/** * GPU Instances Tab — per-instance compute usage view. * * Layout mirrors the existing Token tab: * 1. Top filter bar (date range + scope) * 2. KPI row (GPU-Hours / GPU-Minutes / Instances / GPU Types / Active Users) * 3. Daily bar chart with metric + group_by switches * 4. Bottom tab table grouped by GPU Type / Instance / User * * Talks to the new ``/usage/gpu-instances/{meta,breakdown}`` endpoints. */ import useCoolColors from '@/hooks/use-cool-colors'; import { formatLargeNumber } from '@/utils'; import { SimpleCard } from '@gpustack/core-ui'; import { useAccess, useIntl } from '@umijs/max'; import { Tabs } from 'antd'; import dayjs from 'dayjs'; import React, { useEffect, useMemo, useState } from 'react'; import { queryGpuInstancesBreakdown, ResourceBreakdownRequest } from '../apis/resource'; import MetricChartCard from '../components/metric-chart-card'; import MetricLabel from '../components/metric-label'; import ResourceExportData from '../components/resource-export-data'; import ResourceFilterBar from '../components/resource-filter-bar'; import useResourceMeta from '../hooks/use-resource-meta'; import { exportBreakdownSheets, toExportColumns } from '../utils/export-breakdown'; import { bucketKey, generateBucketRange, Granularity } from '../utils/time-buckets'; import { buildTrendSeries } from '../utils/trend-series'; import useQueryGpuInstancesBreakdown from './services/use-query-gpu-instances-breakdown'; import InstancesBreakdownTable from './tables/instances-breakdown-table'; import useInstancesColumns from './tables/use-instances-columns'; type Scope = 'self' | 'all'; type Metric = 'gpu_hours' | 'instance_hours'; type GroupKey = 'gpu_type' | 'instance' | 'user'; const GpuInstancesTab: React.FC = () => { const access = useAccess(); const intl = useIntl(); const METRIC_OPTIONS: { value: Metric; label: string }[] = useMemo( () => [ { value: 'gpu_hours', label: intl.formatMessage({ id: 'usage.metric.gpuHours' }) }, { value: 'instance_hours', label: intl.formatMessage({ id: 'usage.metric.instanceHours' }) } ], [intl] ); const TABLE_TABS: { key: GroupKey; label: string }[] = useMemo(() => { const tabs = [ { key: 'gpu_type' as GroupKey, label: intl.formatMessage({ id: 'usage.table.instanceTypes' }) }, { key: 'instance' as GroupKey, label: intl.formatMessage({ id: 'usage.table.instances' }) } ]; // Managers see the org-wide User breakdown; members only their own rows. if (access.canSeeOrgAdmin) { tabs.push({ key: 'user' as GroupKey, label: intl.formatMessage({ id: 'usage.table.users' }) }); } return tabs; }, [intl, access.canSeeOrgAdmin]); // ``useCoolColors`` returns a memoized factory; resolve a fixed 5-slot // palette for the KPI cards, and keep the factory for the grouped trend // (sized to the group count). const colorFactory = useCoolColors(); const coolColors = colorFactory(5); // No All/My dropdown (matches the Tokens tab): managers see the org-wide // view and narrow it with the user filter, others only their own rows. const canManageUsers = !!access.canSeeOrgAdmin; const scope: Scope = canManageUsers ? 'all' : 'self'; const [dateRange, setDateRange] = useState<[dayjs.Dayjs, dayjs.Dayjs]>([ dayjs().subtract(29, 'day'), dayjs() ]); const [selectedUsers, setSelectedUsers] = useState([]); const [selectedInstances, setSelectedInstances] = useState([]); const [refreshKey, setRefreshKey] = useState(0); const [metric, setMetric] = useState('gpu_hours'); const [granularity, setGranularity] = useState('day'); // Optional trend group-by (split the chart into one series per group). const [chartGroupBy, setChartGroupBy] = useState(null); // ``null`` group_by = no row grouping, just the summary KPIs. // The chart needs the ``date`` group; tables use the active table tab. const [activeTableTab, setActiveTableTab] = useState('gpu_type'); const { creators: userOptions, instances: instanceOptions } = useResourceMeta(scope); // The daily chart fetches group_by=date here; each bottom table owns its own // fetch (group_by=tab key) inside InstancesBreakdownTable. Bumped on any // filter change to snap every mounted table back to page 1. const [pageResetKey, setPageResetKey] = useState(0); const { detailData: chartData, loading: chartLoading, fetchData: fetchChartData } = useQueryGpuInstancesBreakdown({ key: 'gpuInstancesBreakdownChart' }); const baseRequest = (): Omit => ({ start_date: dateRange[0].format('YYYY-MM-DD'), end_date: dateRange[1].format('YYYY-MM-DD'), scope, granularity, filters: selectedUsers.length || selectedInstances.length ? { ...