-
Notifications
You must be signed in to change notification settings - Fork 38.6k
/
util.go
453 lines (402 loc) · 15 KB
/
util.go
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
/*
Copyright 2015 The Kubernetes Authors.
Licensed under the Apache License, Version 2.0 (the "License");
you may not use this file except in compliance with the License.
You may obtain a copy of the License at
http://www.apache.org/licenses/LICENSE-2.0
Unless required by applicable law or agreed to in writing, software
distributed under the License is distributed on an "AS IS" BASIS,
WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
See the License for the specific language governing permissions and
limitations under the License.
*/
package benchmark
import (
"bytes"
"context"
"encoding/json"
"flag"
"fmt"
"math"
"os"
"path"
"sort"
"testing"
"time"
v1 "k8s.io/api/core/v1"
resourcev1alpha2 "k8s.io/api/resource/v1alpha2"
metav1 "k8s.io/apimachinery/pkg/apis/meta/v1"
"k8s.io/apimachinery/pkg/labels"
"k8s.io/apimachinery/pkg/util/sets"
"k8s.io/client-go/dynamic"
"k8s.io/client-go/informers"
coreinformers "k8s.io/client-go/informers/core/v1"
clientset "k8s.io/client-go/kubernetes"
restclient "k8s.io/client-go/rest"
cliflag "k8s.io/component-base/cli/flag"
"k8s.io/component-base/featuregate"
"k8s.io/component-base/metrics/legacyregistry"
"k8s.io/component-base/metrics/testutil"
"k8s.io/klog/v2"
kubeschedulerconfigv1 "k8s.io/kube-scheduler/config/v1"
"k8s.io/kubernetes/cmd/kube-apiserver/app/options"
"k8s.io/kubernetes/pkg/features"
"k8s.io/kubernetes/pkg/scheduler/apis/config"
kubeschedulerscheme "k8s.io/kubernetes/pkg/scheduler/apis/config/scheme"
"k8s.io/kubernetes/test/integration/framework"
"k8s.io/kubernetes/test/integration/util"
testutils "k8s.io/kubernetes/test/utils"
)
const (
dateFormat = "2006-01-02T15:04:05Z"
testNamespace = "sched-test"
setupNamespace = "sched-setup"
throughputSampleInterval = time.Second
)
var dataItemsDir = flag.String("data-items-dir", "", "destination directory for storing generated data items for perf dashboard")
func newDefaultComponentConfig() (*config.KubeSchedulerConfiguration, error) {
gvk := kubeschedulerconfigv1.SchemeGroupVersion.WithKind("KubeSchedulerConfiguration")
cfg := config.KubeSchedulerConfiguration{}
_, _, err := kubeschedulerscheme.Codecs.UniversalDecoder().Decode(nil, &gvk, &cfg)
if err != nil {
return nil, err
}
return &cfg, nil
}
// mustSetupCluster starts the following components:
// - k8s api server
// - scheduler
// - some of the kube-controller-manager controllers
//
// It returns regular and dynamic clients, and destroyFunc which should be used to
// remove resources after finished.
// Notes on rate limiter:
// - client rate limit is set to 5000.
func mustSetupCluster(ctx context.Context, tb testing.TB, config *config.KubeSchedulerConfiguration, enabledFeatures map[featuregate.Feature]bool) (informers.SharedInformerFactory, clientset.Interface, dynamic.Interface) {
// Run API server with minimimal logging by default. Can be raised with -v.
framework.MinVerbosity = 0
_, kubeConfig, tearDownFn := framework.StartTestServer(ctx, tb, framework.TestServerSetup{
ModifyServerRunOptions: func(opts *options.ServerRunOptions) {
// Disable ServiceAccount admission plugin as we don't have serviceaccount controller running.
opts.Admission.GenericAdmission.DisablePlugins = []string{"ServiceAccount", "TaintNodesByCondition", "Priority"}
// Enable DRA API group.
