cc-backend/web/frontend/src/Analysis.root.svelte

266 lines
9.8 KiB
Svelte
Raw Normal View History

2022-06-22 11:20:57 +02:00
<script>
import { init } from './utils.js'
import { getContext, onMount } from 'svelte'
import { operationStore, query } from '@urql/svelte'
import { Row, Col, Spinner, Card, Table } from 'sveltestrap'
import Filters from './filters/Filters.svelte'
import PlotSelection from './PlotSelection.svelte'
import Histogram, { binsFromFootprint } from './plots/Histogram.svelte'
import ScatterPlot from './plots/Scatter.svelte'
import PlotTable from './PlotTable.svelte'
import Roofline from './plots/Roofline.svelte'
const { query: initq } = init()
export let filterPresets
// By default, look at the jobs of the last 6 hours:
if (filterPresets?.startTime == null) {
if (filterPresets == null)
filterPresets = {}
let now = new Date(Date.now())
let hourAgo = new Date(now)
hourAgo.setHours(hourAgo.getHours() - 6)
filterPresets.startTime = { from: hourAgo.toISOString(), to: now.toISOString() }
}
let cluster
let filters
let rooflineMaxY
let colWidth
let numBins = 50
const ccconfig = getContext('cc-config'),
metricConfig = getContext('metrics')
let metricsInHistograms = ccconfig.analysis_view_histogramMetrics,
metricsInScatterplots = ccconfig.analysis_view_scatterPlotMetrics
$: metrics = [...new Set([...metricsInHistograms, ...metricsInScatterplots.flat()])]
getContext('on-init')(({ data }) => {
if (data != null) {
cluster = data.clusters.find(c => c.name == filterPresets.cluster)
console.assert(cluster != null, `This cluster could not be found: ${filterPresets.cluster}`)
rooflineMaxY = cluster.subClusters.reduce((max, part) => Math.max(max, part.flopRateSimd), 0)
$rooflineQuery.variables.maxY = rooflineMaxY
$rooflineQuery.context.pause = false
$rooflineQuery.reexecute()
}
})
const statsQuery = operationStore(`
query($filter: [JobFilter!]!) {
stats: jobsStatistics(filter: $filter) {
totalJobs
shortJobs
totalWalltime
totalCoreHours
histDuration { count, value }
histNumNodes { count, value }
}
topUsers: jobsCount(filter: $filter, groupBy: USER, weight: NODE_HOURS, limit: 5) { name, count }
}
`, { filter: [] }, { pause: true })
const footprintsQuery = operationStore(`
query($filter: [JobFilter!]!, $metrics: [String!]!) {
footprints: jobsFootprints(filter: $filter, metrics: $metrics) {
nodehours,
metrics { metric, data }
}
}
`, { filter: [], metrics }, { pause: true })
$: $footprintsQuery.variables = { ...$footprintsQuery.variables, metrics }
const rooflineQuery = operationStore(`
query($filter: [JobFilter!]!, $rows: Int!, $cols: Int!,
$minX: Float!, $minY: Float!, $maxX: Float!, $maxY: Float!) {
rooflineHeatmap(filter: $filter, rows: $rows, cols: $cols,
minX: $minX, minY: $minY, maxX: $maxX, maxY: $maxY)
}
`, {
filter: [],
rows: 50, cols: 50,
minX: 0.01, minY: 1., maxX: 1000., maxY: -1
}, { pause: true });
query(statsQuery)
query(footprintsQuery)
query(rooflineQuery)
onMount(() => filters.update())
</script>
<Row>
{#if $initq.fetching || $statsQuery.fetching || $footprintsQuery.fetching}
<Col xs="auto">
<Spinner />
</Col>
{/if}
<Col xs="auto">
{#if $initq.error}
<Card body color="danger">{$initq.error.message}</Card>
{:else if cluster}
<PlotSelection
availableMetrics={cluster.metricConfig.map(mc => mc.name)}
bind:metricsInHistograms={metricsInHistograms}
bind:metricsInScatterplots={metricsInScatterplots} />
{/if}
</Col>
<Col xs="auto">
<Filters
bind:this={filters}
filterPresets={filterPresets}
disableClusterSelection={true}
startTimeQuickSelect={true}
on:update={({ detail }) => {
$statsQuery.context.pause = false
$statsQuery.variables = { filter: detail.filters }
$footprintsQuery.context.pause = false
$footprintsQuery.variables = { metrics, filter: detail.filters }