(selectedUsers.length ? { creator_ids: selectedUsers } : {}), ...(selectedInstances.length ? { instance_ids: selectedInstances } : {}) } : undefined, page: 1, perPage: 50 }); const fetchChart = () => fetchChartData({ ...baseRequest(), // Split each bucket by the chosen dimension when grouping. group_by: chartGroupBy ? ['date', chartGroupBy] : ['date'], // A trend is a time series, not a paginated table: always fetch the // whole range. The default order is metric-desc, so partial (current/ // recent) buckets have smaller values and would be pushed onto later // pages — dropping the newest hours from the chart under a small page. // ``page: -1`` is the backend's no-pagination sentinel. page: -1 }); useEffect(() => { fetchChart(); }, [ dateRange, selectedUsers, selectedInstances, granularity, chartGroupBy, refreshKey ]); // KPI summary cards — pull from the chart summary since both queries // return the same scope-wide totals. const summary = chartData?.summary; const summaryCards = useMemo( () => [ { label: formatLargeNumber( Math.round((summary?.gpu_hours ?? 0) * 10) / 10 ) as string, value: ( ), color: coolColors[0] }, { label: formatLargeNumber( Math.round((summary?.instance_hours ?? 0) * 10) / 10 ) as string, value: ( ), color: coolColors[1] }, { label: (summary?.active_instances ?? 0).toString(), value: intl.formatMessage({ id: 'usage.metric.activeInstances' }), color: coolColors[2] }, { label: (summary?.active_users ?? 0).toString(), value: intl.formatMessage({ id: 'usage.metric.activeUsers' }), color: coolColors[3] } ], [summary, coolColors, intl] ); // x-axis = the contiguous date range plus any buckets present in the data. const xAxis = useMemo(() => { const keys = new Set( generateBucketRange( dateRange[0].format('YYYY-MM-DD'), dateRange[1].format('YYYY-MM-DD'), granularity ) ); chartData?.items?.forEach((i) => { if (i.date) keys.add(bucketKey(i.date, granularity)); }); return Array.from(keys).sort(); }, [chartData, dateRange, granularity]); // Single series, or one stacked series per group when grouping is on. const seriesData = useMemo( () => buildTrendSeries({ items: chartData?.items, metric, granularity, xAxis, groupBy: chartGroupBy, palette: colorFactory, singleName: METRIC_OPTIONS.find((m) => m.value === metric)?.label || metric }), [chartData, metric, granularity, xAxis, chartGroupBy, colorFactory] ); // Group-by options for the trend = the same dimensions as the bottom tables // (TABLE_TABS already gates Users to org admins). const chartGroupByOptions = useMemo( () => TABLE_TABS.map((t) => ({ value: t.key, label: t.label })), [TABLE_TABS] ); // Columns for each bottom-table grouping — same factory the tables render // with — used to build the export sheets below. const gpuTypeColumns = useInstancesColumns('gpu_type'); const instanceColumns = useInstancesColumns('instance'); const userColumns = useInstancesColumns('user'); // "Export Table Data" writes every bottom table at once — one sheet per // grouping (Instance Types / Instances / Users) — straight to a workbook, // no preview dialog (mirrors the Tokens tab's `useExportTable`). The User // sheet is included only when the org-wide view is available. const tableExportGroups = useMemo(() => { const groups = [ { key: 'gpu_type' as GroupKey, columns: gpuTypeColumns, sheetName: intl.formatMessage({ id: 