if enabledFeatures[features.DynamicResourceAllocation] {
opts.APIEnablement.RuntimeConfig = cliflag.ConfigurationMap{
resourcev1alpha2.SchemeGroupVersion.String(): "true",
}
}
},
})
tb.Cleanup(tearDownFn)
// Cleanup will be in reverse order: first the clients get cancelled,
// then the apiserver is torn down.
ctx, cancel := context.WithCancel(ctx)
tb.Cleanup(cancel)
// TODO: client connection configuration, such as QPS or Burst is configurable in theory, this could be derived from the `config`, need to
// support this when there is any testcase that depends on such configuration.
cfg := restclient.CopyConfig(kubeConfig)
cfg.QPS = 5000.0
cfg.Burst = 5000
// use default component config if config here is nil
if config == nil {
var err error
config, err = newDefaultComponentConfig()
if err != nil {
tb.Fatalf("Error creating default component config: %v", err)
}
}
client := clientset.NewForConfigOrDie(cfg)
dynClient := dynamic.NewForConfigOrDie(cfg)
// Not all config options will be effective but only those mostly related with scheduler performance will
// be applied to start a scheduler, most of them are defined in `scheduler.schedulerOptions`.
_, informerFactory := util.StartScheduler(ctx, client, cfg, config)
util.StartFakePVController(ctx, client, informerFactory)
runGC := util.CreateGCController(ctx, tb, *cfg, informerFactory)
runNS := util.CreateNamespaceController(ctx, tb, *cfg, informerFactory)
runResourceClaimController := func() {}
if enabledFeatures[features.DynamicResourceAllocation] {
// Testing of DRA with inline resource claims depends on this
// controller for creating and removing ResourceClaims.
runResourceClaimController = util.CreateResourceClaimController(ctx, tb, client, informerFactory)
}
informerFactory.Start(ctx.Done())
informerFactory.WaitForCacheSync(ctx.Done())
go runGC()
go runNS()
go runResourceClaimController()
return informerFactory, client, dynClient
}
// Returns the list of scheduled pods in the specified namespaces.
// Note that no namespaces specified matches all namespaces.
func getScheduledPods(podInformer coreinformers.PodInformer, namespaces ...string) ([]*v1.Pod, error) {
pods, err := podInformer.Lister().List(labels.Everything())
if err != nil {
return nil, err
}
s := sets.New(namespaces...)
scheduled := make([]*v1.Pod, 0, len(pods))
for i := range pods {
pod := pods[i]
if len(pod.Spec.NodeName) > 0 && (len(s) == 0 || s.Has(pod.Namespace)) {
scheduled = append(scheduled, pod)
}
}
return scheduled, nil
}
// DataItem is the data point.
type DataItem struct {
// Data is a map from bucket to real data point (e.g. "Perc90" -> 23.5). Notice
// that all data items with the same label combination should have the same buckets.
Data map[string]float64 `json:"data"`
// Unit is the data unit. Notice that all data items with the same label combination
// should have the same unit.
Unit string `json:"unit"`
// Labels is the labels of the data item.
Labels map[string]string `json:"labels,omitempty"`
}
// DataItems is the data point set. It is the struct that perf dashboard expects.
type DataItems struct {
Version string `json:"version"`
DataItems []DataItem `json:"dataItems"`
}
// makeBasePod creates a Pod object to be used as a template.
func makeBasePod() *v1.Pod {
basePod := &v1.Pod{
ObjectMeta: metav1.ObjectMeta{
GenerateName: "pod-",
},
Spec: testutils.MakePodSpec(),
}
return basePod
}
func dataItems2JSONFile(dataItems DataItems, namePrefix string) error {
// perfdash expects all data items to have the same set of labels. It
// then renders drop-down buttons for each label with all values found
// for each label. If we were to store data items that don't have a
// certain label, then perfdash will never show those data items
// because it will only show data items that have the currently
// selected label value. To avoid that, we collect all labels used
// anywhere and then add missing labels with "not applicable" as value.
