+        vary_by=["skip", "limit", "is_active", "search"],
+    )
     async def handle(self, context: ApiRequestContext) -> Any:  # type: ignore[override]
         db = context.db
         logger.debug("公共API请求 GlobalModel 列表")
@@ -691,8 +784,8 @@ class PublicGlobalModelsAdapter(PublicApiAdapter):
                 )
             )
 
-        # 统计总数
+        # 统计总数（避免 Query.count() 生成大子查询）
-        total = query.count()
+        total = int(query.with_entities(func.count(GlobalModel.id)).scalar() or 0)
 
         # 分页
         models = query.order_by(GlobalModel.name).offset(self.skip).limit(self.limit).all()
@@ -714,4 +807,4 @@ class PublicGlobalModelsAdapter(PublicApiAdapter):
             )
 
         logger.debug(f"返回 {len(model_responses)} 个 GlobalModel")
-        return PublicGlobalModelListResponse(models=model_responses, total=total)
+        return PublicGlobalModelListResponse(models=model_responses, total=total).model_dump()

src/api/public/system_catalog.py

@@ -12,15 +12,18 @@ from typing import Any
 import httpx
 from fastapi import APIRouter, Depends, HTTPException, Query, Request
 from sqlalchemy import func
-from sqlalchemy.orm import Session, selectinload
+from sqlalchemy.orm import Session, load_only, selectinload
 
-from src.api.handlers.base.request_builder import PassthroughRequestBuilder
+from src.api.handlers.base.request_builder import (
-from src.api.handlers.base.request_builder import build_test_request_body, get_provider_auth
+    PassthroughRequestBuilder,
+    build_test_request_body,
+    get_provider_auth,
+)
 from src.clients.redis_client import get_redis_client
 from src.core.logger import logger
 from src.database import get_db
 from src.database.database import get_pool_status
-from src.models.database import Model, Provider, ProviderAPIKey, ProviderEndpoint
+from src.models.database import GlobalModel, Model, Provider, ProviderAPIKey, ProviderEndpoint
 from src.services.provider.transport import build_provider_url
 from src.utils.ssl_utils import get_ssl_context
 
@@ -85,7 +88,7 @@ def _serialize_provider(
 
 def _select_provider(db: Session, provider_name: str | None) -> Provider | None:
     """选择 Provider（按 provider_priority 优先级选择）"""
-    query = db.query(Provider).filter(Provider.is_active == True)
+    query = db.query(Provider).filter(Provider.is_active.is_(True))
     if provider_name:
         provider = query.filter(Provider.name == provider_name).first()
         if provider:
@@ -102,9 +105,9 @@ def _select_provider(db: Session, provider_name: str | None) -> Provider | None:
 async def service_health(db: Session = Depends(get_db)) -> Any:
     """返回服务健康状态与依赖信息"""
     active_providers = (
-        db.query(func.count(Provider.id)).filter(Provider.is_active == True).scalar() or 0
+        db.query(func.count(Provider.id)).filter(Provider.is_active.is_(True)).scalar() or 0
     )
-    active_models = db.query(func.count(Model.id)).filter(Model.is_active == True).scalar() or 0
+    active_models = db.query(func.count(Model.id)).filter(Model.is_active.is_(True)).scalar() or 0
 
     redis_info: dict[str, Any] = {"status": "unknown"}
     try:
@@ -163,11 +166,14 @@ async def root(db: Session = Depends(get_db)) -> Any:
     # 按优先级选择最高优先级的提供商
     top_provider = (
         db.query(Provider)
-        .filter(Provider.is_active == True)
+        .options(load_only(Provider.id, Provider.name, Provider.provider_priority))
+        .filter(Provider.is_active.is_(True))
         .order_by(Provider.provider_priority.asc())
         .first()
     )
-    active_providers = db.query(Provider).filter(Provider.is_active == True).count()
+    active_providers = (
+        db.query(func.count(Provider.id)).filter(Provider.is_active.is_(True)).scalar() or 0
+    )
 
