perf(usage): 优化 admin usage records 查询性能

- count 查询按需 JOIN，避免不必要的表关联
- 前端 onMounted 将 stats/heatmap/records/users 全部并行加载
- 调整缓存 TTL（聚合 30s->60s，列表 10s->15s）
- 为 request_candidates 添加复合索引优化 fallback/retry 查询

alembic/versions/20260228_1448_00b9161b8729_add_request_candidates_composite_indexes.py

@@ -0,0 +1,49 @@
+"""add_request_candidates_composite_indexes
+
+Revision ID: 00b9161b8729
+Revises: 48afe197cc15
+Create Date: 2026-02-28 14:48:00.000000+00:00
+
+"""
+
+from sqlalchemy import inspect
+
+from alembic import op
+
+# revision identifiers, used by Alembic.
+revision = "00b9161b8729"
+down_revision = "48afe197cc15"
+branch_labels = None
+depends_on = None
+
+
+def _index_exists(index_name: str) -> bool:
+    bind = op.get_bind()
+    insp = inspect(bind)
+    indexes = insp.get_indexes("request_candidates")
+    return any(idx["name"] == index_name for idx in indexes)
+
+
+def upgrade() -> None:
+    # (request_id, status) - fallback/retry 查询优化
+    if not _index_exists("idx_rc_request_id_status"):
+        op.create_index(
+            "idx_rc_request_id_status",
+            "request_candidates",
+            ["request_id", "status"],
+        )
+
+    # (provider_id, status, created_at) - provider 聚合统计优化
+    if not _index_exists("idx_rc_provider_status_created"):
+        op.create_index(
+            "idx_rc_provider_status_created",
+            "request_candidates",
+            ["provider_id", "status", "created_at"],
+        )
+
+
+def downgrade() -> None:
+    if _index_exists("idx_rc_provider_status_created"):
+        op.drop_index("idx_rc_provider_status_created", table_name="request_candidates")
+    if _index_exists("idx_rc_request_id_status"):
+        op.drop_index("idx_rc_request_id_status", table_name="request_candidates")

frontend/src/views/shared/Usage.vue

@@ -411,34 +411,30 @@ const selectedRequestId = ref<string | null>(null)
 
 // 初始化加载
 onMounted(async () => {
-  // 并行加载统计数据和热力图（使用 allSettled 避免其中一个失败影响另一个）
-  const [statsResult, heatmapResult] = await Promise.allSettled([
-    loadStats(timeRange.value),
-    loadHeatmapData()
-  ])
-
-  // 检查加载结果并通知用户
-  if (statsResult.status === 'rejected') {
-    log.error('加载统计数据失败:', statsResult.reason)
+  // 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)
     warning('统计数据加载失败，请刷新重试')
-  }
-  if (heatmapResult.status === 'rejected') {
-    log.error('加载热力图数据失败:', heatmapResult.reason)
-    // 热力图加载失败不提示，因为 UI 已显示占位符
+  })
+  const heatmapTask = loadHeatmapData().catch(err => {
+    log.error('加载热力图数据失败:', err)
+  })
+  const recordsTask = loadRecords(
+    { page: currentPage.value, pageSize: pageSize.value },
+    getCurrentFilters()
+  )
+
+  const tasks: Promise<unknown>[] = [statsTask, heatmapTask, recordsTask]
+
+  if (isAdminPage.value) {
+    tasks.push(
+      usersApi.getAllUsers().then(users => {
+        availableUsers.value = users.map(u => ({ id: u.id, username: u.username, email: u.email }))
+      })
+    )
   }
 
-  // 加载记录和用户列表
-  if (isAdminPage.value) {
-    // 管理员页面：并行加载用户列表和记录
-    const [users] = await Promise.all([
-      usersApi.getAllUsers(),
-      loadRecords({ page: currentPage.value, pageSize: pageSize.value }, getCurrentFilters())
-    ])
-    availableUsers.value = users.map(u => ({ id: u.id, username: u.username, email: u.email }))
-  } else {
-    // 用户页面：加载记录
-    await loadRecords({ page: currentPage.value, pageSize: pageSize.value }, getCurrentFilters())
-  }
+  await Promise.allSettled(tasks)
 })
 
