add resource compare graph, add cursor sync, handle jobIds fitler

This commit is contained in:
Christoph Kluge 2025-05-06 17:54:13 +02:00
parent d3d752f90c
commit aed2bd48fc
7 changed files with 324 additions and 101 deletions

View File

@ -174,6 +174,9 @@ type JobStats {
jobId: Int!
startTime: Int!
duration: Int!
numNodes: Int!
numHWThreads: Int
numAccelerators: Int
stats: [NamedStats!]!
}

View File

@ -173,6 +173,9 @@ type ComplexityRoot struct {
JobStats struct {
Duration func(childComplexity int) int
JobID func(childComplexity int) int
NumAccelerators func(childComplexity int) int
NumHWThreads func(childComplexity int) int
NumNodes func(childComplexity int) int
StartTime func(childComplexity int) int
Stats func(childComplexity int) int
}
@ -956,6 +959,27 @@ func (e *executableSchema) Complexity(typeName, field string, childComplexity in
return e.complexity.JobStats.JobID(childComplexity), true
case "JobStats.numAccelerators":
if e.complexity.JobStats.NumAccelerators == nil {
break
}
return e.complexity.JobStats.NumAccelerators(childComplexity), true
case "JobStats.numHWThreads":
if e.complexity.JobStats.NumHWThreads == nil {
break
}
return e.complexity.JobStats.NumHWThreads(childComplexity), true
case "JobStats.numNodes":
if e.complexity.JobStats.NumNodes == nil {
break
}
return e.complexity.JobStats.NumNodes(childComplexity), true
case "JobStats.startTime":
if e.complexity.JobStats.StartTime == nil {
break
@ -2299,6 +2323,9 @@ type JobStats {
jobId: Int!
startTime: Int!
duration: Int!
numNodes: Int!
numHWThreads: Int
numAccelerators: Int
stats: [NamedStats!]!
}
@ -7506,6 +7533,132 @@ func (ec *executionContext) fieldContext_JobStats_duration(_ context.Context, fi
return fc, nil
}
func (ec *executionContext) _JobStats_numNodes(ctx context.Context, field graphql.CollectedField, obj *model.JobStats) (ret graphql.Marshaler) {
fc, err := ec.fieldContext_JobStats_numNodes(ctx, field)
if err != nil {
return graphql.Null
}
ctx = graphql.WithFieldContext(ctx, fc)
defer func() {
if r := recover(); r != nil {
ec.Error(ctx, ec.Recover(ctx, r))
ret = graphql.Null
}
}()
resTmp, err := ec.ResolverMiddleware(ctx, func(rctx context.Context) (any, error) {
ctx = rctx // use context from middleware stack in children
return obj.NumNodes, nil
})
if err != nil {
ec.Error(ctx, err)
return graphql.Null
}
if resTmp == nil {
if !graphql.HasFieldError(ctx, fc) {
ec.Errorf(ctx, "must not be null")
}
return graphql.Null
}
res := resTmp.(int)
fc.Result = res
return ec.marshalNInt2int(ctx, field.Selections, res)
}
func (ec *executionContext) fieldContext_JobStats_numNodes(_ context.Context, field graphql.CollectedField) (fc *graphql.FieldContext, err error) {
fc = &graphql.FieldContext{
Object: "JobStats",
Field: field,
IsMethod: false,
IsResolver: false,
Child: func(ctx context.Context, field graphql.CollectedField) (*graphql.FieldContext, error) {
return nil, errors.New("field of type Int does not have child fields")
},
}
return fc, nil
}
func (ec *executionContext) _JobStats_numHWThreads(ctx context.Context, field graphql.CollectedField, obj *model.JobStats) (ret graphql.Marshaler) {
fc, err := ec.fieldContext_JobStats_numHWThreads(ctx, field)
if err != nil {
return graphql.Null
}
ctx = graphql.WithFieldContext(ctx, fc)
defer func() {
if r := recover(); r != nil {
ec.Error(ctx, ec.Recover(ctx, r))