labels := sets.New[string]()
for _, item := range dataItems.DataItems {
for label := range item.Labels {
labels.Insert(label)
}
}
for _, item := range dataItems.DataItems {
for label := range labels {
if _, ok := item.Labels[label]; !ok {
item.Labels[label] = "not applicable"
}
}
}
b, err := json.Marshal(dataItems)
if err != nil {
return err
}
destFile := fmt.Sprintf("%v_%v.json", namePrefix, time.Now().Format(dateFormat))
if *dataItemsDir != "" {
// Ensure the "dataItemsDir" path to be valid.
if err := os.MkdirAll(*dataItemsDir, 0750); err != nil {
return fmt.Errorf("dataItemsDir path %v does not exist and cannot be created: %v", *dataItemsDir, err)
}
destFile = path.Join(*dataItemsDir, destFile)
}
formatted := &bytes.Buffer{}
if err := json.Indent(formatted, b, "", " "); err != nil {
return fmt.Errorf("indenting error: %v", err)
}
return os.WriteFile(destFile, formatted.Bytes(), 0644)
}
type labelValues struct {
label string
values []string
}
// metricsCollectorConfig is the config to be marshalled to YAML config file.
// NOTE: The mapping here means only one filter is supported, either value in the list of `values` is able to be collected.
type metricsCollectorConfig struct {
Metrics map[string]*labelValues
}
// metricsCollector collects metrics from legacyregistry.DefaultGatherer.Gather() endpoint.
// Currently only Histogram metrics are supported.
type metricsCollector struct {
*metricsCollectorConfig
labels map[string]string
}
func newMetricsCollector(config *metricsCollectorConfig, labels map[string]string) *metricsCollector {
return &metricsCollector{
metricsCollectorConfig: config,
labels: labels,
}
}
func (*metricsCollector) run(ctx context.Context) {
// metricCollector doesn't need to start before the tests, so nothing to do here.
}
func (pc *metricsCollector) collect() []DataItem {
var dataItems []DataItem
for metric, labelVals := range pc.Metrics {
// no filter is specified, aggregate all the metrics within the same metricFamily.
if labelVals == nil {
dataItem := collectHistogramVec(metric, pc.labels, nil)
if dataItem != nil {
dataItems = append(dataItems, *dataItem)
}
} else {
// fetch the metric from metricFamily which match each of the lvMap.
for _, value := range labelVals.values {
lvMap := map[string]string{labelVals.label: value}
dataItem := collectHistogramVec(metric, pc.labels, lvMap)
if dataItem != nil {
dataItems = append(dataItems, *dataItem)
}
}
}
}
return dataItems
}
func collectHistogramVec(metric string, labels map[string]string, lvMap map[string]string) *DataItem {
vec, err := testutil.GetHistogramVecFromGatherer(legacyregistry.DefaultGatherer, metric, lvMap)
if err != nil {
klog.Error(err)
return nil
}
if err := vec.Validate(); err != nil {
klog.ErrorS(err, "the validation for HistogramVec is failed. The data for this metric won't be stored in a benchmark result file", "metric", metric, "labels", labels)
return nil
}
if vec.GetAggregatedSampleCount() == 0 {
klog.InfoS("It is expected that this metric wasn't recorded. The data for this metric won't be stored in a benchmark result file", "metric", metric, "labels", labels)
return nil
}
q50 := vec.Quantile(0.50)
q90 := vec.Quantile(0.90)
q95 := vec.Quantile(0.95)
q99 := vec.Quantile(0.99)
avg := vec.Average()
msFactor := float64(time.Second) / float64(time.Millisecond)
// Copy labels and add "Metric" label for this metric.