 // 处理时间范围变化

src/api/admin/usage/routes.py

@@ -869,6 +869,21 @@ class AdminUsageRecordsAdapter(AdminApiAdapter):
         from src.utils.database_helpers import escape_like_pattern, safe_truncate_escaped
 
         db = context.db
+
+        # -- 构建轻量 count 查询（仅按需 JOIN） --
+        needs_user_join = bool(self.search or self.username)
+        needs_provider_join = bool(self.search or self.provider)
+        needs_apikey_join = bool(self.search)
+
+        count_query = db.query(func.count(Usage.id))
+        if needs_user_join:
+            count_query = count_query.outerjoin(User, Usage.user_id == User.id)
+        if needs_provider_join:
+            count_query = count_query.outerjoin(Provider, Usage.provider_id == Provider.id)
+        if needs_apikey_join:
+            count_query = count_query.outerjoin(ApiKey, Usage.api_key_id == ApiKey.id)
+
+        # -- 构建数据查询（完整 JOIN） --
         query = (
             db.query(Usage, User, ProviderEndpoint, ProviderAPIKey, ApiKey)
             .outerjoin(User, Usage.user_id == User.id)
@@ -889,57 +904,65 @@ class AdminUsageRecordsAdapter(AdminApiAdapter):
             for keyword in keywords:
                 escaped = safe_truncate_escaped(escape_like_pattern(keyword), 100)
                 search_pattern = f"%{escaped}%"
-                query = query.filter(
-                    or_(
-                        User.username.ilike(search_pattern, escape="\\"),
-                        ApiKey.name.ilike(search_pattern, escape="\\"),
-                        Usage.model.ilike(search_pattern, escape="\\"),
-                        Provider.name.ilike(search_pattern, escape="\\"),
-                    )
+                search_filter = or_(
+                    User.username.ilike(search_pattern, escape="\\"),
+                    ApiKey.name.ilike(search_pattern, escape="\\"),
+                    Usage.model.ilike(search_pattern, escape="\\"),
+                    Provider.name.ilike(search_pattern, escape="\\"),
                 )
+                query = query.filter(search_filter)
+                count_query = count_query.filter(search_filter)
 