ret = graphql.Null
}
}()
resTmp, err := ec.ResolverMiddleware(ctx, func(rctx context.Context) (any, error) {
ctx = rctx // use context from middleware stack in children
return obj.NumHWThreads, nil
})
if err != nil {
ec.Error(ctx, err)
return graphql.Null
}
if resTmp == nil {
return graphql.Null
}
res := resTmp.(*int)
fc.Result = res
return ec.marshalOInt2ᚖint(ctx, field.Selections, res)
}
func (ec *executionContext) fieldContext_JobStats_numHWThreads(_ context.Context, field graphql.CollectedField) (fc *graphql.FieldContext, err error) {
fc = &graphql.FieldContext{
Object: "JobStats",
Field: field,
IsMethod: false,
IsResolver: false,
Child: func(ctx context.Context, field graphql.CollectedField) (*graphql.FieldContext, error) {
return nil, errors.New("field of type Int does not have child fields")
},
}
return fc, nil
}
func (ec *executionContext) _JobStats_numAccelerators(ctx context.Context, field graphql.CollectedField, obj *model.JobStats) (ret graphql.Marshaler) {
fc, err := ec.fieldContext_JobStats_numAccelerators(ctx, field)
if err != nil {
return graphql.Null
}
ctx = graphql.WithFieldContext(ctx, fc)
defer func() {
if r := recover(); r != nil {
ec.Error(ctx, ec.Recover(ctx, r))
ret = graphql.Null
}
}()
resTmp, err := ec.ResolverMiddleware(ctx, func(rctx context.Context) (any, error) {
ctx = rctx // use context from middleware stack in children
return obj.NumAccelerators, nil
})
if err != nil {
ec.Error(ctx, err)
return graphql.Null
}
if resTmp == nil {
return graphql.Null
}
res := resTmp.(*int)
fc.Result = res
return ec.marshalOInt2ᚖint(ctx, field.Selections, res)
}
func (ec *executionContext) fieldContext_JobStats_numAccelerators(_ context.Context, field graphql.CollectedField) (fc *graphql.FieldContext, err error) {
fc = &graphql.FieldContext{
Object: "JobStats",
Field: field,
IsMethod: false,
IsResolver: false,
Child: func(ctx context.Context, field graphql.CollectedField) (*graphql.FieldContext, error) {
return nil, errors.New("field of type Int does not have child fields")
},
}
return fc, nil
}
func (ec *executionContext) _JobStats_stats(ctx context.Context, field graphql.CollectedField, obj *model.JobStats) (ret graphql.Marshaler) {
fc, err := ec.fieldContext_JobStats_stats(ctx, field)
if err != nil {
@ -11307,6 +11460,12 @@ func (ec *executionContext) fieldContext_Query_jobsMetricStats(ctx context.Conte
return ec.fieldContext_JobStats_startTime(ctx, field)
case "duration":
return ec.fieldContext_JobStats_duration(ctx, field)
case "numNodes":
return ec.fieldContext_JobStats_numNodes(ctx, field)
case "numHWThreads":
return ec.fieldContext_JobStats_numHWThreads(ctx, field)
case "numAccelerators":
return ec.fieldContext_JobStats_numAccelerators(ctx, field)
case "stats":
return ec.fieldContext_JobStats_stats(ctx, field)
}
@ -17647,6 +17806,15 @@ func (ec *executionContext) _JobStats(ctx context.Context, sel ast.SelectionSet,
if out.Values[i] == graphql.Null {
out.Invalids++
}
case "numNodes":
out.Values[i] = ec._JobStats_numNodes(ctx, field, obj)
if out.Values[i] == graphql.Null {
out.Invalids++
}
case "numHWThreads":
out.Values[i] = ec._JobStats_numHWThreads(ctx, field, obj)
case "numAccelerators":
out.Values[i] = ec._JobStats_numAccelerators(ctx, field, obj)
case "stats":
out.Values[i] = ec._JobStats_stats(ctx, field, obj)
if out.Values[i] == graphql.Null {

