/
hpa.go
303 lines (258 loc) · 8.5 KB
/
hpa.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
package custom
import (
"context"
"reflect"
"sync"
"time"
"github.com/kaladaOpuiyo/k8s-logicmonitor-adapter/pkg/collector"
m "github.com/kaladaOpuiyo/k8s-logicmonitor-adapter/pkg/metrics-adapter/metrics-store"
"github.com/kaladaOpuiyo/k8s-logicmonitor-adapter/pkg/recorder"
"github.com/kubernetes-incubator/custom-metrics-apiserver/pkg/provider"
log "github.com/sirupsen/logrus"
autoscalingv2 "k8s.io/api/autoscaling/v2beta2"
metav1 "k8s.io/apimachinery/pkg/apis/meta/v1"
"k8s.io/apimachinery/pkg/labels"
"k8s.io/client-go/kubernetes"
kube_record "k8s.io/client-go/tools/record"
"k8s.io/metrics/pkg/apis/external_metrics"
)
type metricCollection struct {
Values []collector.CollectedMetric
Error error
}
type resourceReference struct {
Name string
Namespace string
}
// CollectorScheduler is a scheduler for running metric collection jobs.
// It keeps track of all running collectors and stops them if they are to be
// removed.
type CollectorScheduler struct {
ctx context.Context
table map[resourceReference]map[collector.MetricTypeName]context.CancelFunc
metricSink chan<- metricCollection
sync.RWMutex
}
// HPAProvider is a base provider for initializing metric collectors based on
// HPA resources.
type HPAProvider struct {
client kubernetes.Interface
interval time.Duration
collectorScheduler *CollectorScheduler
collectorInterval time.Duration
metricSink chan metricCollection
hpaCache map[resourceReference]autoscalingv2.HorizontalPodAutoscaler
metricStore *m.MetricStore
collectorFactory *collector.CollectorFactory
recorder kube_record.EventRecorder
logger *log.Entry
}
// NewHPAProvider ...
func NewHPAProvider(client kubernetes.Interface, interval, collectorInterval time.Duration, collectorFactory *collector.CollectorFactory) *HPAProvider {
metricsc := make(chan metricCollection)
return &HPAProvider{
client: client,
interval: interval,
collectorInterval: collectorInterval,
metricSink: metricsc,
metricStore: m.NewMetricStore(func() time.Time {
return time.Now().UTC().Add(15 * time.Minute)
}),
collectorFactory: collectorFactory,
recorder: recorder.CreateEventRecorder(client),
logger: log.WithFields(log.Fields{"provider": "hpa"}),
}
}
// Run runs the HPA resource discovery and metric collection.
func (h *HPAProvider) Run(ctx context.Context) {
// initialize collector table
h.collectorScheduler = NewCollectorScheduler(ctx, h.metricSink)
go h.collectMetrics(ctx)
for {
err := h.update()
if err != nil {
h.logger.Error(err)
}
select {
case <-time.After(h.interval):
case <-ctx.Done():
h.logger.Info("Stopped HPA provider.")
return
}
}
}
// update discovers all HPA resources and sets upmetric collectors for new
// HPAs.
func (h *HPAProvider) update() error {
h.logger.Info("Looking for HPAs")
hpas, err := h.client.AutoscalingV2beta2().HorizontalPodAutoscalers(metav1.NamespaceAll).List(metav1.ListOptions{})
if err != nil {
return err
}
newHPACache := make(map[resourceReference]autoscalingv2.HorizontalPodAutoscaler, len(hpas.Items))
newHPAs := 0
for _, hpa := range hpas.Items {
resourceRef := resourceReference{
Name: hpa.Name,
Namespace: hpa.Namespace,
}
cachedHPA, ok := h.hpaCache[resourceRef]
hpaUpdated := !equalHPA(cachedHPA, hpa)
if !ok || hpaUpdated {
// if the hpa has changed then remove the previous
// scheduled collector.
if hpaUpdated {
h.logger.Infof("Removing previously scheduled metrics collector: %s", resourceRef)
h.collectorScheduler.Remove(resourceRef)
}
metricConfigs, err := collector.Parse(&hpa)
if err != nil {
h.logger.Errorf("Failed to parse HPA metrics: %v", err)
continue
}
cache := true
for _, config := range metricConfigs {
interval := config.Interval
if interval == 0 {
interval = h.collectorInterval
}
c, err := h.collectorFactory.NewCollector(&hpa, config, interval)
if err != nil {
cache = false
continue
}
h.logger.Infof("Adding new metrics collector: %T", c)
h.collectorScheduler.Add(resourceRef, config.MetricTypeName, c)
}
newHPAs++
// if we get an error setting upthe collectors for the
// HPA, don't cache it, but try again later.