'usage.table.instanceTypes' }) }, { key: 'instance' as GroupKey, columns: instanceColumns, sheetName: intl.formatMessage({ id: 'usage.table.instances' }) } ]; if (canManageUsers) { groups.push({ key: 'user' as GroupKey, columns: userColumns, sheetName: intl.formatMessage({ id: 'usage.table.users' }) }); } return groups; }, [gpuTypeColumns, instanceColumns, userColumns, canManageUsers, intl]); // Chart export still opens the preview modal (matches the Tokens tab); the // table export is direct, so this only ever holds 'chart'. const [exportMode, setExportMode] = useState<'chart' | null>(null); const dateSuffix = `${dateRange[0].format('YYYY-MM-DD')}_${dateRange[1].format( 'YYYY-MM-DD' )}`; const handleExportTable = async () => { const results = await Promise.all( tableExportGroups.map((g) => queryGpuInstancesBreakdown({ ...baseRequest(), group_by: [g.key], // A breakdown export is the full filtered set, not a page. // ``page: -1`` is the backend's no-pagination sentinel. page: -1 }) ) ); exportBreakdownSheets( tableExportGroups.map((g, i) => ({ rows: results[i]?.items ?? [], columns: toExportColumns(g.columns), sheetName: g.sheetName })), `gpu-instances_tables_${dateSuffix}.xlsx` ); }; const chartExportColumns = [ { title: intl.formatMessage({ id: 'usage.table.date' }), dataIndex: 'date', key: 'date' }, { title: intl.formatMessage({ id: 'usage.metric.gpuHours' }), dataIndex: 'gpu_hours', key: 'gpu_hours', render: (v: number) => (v ?? 0).toFixed(2) }, { title: intl.formatMessage({ id: 'usage.metric.instanceHours' }), dataIndex: 'instance_hours', key: 'instance_hours', render: (v: number) => (v ?? 0).toFixed(2) }, { title: intl.formatMessage({ id: 'usage.metric.activeInstances' }), dataIndex: 'active_instances', key: 'active_instances' }, { title: intl.formatMessage({ id: 'usage.metric.activeUsers' }), dataIndex: 'active_users', key: 'active_users' } ]; // The preview modal now only backs the by-date chart export. const exportConfig = { groupBy: ['date'], columns: chartExportColumns, fileName: `gpu-instances_chart_${dateSuffix}.xlsx`, sheetName: intl.formatMessage({ id: 'usage.tabs.gpuInstances' }) }; return (
{/* Top filter row */} { setDateRange(dates); setPageResetKey((k) => k + 1); }} canManageUsers={canManageUsers} userOptions={userOptions} selectedUsers={selectedUsers} onUsersChange={(ids) => { setSelectedUsers(ids); setPageResetKey((k) => k + 1); }} resourceFilter={{ options: instanceOptions, value: selectedInstances, onChange: (ids) => { setSelectedInstances(ids); setPageResetKey((k) => k + 1); }, placeholder: intl.formatMessage({ id: 'usage.filter.instance' }) }} onRefresh={() => setRefreshKey((k) => k + 1)} onExportChart={() => setExportMode('chart')} onExportTable={handleExportTable} /> {/* KPI cards */}
{/* Daily trend chart */}
setMetric(v as Metric)} onGranularityChange={(v) => setGranularity(v as Granularity)} seriesData={seriesData} xAxisData={xAxis} groupBy={chartGroupBy} groupByOptions={chartGroupByOptions} onGroupByChange={(v) => setChartGroupBy(v as GroupKey | null)} loading={chartLoading} />
{/* Bottom tabs + table */} setActiveTableTab(k as GroupKey)} items={TABLE_TABS.map((t) => ({ key: t.key, label: t.label, // Keep every pane mounted so each table holds its own page/sort and // switching tabs neither refetches nor resets the others. forceRender: true, children: ( ) }))} /> setExportMode(null)} title={intl.formatMessage({ id: 'usage.export.chart' })} queryFn={queryGpuInstancesBreakdown} groupBy={exportConfig.groupBy} columns={exportConfig.columns} fileName={exportConfig.fileName} sheetName={exportConfig.sheetName} scope={scope} canManageUsers={canManageUsers} userOptions={userOptions} resourceFilter={{ options: instanceOptions, placeholder: intl.formatMessage({ id: 'usage.filter.instance' }), key: 'instance_ids' }} initialDateRange={dateRange} initialSelectedUsers={selectedUsers} initialSelectedResources={selectedInstances} />
); }; export default GpuInstancesTab;