     return {
         "message": "AI Proxy with Modular Architecture v4.0.0",
@@ -193,17 +199,37 @@ async def list_providers(
     active_only: bool = Query(True),
 ) -> Any:
     """列出所有 Provider"""
-    load_options = []
+    load_options = [
+        load_only(Provider.id, Provider.name, Provider.is_active, Provider.provider_priority)
+    ]
     if include_models:
-        load_options.append(selectinload(Provider.models).selectinload(Model.global_model))
+        load_options.append(
+            selectinload(Provider.models)
+            .load_only(
+                Model.id,
+                Model.provider_model_name,
+                Model.is_active,
+                Model.supports_streaming,
+                Model.global_model_id,
+            )
+            .selectinload(Model.global_model)
+            .load_only(GlobalModel.id, GlobalModel.name, GlobalModel.display_name)
+        )
     if include_endpoints:
-        load_options.append(selectinload(Provider.endpoints))
+        load_options.append(
+            selectinload(Provider.endpoints).load_only(
+                ProviderEndpoint.id,
+                ProviderEndpoint.base_url,
+                ProviderEndpoint.api_format,
+                ProviderEndpoint.is_active,
+            )
+        )
 
     base_query = db.query(Provider)
     if load_options:
         base_query = base_query.options(*load_options)
     if active_only:
-        base_query = base_query.filter(Provider.is_active == True)
+        base_query = base_query.filter(Provider.is_active.is_(True))
     base_query = base_query.order_by(Provider.provider_priority.asc(), Provider.name.asc())
 
     providers = base_query.all()
@@ -223,11 +249,31 @@ async def provider_detail(
     include_endpoints: bool = Query(False),
 ) -> Any:
     """获取单个 Provider 详情"""
-    load_options = []
+    load_options = [
+        load_only(Provider.id, Provider.name, Provider.is_active, Provider.provider_priority)
+    ]
     if include_models:
-        load_options.append(selectinload(Provider.models).selectinload(Model.global_model))
+        load_options.append(
+            selectinload(Provider.models)
+            .load_only(
+                Model.id,
+                Model.provider_model_name,
+                Model.is_active,
+                Model.supports_streaming,
+                Model.global_model_id,
+            )
+            .selectinload(Model.global_model)
+            .load_only(GlobalModel.id, GlobalModel.name, GlobalModel.display_name)
+        )
     if include_endpoints:
-        load_options.append(selectinload(Provider.endpoints))
+        load_options.append(
+            selectinload(Provider.endpoints).load_only(
+                ProviderEndpoint.id,
+                ProviderEndpoint.base_url,
+                ProviderEndpoint.api_format,
+                ProviderEndpoint.is_active,
+            )
+        )
 
     base_query = db.query(Provider)
     if load_options:

src/api/user_me/routes.py

@@ -14,6 +14,7 @@ from sqlalchemy.orm import Session
 from src.api.base.authenticated_adapter import AuthenticatedApiAdapter
 from src.api.base.context import ApiRequestContext
 from src.api.base.pipeline import ApiRequestPipeline
+from src.config.constants import CacheTTL
 from src.core.crypto import crypto_service
 from src.core.exceptions import (
     ForbiddenException,
@@ -45,6 +46,7 @@ from src.services.system.time_range import TimeRangeParams
 from src.services.usage.service import UsageService
 from src.services.user.apikey import ApiKeyService
 from src.services.user.preference import PreferenceService
+from src.utils.cache_decorator import cache_result
 
 router = APIRouter(prefix="/api/users/me", tags=["User Profile"])
 pipeline = ApiRequestPipeline()
@@ -259,7 +261,7 @@ async def get_my_active_requests(
 async def get_my_interval_timeline(
     request: Request,
     hours: int = Query(24, ge=1, le=720, description="分析最近多少小时的数据"),
-    limit: int = Query(5000, ge=100, le=20000, description="最大返回数据点数量"),
+    limit: int = Query(2000, ge=100, le=20000, description="最大返回数据点数量"),
     db: Session = Depends(get_db),
 ) -> Any:
     """
@@ -767,6 +769,21 @@ class GetUsageAdapter(AuthenticatedApiAdapter):
     limit: int = 100
     offset: int = 0
 