if !cache {
continue
}
}
newHPACache[resourceRef] = hpa
}
for ref := range h.hpaCache {
if _, ok := newHPACache[ref]; ok {
continue
}
h.logger.Infof("Removing previously scheduled metrics collector: %s", ref)
h.collectorScheduler.Remove(ref)
}
h.logger.Infof("Found %d new/updated HPA(s)", newHPAs)
h.hpaCache = newHPACache
return nil
}
// equalHPA returns true if two HPAs are identical (apart from their status).
func equalHPA(a, b autoscalingv2.HorizontalPodAutoscaler) bool {
// reset resource version to not compare it since this will change
// whenever the status of the object is updated. We only want to
// compare the metadata and the spec.
a.ObjectMeta.ResourceVersion = ""
b.ObjectMeta.ResourceVersion = ""
return reflect.DeepEqual(a.ObjectMeta, b.ObjectMeta) && reflect.DeepEqual(a.Spec, b.Spec)
}
// collectMetrics collects all metrics from collectors and manages a central
// metric store.
func (h *HPAProvider) collectMetrics(ctx context.Context) {
// run garbage collection every 10 minutes
go func(ctx context.Context) {
for {
select {
case <-time.After(10 * time.Minute):
h.metricStore.RemoveExpired()
case <-ctx.Done():
h.logger.Info("Stopped metrics store garbage collection.")
return
}
}
}(ctx)
for {
select {
case collection := <-h.metricSink:
if collection.Error != nil {
h.logger.Errorf("Failed to collect metrics: %v", collection.Error)
}
h.logger.Infof("Collected %d new metric(s)", len(collection.Values))
for _, value := range collection.Values {
switch value.Type {
case autoscalingv2.ExternalMetricSourceType:
h.logger.Infof("Collected new external metric '%s' (%s) [%s]",
value.External.MetricName,
value.External.Value.String(),
labels.Set(value.External.MetricLabels).String(),
)
}
h.metricStore.Insert(value)
}
case <-ctx.Done():
h.logger.Info("Stopped metrics collection.")
return
}
}
}
// GetExternalMetric ...
func (h *HPAProvider) GetExternalMetric(namespace string, metricSelector labels.Selector, info provider.ExternalMetricInfo) (*external_metrics.ExternalMetricValueList, error) {
return h.metricStore.GetExternalMetric(namespace, metricSelector, info)
}
// ListAllExternalMetrics ...
func (h *HPAProvider) ListAllExternalMetrics() []provider.ExternalMetricInfo {
return h.metricStore.ListAllExternalMetrics()
}
// NewCollectorScheduler initializes a new CollectorScheduler.
func NewCollectorScheduler(ctx context.Context, metricsc chan<- metricCollection) *CollectorScheduler {
return &CollectorScheduler{
ctx: ctx,
table: map[resourceReference]map[collector.MetricTypeName]context.CancelFunc{},
metricSink: metricsc,
}
}
// Add adds a new collector to the collector scheduler. Once the collector is
// added it will be started to collect metrics.
func (t *CollectorScheduler) Add(resourceRef resourceReference, typeName collector.MetricTypeName, metricCollector collector.Collector) {
t.Lock()
defer t.Unlock()
collectors, ok := t.table[resourceRef]
if !ok {
collectors = map[collector.MetricTypeName]context.CancelFunc{}
t.table[resourceRef] = collectors
}
if cancelCollector, ok := collectors[typeName]; ok {
// stopold collector
cancelCollector()
}
ctx, cancel := context.WithCancel(t.ctx)
collectors[typeName] = cancel
// start runner for new collector
go collectorRunner(ctx, metricCollector, t.metricSink)
}
// collectorRunner runs a collector at the desired interval. If the passed
// context is canceled the collection will be stopped.
func collectorRunner(ctx context.Context, collector collector.Collector, metricsc chan<- metricCollection) {
for {
values, err := collector.GetMetrics()
metricsc <- metricCollection{
Values: values,
Error: err,
}
select {
case <-time.After(collector.Interval()):
case <-ctx.Done():
log.Info("stopping collector runner...")
return
}
}
}
// Remove removes a collector from the Collector scheduler. The collector is
// stopped before it's removed.
func (t *CollectorScheduler) Remove(resourceRef resourceReference) {
t.Lock()
defer t.Unlock()
if collectors, ok := t.table[resourceRef]; ok {
for _, cancelCollector := range collectors {
cancelCollector()
}
delete(t.table, resourceRef)
}
}