View File

@ -100,6 +100,9 @@ type JobStats struct {
JobID int `json:"jobId"`
StartTime int `json:"startTime"`
Duration int `json:"duration"`
NumNodes int `json:"numNodes"`
NumHWThreads *int `json:"numHWThreads,omitempty"`
NumAccelerators *int `json:"numAccelerators,omitempty"`
Stats []*NamedStats `json:"stats"`
}

View File

@ -615,10 +615,15 @@ func (r *queryResolver) JobsMetricStats(ctx context.Context, filter []*model.Job
})
}
numThreadsInt := int(job.NumHWThreads)
numAccsInt := int(job.NumAcc)
res = append(res, &model.JobStats{
JobID: int(job.JobID),
StartTime: int(job.StartTime.Unix()),
Duration: int(job.Duration),
NumNodes: int(job.NumNodes),
NumHWThreads: &numThreadsInt,
NumAccelerators: &numAccsInt,
Stats: sres,
})
}

View File

@ -149,6 +149,14 @@ func BuildWhereClause(filter *model.JobFilter, query sq.SelectBuilder) sq.Select
if filter.JobID != nil {
query = buildStringCondition("job.job_id", filter.JobID, query)
}
if filter.JobIds != nil {
jobIds := make([]string, len(filter.JobIds))
for i, val := range filter.JobIds {
jobIds[i] = string(val)
}
query = query.Where(sq.Eq{"job.job_id": jobIds})
}
if filter.ArrayJobID != nil {
query = query.Where("job.array_job_id = ?", *filter.ArrayJobID)
}

View File

@ -15,6 +15,7 @@
<script>
import { getContext } from "svelte";
import uPlot from "uplot";
import {
queryStore,
gql,
@ -40,10 +41,11 @@
let filter = [...filterBuffer];
let comparePlotData = {};
let jobIds = [];
const sorting = { field: "startTime", type: "col", order: "DESC" };
/*uPlot*/
let plotSync = uPlot.sync("compareJobsView");
/* GQL */
const client = getContextClient();
// Pull All Series For Metrics Statistics Only On Node Scope
const compareQuery = gql`
@ -52,6 +54,9 @@
jobId
startTime
duration
numNodes
numHWThreads
numAccelerators
stats {
name
data {
@ -111,11 +116,13 @@
function jobs2uplot(jobs, metrics) {
// Prep
// Resources Init
comparePlotData['resources'] = {unit:'', data: [[],[],[],[],[],[]]} // data: [X, XST, XRT, YNODES, YTHREADS, YACCS]
// Metric Init
for (let m of metrics) {
// Get Unit
const rawUnit = globalMetrics.find((gm) => gm.name == m)?.unit
const metricUnit = (rawUnit?.prefix ? rawUnit.prefix : "") + (rawUnit?.base ? rawUnit.base : "")
// Init
comparePlotData[m] = {unit: metricUnit, data: [[],[],[],[],[],[]]} // data: [X, XST, XRT, YMIN, YAVG, YMAX]
}
@ -123,7 +130,16 @@
if (jobs) {
let plotIndex = 0
jobs.forEach((j) => {
// Collect JobIDs for X-Ticks
jobIds.push(j.jobId)
// Resources
comparePlotData['resources'].data[0].push(plotIndex)
comparePlotData['resources'].data[1].push(j.startTime)
comparePlotData['resources'].data[2].push(j.duration)
comparePlotData['resources'].data[3].push(j.numNodes)
comparePlotData['resources'].data[4].push(j?.numHWThreads?j.numHWThreads:0)
comparePlotData['resources'].data[5].push(j?.numAccelerators?j.numAccelerators:0)
// Metrics
for (let s of j.stats) {
comparePlotData[s.name].data[0].push(plotIndex)
comparePlotData[s.name].data[1].push(j.startTime)
@ -181,16 +197,34 @@
</Col>
</Row>
{:else}
{#each metrics as m}
<Row>
<Col>
<Comparogram
title={'Compare '+ m}
xlabel="JobIds"
title={'Compare Resources'}
xlabel="JobIDs"
xticks={jobIds}
ylabel={'Resource Counts'}
data={comparePlotData['resources'].data}
{plotSync}
forResources
/>
</Col>
</Row>
{#each metrics as m}
<Row>
<Col>
<Comparogram
title={`Compare Metric '${m}'`}
xlabel="JobIDs"
xticks={jobIds}
ylabel={m}
metric={m}
yunit={comparePlotData[m].unit}
data={comparePlotData[m].data}
{plotSync}
/>
</Col>
</Row>
{/each}
<hr/><hr/>
{#each $compareData.data.jobsMetricStats as job, jindex (job.jobId)}