+    @cache_result(
+        key_prefix="user:usage:records",
+        ttl=CacheTTL.ADMIN_USAGE_RECORDS,
+        user_specific=True,
+        vary_by=[
+            "time_range.start_date",
+            "time_range.end_date",
+            "time_range.preset",
+            "time_range.timezone",
+            "time_range.tz_offset_minutes",
+            "search",
+            "limit",
+            "offset",
+        ],
+    )
     async def handle(self, context: ApiRequestContext) -> Any:  # type: ignore[override]
         from sqlalchemy import or_
         from sqlalchemy.orm import load_only
@@ -954,17 +971,18 @@ class GetUsageAdapter(AuthenticatedApiAdapter):
             query.order_by(Usage.created_at.desc()).offset(self.offset).limit(self.limit).all()
         )
 
-        avg_resp_query = db.query(func.avg(Usage.response_time_ms)).filter(
+        # 复用 summary 聚合中的成功请求响应时间，避免额外 AVG SQL
-            Usage.user_id == user.id,
+        total_success_response_time_ms = sum(
-            Usage.status_code == 200,
+            float(item.get("success_response_time_sum_ms", 0.0) or 0.0) for item in summary_list
-            Usage.response_time_ms.isnot(None),
+        )
+        total_success_response_count = sum(
+            int(item.get("success_response_time_count", 0) or 0) for item in summary_list
+        )
+        avg_response_time = (
+            total_success_response_time_ms / total_success_response_count / 1000.0
+            if total_success_response_count > 0
+            else 0.0
         )
-        if start_utc and end_utc:
-            avg_resp_query = avg_resp_query.filter(
-                Usage.created_at >= start_utc, Usage.created_at < end_utc
-            )
-        avg_response_ms = avg_resp_query.scalar() or 0
-        avg_response_time = float(avg_response_ms) / 1000.0 if avg_response_ms else 0
 
         # 构建响应数据
         response_data = {
@@ -1100,6 +1118,12 @@ class GetMyIntervalTimelineAdapter(AuthenticatedApiAdapter):
     hours: int
     limit: int
 
+    @cache_result(
+        key_prefix="user:usage:interval_timeline",
+        ttl=CacheTTL.ADMIN_USAGE_AGGREGATION,
+        user_specific=True,
+        vary_by=["hours", "limit"],
+    )
     async def handle(self, context: ApiRequestContext) -> Any:  # type: ignore[override]
         db = context.db
         user = context.user

src/models/database.py

@@ -302,6 +302,7 @@ class Usage(Base):
     __table_args__ = (
         # Composite indexes for common query patterns (analytics / list pages)
         Index("idx_usage_user_created", "user_id", "created_at"),
+        Index("idx_usage_status_user_created", "status", "user_id", "created_at"),
         Index("idx_usage_apikey_created", "api_key_id", "created_at"),
         Index("idx_usage_provider_model_created", "provider_name", "model", "created_at"),
         Index("idx_usage_provider_created", "provider_name", "created_at"),

src/services/management_token/service.py

@@ -6,6 +6,7 @@ import ipaddress
 from datetime import datetime, timezone
 from typing import Any
 
+from sqlalchemy import func
 from sqlalchemy.exc import IntegrityError
 from sqlalchemy.orm import Session
 
@@ -159,7 +160,12 @@ class ManagementTokenService:
             ValueError: 如果名称已存在或超过数量限制
         """
         # 检查用户 Token 数量限制
-        token_count = db.query(ManagementToken).filter(ManagementToken.user_id == user_id).count()
+        token_count = int(
+            db.query(func.count(ManagementToken.id))
+            .filter(ManagementToken.user_id == user_id)
+            .scalar()
+            or 0
+        )
         max_tokens = config.management_token_max_per_user
         if token_count >= max_tokens:
             raise ValueError(f"已达到 Token 数量上限（{max_tokens}）")
@@ -245,7 +251,7 @@ class ManagementTokenService:
         if is_active is not None:
             query = query.filter(ManagementToken.is_active == is_active)
 
-        total = query.count()
+        total = int(query.with_entities(func.count(ManagementToken.id)).scalar() or 0)
         tokens = query.order_by(ManagementToken.created_at.desc()).offset(skip).limit(limit).all()
 
         return tokens, total

src/services/model/service.py

@@ -6,7 +6,7 @@ from __future__ import annotations
 
 import asyncio
 
-from sqlalchemy import and_
+from sqlalchemy import and_, func
 from sqlalchemy.exc import IntegrityError
 from sqlalchemy.orm import Session
 