labelMap := map[string]string{"Metric": metric}
for k, v := range labels {
labelMap[k] = v
}
for k, v := range lvMap {
labelMap[k] = v
}
return &DataItem{
Labels: labelMap,
Data: map[string]float64{
"Perc50": q50 * msFactor,
"Perc90": q90 * msFactor,
"Perc95": q95 * msFactor,
"Perc99": q99 * msFactor,
"Average": avg * msFactor,
},
Unit: "ms",
}
}
type throughputCollector struct {
tb testing.TB
podInformer coreinformers.PodInformer
schedulingThroughputs []float64
labels map[string]string
namespaces []string
errorMargin float64
}
func newThroughputCollector(tb testing.TB, podInformer coreinformers.PodInformer, labels map[string]string, namespaces []string, errorMargin float64) *throughputCollector {
return &throughputCollector{
tb: tb,
podInformer: podInformer,
labels: labels,
namespaces: namespaces,
errorMargin: errorMargin,
}
}
func (tc *throughputCollector) run(ctx context.Context) {
podsScheduled, err := getScheduledPods(tc.podInformer, tc.namespaces...)
if err != nil {
klog.Fatalf("%v", err)
}
lastScheduledCount := len(podsScheduled)
ticker := time.NewTicker(throughputSampleInterval)
defer ticker.Stop()
lastSampleTime := time.Now()
started := false
skipped := 0
for {
select {
case <-ctx.Done():
return
case <-ticker.C:
now := time.Now()
podsScheduled, err := getScheduledPods(tc.podInformer, tc.namespaces...)
if err != nil {
klog.Fatalf("%v", err)
}
scheduled := len(podsScheduled)
// Only do sampling if number of scheduled pods is greater than zero.
if scheduled == 0 {
continue
}
if !started {
started = true
// Skip the initial sample. It's likely to be an outlier because
// sampling and creating pods get started independently.
lastScheduledCount = scheduled
lastSampleTime = now
continue
}
newScheduled := scheduled - lastScheduledCount
if newScheduled == 0 {
// Throughput would be zero for the interval.
// Instead of recording 0 pods/s, keep waiting
// until we see at least one additional pod
// being scheduled.
skipped++
continue
}
// This should be roughly equal to
// throughputSampleInterval * (skipped + 1), but we
// don't count on that because the goroutine might not
// be scheduled immediately when the timer
// triggers. Instead we track the actual time stamps.
duration := now.Sub(lastSampleTime)
durationInSeconds := duration.Seconds()
throughput := float64(newScheduled) / durationInSeconds
expectedDuration := throughputSampleInterval * time.Duration(skipped+1)
errorMargin := (duration - expectedDuration).Seconds() / expectedDuration.Seconds() * 100
if tc.errorMargin > 0 && math.Abs(errorMargin) > tc.errorMargin {
// This might affect the result, report it.
tc.tb.Errorf("ERROR: Expected throuput collector to sample at regular time intervals. The %d most recent intervals took %s instead of %s, a difference of %0.1f%%.", skipped+1, duration, expectedDuration, errorMargin)
}
// To keep percentiles accurate, we have to record multiple samples with the same
// throughput value if we skipped some intervals.
for i := 0; i <= skipped; i++ {
tc.schedulingThroughputs = append(tc.schedulingThroughputs, throughput)
}
lastScheduledCount = scheduled
klog.Infof("%d pods scheduled", lastScheduledCount)
skipped = 0
lastSampleTime = now
}
}
}
func (tc *throughputCollector) collect() []DataItem {
throughputSummary := DataItem{Labels: tc.labels}
if length := len(tc.schedulingThroughputs); length > 0 {
sort.Float64s(tc.schedulingThroughputs)
sum := 0.0
for i := range tc.schedulingThroughputs {
sum += tc.schedulingThroughputs[i]
}
throughputSummary.Labels["Metric"] = "SchedulingThroughput"
throughputSummary.Data = map[string]float64{
"Average": sum / float64(length),
"Perc50": tc.schedulingThroughputs[int(math.Ceil(float64(length*50)/100))-1],
"Perc90": tc.schedulingThroughputs[int(math.Ceil(float64(length*90)/100))-1],
"Perc95": tc.schedulingThroughputs[int(math.Ceil(float64(length*95)/100))-1],
"Perc99": tc.schedulingThroughputs[int(math.Ceil(float64(length*99)/100))-1],
}
throughputSummary.Unit = "pods/s"
}
return []DataItem{throughputSummary}
}