View File

@ -14,24 +14,24 @@
<script>
import uPlot from "uplot";
import { roundTwoDigits, formatTime } from "../units.js";
import { roundTwoDigits, formatTime, formatNumber } from "../units.js";
import { getContext, onMount, onDestroy } from "svelte";
import { Card } from "@sveltestrap/sveltestrap";
export let metric;
export let metric = "";
export let width = 0;
export let height = 300;
export let data;
export let xlabel;
export let xticks;
export let ylabel;
export let yunit;
export let title;
// export let cluster = "";
// export let subCluster = "";
export let data = null;
export let xlabel = "";
export let xticks = [];
export let ylabel = "";
export let yunit = "";
export let title = "";
export let forResources = false;
export let plotSync;
// NOTE: Metric Thresholds non-required, Cluster Mixing Allowed
const metricConfig = null // DEBUG FILLER
// const metricConfig = getContext("getMetricConfig")(cluster, subCluster, metric); // Args woher
const clusterCockpitConfig = getContext("cc-config");
const lineWidth = clusterCockpitConfig.plot_general_lineWidth / window.devicePixelRatio;
const cbmode = clusterCockpitConfig?.plot_general_colorblindMode || false;
@ -80,9 +80,6 @@
overEl.addEventListener("mouseleave", () => {
legendEl.style.display = "none";
});
// let tooltip exit plot
// overEl.style.overflow = "visible";
}
function update(u) {
@ -100,19 +97,6 @@
};
}
let maxY = null;
// TODO: Hilfreich!
// if (metricConfig !== null) {
// maxY = data[3].reduce( // Data[3] is JobMaxs
// (max, x) => Math.max(max, x),
// metricConfig.normal,
// ) || metricConfig.normal
// if (maxY >= 10 * metricConfig.peak) {
// // Hard y-range render limit if outliers in series data
// maxY = 10 * metricConfig.peak;
// }
// }
const plotSeries = [
{
label: "JobID",
@ -135,6 +119,32 @@
return formatTime(ts);
},
},
]
if (forResources) {
const resSeries = [
{
label: "Nodes",
scale: "y",
width: lineWidth,
stroke: "black",
},
{
label: "Threads",
scale: "y",
width: lineWidth,
stroke: "rgb(0,0,255)",
},
{
label: "Accelerators",
scale: "y",
width: lineWidth,
stroke: cbmode ? "rgb(0,255,0)" : "red",
}
];
plotSeries.push(...resSeries)
} else {
const statsSeries = [
{
label: "Min",
scale: "y",
@ -163,6 +173,8 @@
},
}
];
plotSeries.push(...statsSeries)
};
const plotBands = [
{ series: [5, 4], fill: cbmode ? "rgba(0,0,255,0.1)" : "rgba(0,255,0,0.1)" },
@ -198,19 +210,20 @@
scale: "y",
grid: { show: true },
labelFont: "sans-serif",
label: ylabel + (yunit ? ` (${yunit})` : '')
label: ylabel + (yunit ? ` (${yunit})` : ''),
values: (u, vals) => vals.map((v) => formatNumber(v)),
},
],
bands: plotBands,
padding: [5, 10, 0, 0], // 5, 10, -20, 0
bands: forResources ? [] : plotBands,
padding: [5, 10, 0, 0],
hooks: {
draw: [
(u) => {
// Draw plot type label:
let textl = "Metric Min/Avg/Max for Job Duration";
let textl = forResources ? "Job Resources by Type" : "Metric Min/Avg/Max for Job Duration";
let textr = "Earlier <- StartTime -> Later";
u.ctx.save();
u.ctx.textAlign = "start"; // 'end'
u.ctx.textAlign = "start";
u.ctx.fillStyle = "black";
u.ctx.fillText(textl, u.bbox.left + 10, u.bbox.top + 10);
u.ctx.textAlign = "end";
@ -220,24 +233,8 @@
u.bbox.left + u.bbox.width - 10,
u.bbox.top + 10,
);
// u.ctx.fillText(text, u.bbox.left + u.bbox.width - 10, u.bbox.top + u.bbox.height - 10) // Recipe for bottom right
if (!metricConfig) {
u.ctx.restore();
return;
}
// TODO: Braucht MetricConf
let y = u.valToPos(metricConfig?.normal, "y", true);
u.ctx.save();
u.ctx.lineWidth = lineWidth;
u.ctx.strokeStyle = "#000000"; // Black
u.ctx.setLineDash([5, 5]);
u.ctx.beginPath();
u.ctx.moveTo(u.bbox.left, y);
u.ctx.lineTo(u.bbox.left + u.bbox.width, y);
u.ctx.stroke();
u.ctx.restore();
},
]
},
@ -245,7 +242,7 @@
x: { time: false },
xst: { time: false },
xrt: { time: false },
y: maxY ? { min: 0, max: (maxY * 1.1) } : {auto: true}, // Add some space to upper render limit
y: {auto: true, distr: forResources ? 3 : 1},
},
legend: {
// Display legend
@ -254,6 +251,10 @@
},
cursor: {
drag: { x: true, y: true },
sync: {
key: plotSync.key,
scales: ["x", null],
}
}
};
@ -267,6 +268,7 @@
opts.width = ren_width;
opts.height = ren_height;
uplot = new uPlot(opts, data, plotWrapper); // Data is uplot formatted [[X][Ymin][Yavg][Ymax]]
plotSync.sub(uplot)
} else {
uplot.setSize({ width: ren_width, height: ren_height });
}