         if self.user_id:
             query = query.filter(Usage.user_id == self.user_id)
+            count_query = count_query.filter(Usage.user_id == self.user_id)
         if self.username:
             # 支持用户名模糊搜索
             escaped = escape_like_pattern(self.username)
-            query = query.filter(User.username.ilike(f"%{escaped}%", escape="\\"))
+            username_filter = User.username.ilike(f"%{escaped}%", escape="\\")
+            query = query.filter(username_filter)
+            count_query = count_query.filter(username_filter)
         if self.model:
             # 模型筛选：前端为下拉框精确值，使用精确匹配以启用索引
             # 如需模糊搜索，请使用 search 参数。
             query = query.filter(Usage.model == self.model)
+            count_query = count_query.filter(Usage.model == self.model)
         if self.provider:
             # 提供商筛选：前端为下拉框精确值，使用精确匹配以启用索引
             # 如需模糊搜索，请使用 search 参数。
             query = query.filter(Provider.name == self.provider)
+            count_query = count_query.filter(Provider.name == self.provider)
         if self.api_format:
             # API 格式筛选：精确匹配（大小写不敏感）
-            query = query.filter(func.lower(Usage.api_format) == self.api_format.lower())
+            api_format_filter = func.lower(Usage.api_format) == self.api_format.lower()
+            query = query.filter(api_format_filter)
+            count_query = count_query.filter(api_format_filter)
         if self.status:
             # 状态筛选
             # 旧的筛选值（基于 is_stream 和 status_code）：stream, standard, error
             # 新的筛选值（基于 status 字段）：pending, streaming, completed, failed, active
+            status_filter = None
             if self.status == "stream":
-                query = query.filter(Usage.is_stream == True)  # noqa: E712
+                status_filter = Usage.is_stream == True  # noqa: E712
             elif self.status == "standard":
-                query = query.filter(Usage.is_stream == False)  # noqa: E712
+                status_filter = Usage.is_stream == False  # noqa: E712
             elif self.status == "error":
-                query = query.filter((Usage.status_code >= 400) | (Usage.error_message.isnot(None)))
+                status_filter = (Usage.status_code >= 400) | (Usage.error_message.isnot(None))
             elif self.status in ("pending", "streaming", "completed", "cancelled"):
                 # 新的状态筛选：直接按 status 字段过滤
-                query = query.filter(Usage.status == self.status)
+                status_filter = Usage.status == self.status
             elif self.status == "failed":
                 # 失败请求需要同时考虑新旧两种判断方式：
                 # 1. 新方式：status = "failed"
                 # 2. 旧方式：status_code >= 400 或 error_message 不为空
-                query = query.filter(
+                status_filter = (
                     (Usage.status == "failed")
                     | (Usage.status_code >= 400)
                     | (Usage.error_message.isnot(None))
                 )
             elif self.status == "active":
                 # 活跃请求：pending 或 streaming 状态
-                query = query.filter(Usage.status.in_(["pending", "streaming"]))
+                status_filter = Usage.status.in_(["pending", "streaming"])
             elif self.status == "has_retry":
                 # 发生重试：存在 retry_index > 0 的已执行候选
                 retry_subq = (
@@ -951,7 +974,7 @@ class AdminUsageRecordsAdapter(AdminApiAdapter):
                     .distinct()
                     .subquery()
                 )
-                query = query.filter(Usage.request_id.in_(retry_subq))
+                status_filter = Usage.request_id.in_(retry_subq)
             elif self.status == "has_fallback":
                 # 发生转移：同一请求有多个不同 candidate_index 的已执行候选
                 fallback_subq = (
@@ -961,13 +984,21 @@ class AdminUsageRecordsAdapter(AdminApiAdapter):
                     .having(func.count(func.distinct(RequestCandidate.candidate_index)) > 1)
                     .subquery()
                 )
-                query = query.filter(Usage.request_id.in_(fallback_subq))
+                status_filter = Usage.request_id.in_(fallback_subq)
+
+            if status_filter is not None:
+                query = query.filter(status_filter)
+                count_query = count_query.filter(status_filter)
+
         if self.time_range:
             start_utc, end_utc = self.time_range.to_utc_datetime_range()
-            query = query.filter(Usage.created_at >= start_utc, Usage.created_at < end_utc)
+            time_filter_start = Usage.created_at >= start_utc
+            time_filter_end = Usage.created_at < end_utc
+            query = query.filter(time_filter_start, time_filter_end)
+            count_query = count_query.filter(time_filter_start, time_filter_end)
 
-        # Perf: avoid Query.count() building a subquery selecting many columns
-        total = int(query.with_entities(func.count(Usage.id)).scalar() or 0)
+        # Perf: count query uses fewer JOINs than the data query
+        total = int(count_query.scalar() or 0)
 
         # Perf: do not load large request/response columns for list view
         query = query.options(

src/config/constants.py

@@ -29,8 +29,8 @@ class CacheTTL:
     DASHBOARD_DAILY = 600  # 10分钟（每日统计）
 
     # Admin usage pages (heavy DB aggregations / list queries)
-    ADMIN_USAGE_AGGREGATION = 30  # 30秒
-    ADMIN_USAGE_RECORDS = 10  # 10秒（列表页短缓存，避免轮询/重复刷新打爆 DB）
+    ADMIN_USAGE_AGGREGATION = 60  # 60秒（聚合统计变化不频繁，适当延长减少 DB 压力）
+    ADMIN_USAGE_RECORDS = 15  # 15秒（列表页短缓存，活跃请求通过轮询接口实时更新）
 
     # Admin leaderboard (heavier, slower moving)
     ADMIN_LEADERBOARD = 300  # 5分钟

src/models/database.py

@@ -2203,6 +2203,10 @@ class RequestCandidate(Base):
         Index("idx_request_candidates_status", "status"),
         Index("idx_request_candidates_provider_id", "provider_id"),
         Index("idx_request_candidates_created_at", "created_at"),
+        # 复合索引: 按 request_id + status 查询 fallback/retry 场景
+        Index("idx_rc_request_id_status", "request_id", "status"),
+        # 复合索引: 按 provider 聚合统计（provider_id + status + created_at）
+        Index("idx_rc_provider_status_created", "provider_id", "status", "created_at"),
     )
 
     # 关系