$rooflineQuery.variables = { ...$rooflineQuery.variables, filter: detail.filters }
}} />
</Col>
</Row>
<br/>
{#if $statsQuery.error}
<Row>
<Col>
<Card body color="danger">{$statsQuery.error.message}</Card>
</Col>
</Row>
{:else if $statsQuery.data}
<Row>
<div class="col-3" bind:clientWidth={colWidth}>
<div style="height: 40%">
<Table>
<tr>
<th scope="col">Total Jobs</th>
<td>{$statsQuery.data.stats[0].totalJobs}</td>
</tr>
<tr>
<th scope="col">Short Jobs (&#60; 2m)</th>
<td>{$statsQuery.data.stats[0].shortJobs}</td>
</tr>
<tr>
<th scope="col">Total Walltime</th>
<td>{$statsQuery.data.stats[0].totalWalltime}</td>
</tr>
<tr>
<th scope="col">Total Core Hours</th>
<td>{$statsQuery.data.stats[0].totalCoreHours}</td>
</tr>
</Table>
</div>
<div style="height: 60%;">
{#key $statsQuery.data.topUsers}
<h4>Top Users (by node hours)</h4>
<Histogram
width={colWidth - 25} height={300 * 0.5}
data={$statsQuery.data.topUsers.sort((a, b) => b.count - a.count).map(({ count }, idx) => ({ count, value: idx }))}
label={(x) => x < $statsQuery.data.topUsers.length ? $statsQuery.data.topUsers[Math.floor(x)].name : '0'} />
{/key}
</div>
</div>
<div class="col-3">
{#key $statsQuery.data.stats[0].histDuration}
<h4>Walltime Distribution</h4>
<Histogram
width={colWidth - 25} height={300}
data={$statsQuery.data.stats[0].histDuration} />
{/key}
</div>
<div class="col-3">
{#key $statsQuery.data.stats[0].histNumNodes}
<h4>Number of Nodes Distribution</h4>
<Histogram
width={colWidth - 25} height={300}
data={$statsQuery.data.stats[0].histNumNodes} />
{/key}
</div>
<div class="col-3">
{#if $rooflineQuery.fetching}
<Spinner />
{:else if $rooflineQuery.error}
<Card body color="danger">{$rooflineQuery.error.message}</Card>
{:else if $rooflineQuery.data && cluster}
{#key $rooflineQuery.data}
<Roofline
width={colWidth - 25} height={300}
tiles={$rooflineQuery.data.rooflineHeatmap}
cluster={cluster.subClusters.length == 1 ? cluster.subClusters[0] : null}
maxY={rooflineMaxY} />
{/key}
{/if}
</div>
</Row>
{/if}
<br/>
{#if $footprintsQuery.error}
<Row>
<Col>
<Card body color="danger">{$footprintsQuery.error.message}</Card>
</Col>
</Row>
{:else if $footprintsQuery.data && $initq.data}
<Row>
<Col>
<Card body>
These histograms show the distribution of the averages of all jobs matching the filters. Each job/average is weighted by its node hours.
</Card>
<br/>
</Col>
</Row>
<Row>
<Col>
<PlotTable
let:item
let:width
items={metricsInHistograms.map(metric => ({ metric, ...binsFromFootprint(
$footprintsQuery.data.footprints.nodehours,
$footprintsQuery.data.footprints.metrics.find(f => f.metric == metric).data, numBins) }))}
itemsPerRow={ccconfig.plot_view_plotsPerRow}>
<h4>{item.metric} [{metricConfig(cluster.name, item.metric)?.unit}]</h4>
<Histogram
width={width} height={250}
min={item.min} max={item.max}
data={item.bins} label={item.label} />
</PlotTable>
</Col>
</Row>
<br/>
<Row>
<Col>
<Card body>
Each circle represents one job. The size of a circle is proportional to its node hours. Darker circles mean multiple jobs have the same averages for the respective metrics.
</Card>
<br/>
</Col>
</Row>
<Row>
<Col>
<PlotTable
let:item
let:width
items={metricsInScatterplots.map(([m1, m2]) => ({
m1, f1: $footprintsQuery.data.footprints.metrics.find(f => f.metric == m1).data,
m2, f2: $footprintsQuery.data.footprints.metrics.find(f => f.metric == m2).data }))}
itemsPerRow={ccconfig.plot_view_plotsPerRow}>
<ScatterPlot
width={width} height={250} color={"rgba(0, 102, 204, 0.33)"}
xLabel={`${item.m1} [${metricConfig(cluster.name, item.m1)?.unit}]`}
yLabel={`${item.m2} [${metricConfig(cluster.name, item.m2)?.unit}]`}
X={item.f1} Y={item.f2} S={$footprintsQuery.data.footprints.nodehours} />
</PlotTable>
</Col>
</Row>
{/if}