@@ -255,14 +255,15 @@ class ModelService:
 
         # 检查这是否是该 GlobalModel 的最后一个关联提供商
         if model.global_model_id:
-            other_implementations = (
+            other_implementations = int(
-                db.query(Model)
+                db.query(func.count(Model.id))
                 .filter(
                     Model.global_model_id == model.global_model_id,
                     Model.id != model_id,
                     Model.is_active == True,
                 )
-                .count()
+                .scalar()
+                or 0
             )
 
             if other_implementations == 0:

src/services/proxy_node/service.py

@@ -14,7 +14,7 @@ from typing import Any
 from urllib.parse import urlparse
 
 import httpx
-from sqlalchemy import update
+from sqlalchemy import func, update
 from sqlalchemy.orm import Session
 
 from src.core.exceptions import InvalidRequestException, NotFoundException
@@ -452,7 +452,7 @@ class ProxyNodeService:
                 raise InvalidRequestException(f"status 必须是以下之一: {sorted(allowed)}", "status")
             query = query.filter(ProxyNode.status == ProxyNodeStatus(normalized))
 
-        total = query.count()
+        total = int(query.with_entities(func.count(ProxyNode.id)).scalar() or 0)
         nodes = query.order_by(ProxyNode.name.asc()).offset(skip).limit(limit).all()
         return nodes, total
 

src/services/system/announcement.py

@@ -6,7 +6,7 @@ from __future__ import annotations
 
 from datetime import datetime, timezone
 
-from sqlalchemy import or_
+from sqlalchemy import func, or_
 from sqlalchemy.orm import Session
 
 from src.core.exceptions import ForbiddenException, NotFoundException
@@ -87,7 +87,7 @@ class AnnouncementService:
         )
 
         # 分页
-        total = query.count()
+        total = int(query.with_entities(func.count(Announcement.id)).scalar() or 0)
         announcements = query.offset(offset).limit(limit).all()
 
         # 获取已读状态

src/services/system/audit.py

@@ -338,8 +338,8 @@ class AuditService:
         )
 
         # 获取最近的可疑活动
-        recent_suspicious = (
+        recent_suspicious = int(
-            db.query(AuditLog)
+            db.query(func.count(AuditLog.id))
             .filter(
                 AuditLog.user_id == user_id,
                 AuditLog.event_type.in_(
@@ -350,7 +350,8 @@ class AuditService:
                 ),
                 AuditLog.created_at >= cutoff_time,
             )
-            .count()
+            .scalar()
+            or 0
         )
 
         return {

src/services/system/config.py

@@ -116,8 +116,8 @@ class SystemConfigService:
             "description": "详细日志保留天数，超过此天数后压缩 request_body 和 response_body 到压缩字段",
         },
         "compressed_log_retention_days": {
-            "value": 90,
+            "value": 30,
-            "description": "压缩日志保留天数，超过此天数后删除压缩的 body 字段（保留headers和统计）",
+            "description": "压缩记录保留天数，超过此天数后删除压缩的 body 字段（保留headers和统计）",
         },
         "header_retention_days": {
             "value": 90,
@@ -125,7 +125,7 @@ class SystemConfigService:
         },
         "log_retention_days": {
             "value": 365,
-            "description": "完整日志保留天数，超过此天数后删除整条记录（保留核心统计）",
+            "description": "请求记录保存天数，超过此天数后删除整条使用记录",
         },
         "enable_auto_cleanup": {
             "value": True,

src/services/system/maintenance_scheduler.py

@@ -1052,7 +1052,7 @@ class MaintenanceScheduler:
             # 获取配置参数
             detail_retention = SystemConfigService.get_config(db, "detail_log_retention_days", 7)
             compressed_retention = SystemConfigService.get_config(
-                db, "compressed_log_retention_days", 90
+                db, "compressed_log_retention_days", 30
             )
             header_retention = SystemConfigService.get_config(db, "header_retention_days", 90)
             log_retention = SystemConfigService.get_config(db, "log_retention_days", 365)

src/services/usage/active_requests.py

@@ -3,6 +3,7 @@ from __future__ import annotations
 from datetime import datetime, timedelta, timezone
 from typing import Any
 
+from sqlalchemy import func
 from sqlalchemy.orm import Session
 
 from src.core.api_format.metadata import can_passthrough_endpoint
@@ -104,13 +105,14 @@ class UsageActiveRequestsMixin:
         """
         cutoff_time = datetime.now(timezone.utc) - timedelta(minutes=timeout_minutes)
 
-        return (
+        return int(
-            db.query(Usage)
+            db.query(func.count(Usage.id))
             .filter(
                 Usage.status.in_(["pending", "streaming"]),
                 Usage.created_at < cutoff_time,
             )
-            .count()
+            .scalar()
+            or 0
         )
 
     @classmethod

src/services/usage/query.py

@@ -3,7 +3,7 @@ from __future__ import annotations
 from datetime import datetime, timedelta, timezone
 from typing import Any
 
-from sqlalchemy import func
+from sqlalchemy import case, func
 from sqlalchemy.orm import Session
 
 from src.core.logger import logger
@@ -239,6 +239,21 @@ class UsageQueryMixin:
             func.sum(Usage.total_tokens).label("total_tokens"),
             func.sum(Usage.total_cost_usd).label("total_cost_usd"),
             func.avg(Usage.response_time_ms).label("avg_response_time"),
+            func.sum(
+                case(
+                    (
+                        (Usage.status_code == 200) & Usage.response_time_ms.isnot(None),
+                        Usage.response_time_ms,
+                    ),
+                    else_=0,
+                )
+            ).label("success_response_time_sum"),
+            func.sum(
+                case(
+                    ((Usage.status_code == 200) & Usage.response_time_ms.isnot(None), 1),
+                    else_=0,
+                )
+            ).label("success_response_time_count"),
         )
 
         # 过滤掉 pending/streaming 状态的请求（与上方明细查询一致）
@@ -268,6 +283,8 @@ class UsageQueryMixin:
                 "avg_response_time_ms": (
                     float(row.avg_response_time) if row.avg_response_time else 0
                 ),
+                "success_response_time_sum_ms": float(row.success_response_time_sum or 0.0),
+                "success_response_time_count": int(row.success_response_time_count or 0),
             }
             for row in summary
         ]

src/services/user/service.py

@@ -266,7 +266,9 @@ class UserService:
         db.query(AnnouncementRead).filter(AnnouncementRead.user_id == user_id).delete(
             synchronize_session=False
         )
-        api_key_count = db.query(ApiKey).filter(ApiKey.user_id == user_id).count()
+        api_key_count = int(
+            db.query(func.count(ApiKey.id)).filter(ApiKey.user_id == user_id).scalar() or 0
+        )
         db.query(ApiKey).filter(ApiKey.user_id == user_id).delete(synchronize_session=False)
 
         # 现在删除用户（Usage, AuditLog, RequestAttempt 会通过数据库 SET NULL 保留）

src/utils/cache_decorator.py

@@ -25,6 +25,22 @@ def _is_api_context(obj: Any) -> bool:
     return hasattr(obj, "user") and hasattr(obj, "db")
 
 
+def _resolve_attr(obj: Any, dotted_name: str) -> tuple[bool, Any]:
+    """Resolve a possibly dotted attribute path (e.g. 'time_range.start_date').
+
+    Returns (found, value).  When any segment along the chain is missing or
+    the intermediate value is None the lookup stops and returns (False, None).
+    """
+    current = obj
+    for part in dotted_name.split("."):
+        if current is None:
+            return False, None
+        if not hasattr(current, part):
+            return False, None
+        current = getattr(current, part)
+    return True, current
+
+
 def _hash_vary(vary: dict[str, Any]) -> str:
     """Build a short stable hash for cache key variations."""
     try:
@@ -91,8 +107,9 @@ def cache_result(
                     if vary_by:
                         vary: dict[str, Any] = {}
                         for attr_name in vary_by:
-                            if hasattr(adapter_self, attr_name):
+                            found, value = _resolve_attr(adapter_self, attr_name)
-                                vary[attr_name] = getattr(adapter_self, attr_name)
+                            if found:
+                                vary[attr_name] = value
                         if vary:
                             cache_key += f":v:{_hash_vary(vary)}"
                     else: