forked from kubernetes/kubernetes
-
Notifications
You must be signed in to change notification settings - Fork 0
/
hpa.go
938 lines (859 loc) · 33.2 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
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
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
/*
Copyright 2017 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 federatedtypes
import (
"time"
"fmt"
autoscalingv1 "k8s.io/api/autoscaling/v1"
metav1 "k8s.io/apimachinery/pkg/apis/meta/v1"
pkgruntime "k8s.io/apimachinery/pkg/runtime"
"k8s.io/apimachinery/pkg/runtime/schema"
"k8s.io/apimachinery/pkg/util/sets"
"k8s.io/apimachinery/pkg/watch"
kubeclientset "k8s.io/client-go/kubernetes"
restclient "k8s.io/client-go/rest"
federationapi "k8s.io/kubernetes/federation/apis/federation/v1beta1"
federationclientset "k8s.io/kubernetes/federation/client/clientset_generated/federation_clientset"
fedutil "k8s.io/kubernetes/federation/pkg/federation-controller/util"
)
const (
HpaKind = "horizontalpodautoscaler"
HpaControllerName = "horizontalpodautoscalers"
// This is used as the default min for hpa object submitted
// to federation, in a situation where the default is for
// some reason not present (Spec.MinReplicas == nil)
hpaMinReplicaDefault = int32(1)
// This is a tunable which does not change replica nums
// on an existing local hpa, before this timeout, if it
// did scale already (avoids thrashing of replicas around).
ScaleForbiddenWindow = 2 * time.Minute
)
func init() {
RegisterFederatedType(HpaKind, HpaControllerName, []schema.GroupVersionResource{autoscalingv1.SchemeGroupVersion.WithResource(HpaControllerName)}, NewHpaAdapter)
}
type HpaAdapter struct {
client federationclientset.Interface
scaleForbiddenWindow time.Duration
}
func NewHpaAdapter(client federationclientset.Interface, config *restclient.Config, adapterSpecificArgs map[string]interface{}) FederatedTypeAdapter {
var scaleForbiddenWindow time.Duration
if adapterSpecificArgs != nil && adapterSpecificArgs[HpaKind] != nil {
scaleForbiddenWindow = adapterSpecificArgs[HpaKind].(*metav1.Duration).Duration
} else {
scaleForbiddenWindow = ScaleForbiddenWindow
}
return &HpaAdapter{
client: client,
scaleForbiddenWindow: scaleForbiddenWindow,
}
}
func (a *HpaAdapter) Kind() string {
return HpaKind
}
func (a *HpaAdapter) ObjectType() pkgruntime.Object {
return &autoscalingv1.HorizontalPodAutoscaler{}
}
func (a *HpaAdapter) IsExpectedType(obj interface{}) bool {
_, ok := obj.(*autoscalingv1.HorizontalPodAutoscaler)
return ok
}
func (a *HpaAdapter) Copy(obj pkgruntime.Object) pkgruntime.Object {
hpa := obj.(*autoscalingv1.HorizontalPodAutoscaler)
return &autoscalingv1.HorizontalPodAutoscaler{
ObjectMeta: fedutil.DeepCopyRelevantObjectMeta(hpa.ObjectMeta),
Spec: *fedutil.DeepCopyApiTypeOrPanic(&hpa.Spec).(*autoscalingv1.HorizontalPodAutoscalerSpec),
}
}
func (a *HpaAdapter) Equivalent(obj1, obj2 pkgruntime.Object) bool {
return fedutil.ObjectMetaAndSpecEquivalent(obj1, obj2)
}
func (a *HpaAdapter) QualifiedName(obj pkgruntime.Object) QualifiedName {
hpa := obj.(*autoscalingv1.HorizontalPodAutoscaler)
return QualifiedName{Namespace: hpa.Namespace, Name: hpa.Name}
}
func (a *HpaAdapter) ObjectMeta(obj pkgruntime.Object) *metav1.ObjectMeta {
return &obj.(*autoscalingv1.HorizontalPodAutoscaler).ObjectMeta
}
func (a *HpaAdapter) FedCreate(obj pkgruntime.Object) (pkgruntime.Object, error) {
hpa := obj.(*autoscalingv1.HorizontalPodAutoscaler)
return a.client.AutoscalingV1().HorizontalPodAutoscalers(hpa.Namespace).Create(hpa)
}
func (a *HpaAdapter) FedDelete(qualifiedName QualifiedName, options *metav1.DeleteOptions) error {
return a.client.AutoscalingV1().HorizontalPodAutoscalers(qualifiedName.Namespace).Delete(qualifiedName.Name, options)
}
func (a *HpaAdapter) FedGet(qualifiedName QualifiedName) (pkgruntime.Object, error) {
return a.client.AutoscalingV1().HorizontalPodAutoscalers(qualifiedName.Namespace).Get(qualifiedName.Name, metav1.GetOptions{})
}
func (a *HpaAdapter) FedList(namespace string, options metav1.ListOptions) (pkgruntime.Object, error) {
return a.client.AutoscalingV1().HorizontalPodAutoscalers(namespace).List(options)
}
func (a *HpaAdapter) FedUpdate(obj pkgruntime.Object) (pkgruntime.Object, error) {
hpa := obj.(*autoscalingv1.HorizontalPodAutoscaler)
return a.client.AutoscalingV1().HorizontalPodAutoscalers(hpa.Namespace).Update(hpa)
}
func (a *HpaAdapter) FedWatch(namespace string, options metav1.ListOptions) (watch.Interface, error) {
return a.client.AutoscalingV1().HorizontalPodAutoscalers(namespace).Watch(options)
}
func (a *HpaAdapter) ClusterCreate(client kubeclientset.Interface, obj pkgruntime.Object) (pkgruntime.Object, error) {
hpa := obj.(*autoscalingv1.HorizontalPodAutoscaler)
return client.AutoscalingV1().HorizontalPodAutoscalers(hpa.Namespace).Create(hpa)
}
func (a *HpaAdapter) ClusterDelete(client kubeclientset.Interface, qualifiedName QualifiedName, options *metav1.DeleteOptions) error {
return client.AutoscalingV1().HorizontalPodAutoscalers(qualifiedName.Namespace).Delete(qualifiedName.Name, options)
}
func (a *HpaAdapter) ClusterGet(client kubeclientset.Interface, qualifiedName QualifiedName) (pkgruntime.Object, error) {
return client.AutoscalingV1().HorizontalPodAutoscalers(qualifiedName.Namespace).Get(qualifiedName.Name, metav1.GetOptions{})
}
func (a *HpaAdapter) ClusterList(client kubeclientset.Interface, namespace string, options metav1.ListOptions) (pkgruntime.Object, error) {
return client.AutoscalingV1().HorizontalPodAutoscalers(namespace).List(options)
}
func (a *HpaAdapter) ClusterUpdate(client kubeclientset.Interface, obj pkgruntime.Object) (pkgruntime.Object, error) {
hpa := obj.(*autoscalingv1.HorizontalPodAutoscaler)
return client.AutoscalingV1().HorizontalPodAutoscalers(hpa.Namespace).Update(hpa)
}
func (a *HpaAdapter) ClusterWatch(client kubeclientset.Interface, namespace string, options metav1.ListOptions) (watch.Interface, error) {
return client.AutoscalingV1().HorizontalPodAutoscalers(namespace).Watch(options)
}
func (a *HpaAdapter) NewTestObject(namespace string) pkgruntime.Object {
var min int32 = 4
var targetCPU int32 = 70
return &autoscalingv1.HorizontalPodAutoscaler{
ObjectMeta: metav1.ObjectMeta{
GenerateName: "test-hpa-",
Namespace: namespace,
},
Spec: autoscalingv1.HorizontalPodAutoscalerSpec{
ScaleTargetRef: autoscalingv1.CrossVersionObjectReference{
Kind: "replicaset",
Name: "myrs",
},
MinReplicas: &min,
MaxReplicas: int32(10),
TargetCPUUtilizationPercentage: &targetCPU,
},
}
}
func (a *HpaAdapter) IsSchedulingAdapter() bool {
return true
}
func (a *HpaAdapter) EquivalentIgnoringSchedule(obj1, obj2 pkgruntime.Object) bool {
hpa1 := obj1.(*autoscalingv1.HorizontalPodAutoscaler)
hpa2 := a.Copy(obj2).(*autoscalingv1.HorizontalPodAutoscaler)
if hpa1.Spec.MinReplicas == nil {
hpa2.Spec.MinReplicas = nil
} else if hpa2.Spec.MinReplicas == nil {
var r int32 = *hpa1.Spec.MinReplicas
hpa2.Spec.MinReplicas = &r
} else {
*hpa2.Spec.MinReplicas = *hpa1.Spec.MinReplicas
}
hpa2.Spec.MaxReplicas = hpa1.Spec.MaxReplicas
return fedutil.ObjectMetaAndSpecEquivalent(hpa1, hpa2)
}
type replicaNums struct {
min int32
max int32
}
type hpaFederatedStatus struct {
lastScaleTime *metav1.Time
// Indicates how many clusters have hpa/replicas.
// Used to average the cpu utilization which is
// reflected to the federation user.
count int32
aggregateCPUUtilizationPercentage *int32
currentReplicas int32
desiredReplicas int32
}
type hpaSchedulingInfo struct {
scheduleState map[string]*replicaNums
fedStatus hpaFederatedStatus
}
// List of cluster names.
type hpaLists struct {
// Stores names of those clusters which can offer min.
availableMin sets.String
// Stores names of those clusters which can offer max.
availableMax sets.String
// Stores names of those clusters which do not have hpa yet.
noHpa sets.String
}
func (a *HpaAdapter) GetSchedule(obj pkgruntime.Object, key string, clusters []*federationapi.Cluster, informer fedutil.FederatedInformer) (interface{}, error) {
currentClusterObjs, err := getCurrentClusterObjs(informer, key, clusters)
if err != nil {
return nil, err
}
// Initialise averaged cpu utilisation for this reconcile.
var ccup int32 = 0
fedStatus := hpaFederatedStatus{
aggregateCPUUtilizationPercentage: &ccup,
count: int32(0),
desiredReplicas: int32(0),
currentReplicas: int32(0),
}
fedHpa := obj.(*autoscalingv1.HorizontalPodAutoscaler)
// We assign the last known scale time here, which we update with
// the latest time from among all clusters in ScheduleObject()
if fedHpa.Status.LastScaleTime != nil {
t := metav1.NewTime(fedHpa.Status.LastScaleTime.Time)
fedStatus.lastScaleTime = &t
}
return &hpaSchedulingInfo{
scheduleState: a.getHpaScheduleState(obj, currentClusterObjs),
fedStatus: fedStatus,
}, nil
}
func getCurrentClusterObjs(informer fedutil.FederatedInformer, key string, clusters []*federationapi.Cluster) (map[string]pkgruntime.Object, error) {
currentClusterObjs := make(map[string]pkgruntime.Object)
for _, cluster := range clusters {
clusterName := cluster.Name
clusterObj, found, err := informer.GetTargetStore().GetByKey(clusterName, key)
if err != nil {
return nil, err
}
currentClusterObjs[clusterName] = nil
if found {
currentClusterObjs[clusterName] = clusterObj.(pkgruntime.Object)
}
}
return currentClusterObjs, nil
}
// The algorithm used for scheduling is briefed as below:
//
// 1. Find clusters which can offer max and min, if any (lists.availableMax and
// lists.availableMin) in one pass on all clusters.
//
// 2. Reduce the replicas (both min and max) if needed (situation when fedHpa
// has lesser replicas then all cluster local hpa replicas totalled together).
// In this step reduce first from those hpas which already have max (and min)
// reducible. Once such clusters are over and reduction still needed, reduce
// one at a time from all clusters, randomly. This step will ensure that the
// exceeding replicas in local hpas are reduced to match the fedHpa.
// This step would ideally be a noop in most cases because its rare that fedHpa
// would have lesser replicas then the cluster local total (probably when user
// forces update if fedHpa).
//
// 3. Distribute the replicas. In this step we have replicas to distribute (which
// are fed replicas exceeding the sum total of local cluster replicas). If clusters
// already have replicas, one replica from each cluster which can offer replicas
// (both for max and min) are also added to this replicas to distribute numbers (min
// and max).
// 3a. We first do a sub-pass to distribute to clusters which need replicas, considering
// those as clusters in crucial need of replicas.
// 3b. After previous sub-pass, if we still have replicas remaining, in the sub-pass
// we distribute to those clusters which do not yet have any hpa.
// 3c. After previous if we still have more to distribute, then we distribute to all
// clusters randomly, giving replica distribution count (rdc=total-fed-replicas/no-of-clusters)
// to each at a time.
//
// The above algorithm is run to first distribute max and then distribute min to those clusters
// which get max.
func (a *HpaAdapter) getHpaScheduleState(fedObj pkgruntime.Object, currentObjs map[string]pkgruntime.Object) map[string]*replicaNums {
fedHpa := fedObj.(*autoscalingv1.HorizontalPodAutoscaler)
requestedMin := hpaMinReplicaDefault
if fedHpa.Spec.MinReplicas != nil {
requestedMin = *fedHpa.Spec.MinReplicas
}
requestedReplicas := replicaNums{
min: requestedMin,
max: fedHpa.Spec.MaxReplicas,
}
// replica distribution count, per cluster
rdc := replicaNums{
min: requestedReplicas.min / int32(len(currentObjs)),
max: requestedReplicas.max / int32(len(currentObjs)),
}
if rdc.min < 1 {
rdc.min = 1
}
// TODO: Is there a better way?
// We need to cap the lowest limit of Max to 2, because in a
// situation like both min and max become 1 (same) for all clusters,
// no rebalancing would happen.
if rdc.max < 2 {
rdc.max = 2
}
// Pass 1: Analyse existing local hpa's if any.
// clusterLists holds the list of those clusters which can offer
// min and max replicas, to those which want them.
// For example new clusters joining the federation and/or
// those clusters which need to increase or reduce replicas
// beyond min/max limits.
// schedStatus currently have status of existing hpas.
// It will eventually have desired status for this reconcile.
clusterLists, currentReplicas, scheduleState := a.prepareForScheduling(currentObjs)
remainingReplicas := replicaNums{
min: requestedReplicas.min - currentReplicas.min,
max: requestedReplicas.max - currentReplicas.max,
}
// Pass 2: reduction of replicas if needed ( situation that fedHpa updated replicas
// to lesser then existing).
// In this pass, we remain pessimistic and reduce one replica per cluster at a time.
if remainingReplicas.min < 0 {
excessMin := (remainingReplicas.min * int32(-1))
remainingReplicas.min = reduceMinReplicas(excessMin, clusterLists.availableMin, scheduleState)
}
if remainingReplicas.max < 0 {
excessMax := (remainingReplicas.max * int32(-1))
remainingReplicas.max = reduceMaxReplicas(excessMax, clusterLists.availableMax, scheduleState)
}
toDistribute := replicaNums{
min: remainingReplicas.min + int32(clusterLists.availableMin.Len()),
max: remainingReplicas.max + int32(clusterLists.availableMax.Len()),
}
// Pass 3: Distribute Max and then Min.
// Here we first distribute max and then (in the next loop)
// distribute min into those clusters which already get the
// max fixed.
// In this process we might not meet the min limit and total of
// min limits might remain more then the requested federated min.
// This is partially because a min per cluster cannot be lesser
// then 1, but min could be requested as 1 at federation.
// Additionally we first increase replicas into those clusters
// which already have hpa's and are in a condition to increase.
// This will save cluster related resources for the user, such that
// if an already existing cluster can satisfy users request why send
// the workload to another.
// We then go ahead to give the replicas to those which do not
// have any hpa. In this pass however we try to ensure that all
// our Max are consumed in this reconcile.
a.distributeMaxReplicas(toDistribute.max, clusterLists, rdc, currentObjs, scheduleState)
// We distribute min to those clusters which:
// 1 - can adjust min (our increase step would be only 1)
// 2 - which do not have this hpa and got max(increase step rdcMin)
// We might exhaust all min replicas here, with
// some clusters still needing them. We adjust this in finalise by
// assigning min replicas to 1 into those clusters which got max
// but min remains 0.
a.distributeMinReplicas(toDistribute.min, clusterLists, rdc, currentObjs, scheduleState)
return finaliseScheduleState(scheduleState)
}
func (a *HpaAdapter) ScheduleObject(cluster *federationapi.Cluster, clusterObj pkgruntime.Object, federationObjCopy pkgruntime.Object, schedulingInfo interface{}) (pkgruntime.Object, ScheduleAction, error) {
// Update federated status info
typedInfo := schedulingInfo.(*hpaSchedulingInfo)
if clusterObj != nil {
clusterHpa := clusterObj.(*autoscalingv1.HorizontalPodAutoscaler)
if clusterHpa.Status.CurrentCPUUtilizationPercentage != nil {
*typedInfo.fedStatus.aggregateCPUUtilizationPercentage +=
(*clusterHpa.Status.CurrentCPUUtilizationPercentage * clusterHpa.Status.CurrentReplicas)
typedInfo.fedStatus.count += clusterHpa.Status.CurrentReplicas
}
if clusterHpa.Status.LastScaleTime != nil {
t := metav1.NewTime(clusterHpa.Status.LastScaleTime.Time)
if typedInfo.fedStatus.lastScaleTime != nil &&
t.After(typedInfo.fedStatus.lastScaleTime.Time) {
typedInfo.fedStatus.lastScaleTime = &t
}
}
typedInfo.fedStatus.currentReplicas += clusterHpa.Status.CurrentReplicas
typedInfo.fedStatus.desiredReplicas += clusterHpa.Status.DesiredReplicas
}
// Update the cluster obj and the needed action on the cluster
clusterHpaState := typedInfo.scheduleState[cluster.Name]
desiredHpa := federationObjCopy.(*autoscalingv1.HorizontalPodAutoscaler)
if clusterHpaState != nil {
desiredHpa.Spec.MaxReplicas = clusterHpaState.max
if desiredHpa.Spec.MinReplicas == nil {
min := int32(0)
desiredHpa.Spec.MinReplicas = &min
}
*desiredHpa.Spec.MinReplicas = clusterHpaState.min
}
var defaultAction ScheduleAction = ""
switch {
case clusterHpaState != nil && clusterObj != nil:
return desiredHpa, defaultAction, nil
case clusterHpaState != nil && clusterObj == nil:
return desiredHpa, ActionAdd, nil
case clusterHpaState == nil && clusterObj != nil:
return nil, ActionDelete, nil
}
return nil, defaultAction, nil
}
func (a *HpaAdapter) UpdateFederatedStatus(obj pkgruntime.Object, schedulingInfo interface{}) error {
fedHpa := obj.(*autoscalingv1.HorizontalPodAutoscaler)
needUpdate, newFedHpaStatus := updateStatus(fedHpa, schedulingInfo.(*hpaSchedulingInfo).fedStatus)
if needUpdate {
fedHpa.Status = newFedHpaStatus
_, err := a.client.AutoscalingV1().HorizontalPodAutoscalers(fedHpa.Namespace).UpdateStatus(fedHpa)
if err != nil {
return fmt.Errorf("Error updating hpa: %s status in federation: %v", fedHpa.Name, err)
}
}
return nil
}
func updateStatus(fedHpa *autoscalingv1.HorizontalPodAutoscaler, newStatus hpaFederatedStatus) (bool, autoscalingv1.HorizontalPodAutoscalerStatus) {
averageCPUUtilizationPercentage := int32(0)
// Average out the available current utilisation
if *newStatus.aggregateCPUUtilizationPercentage != 0 && newStatus.count != 0 {
averageCPUUtilizationPercentage = *newStatus.aggregateCPUUtilizationPercentage / newStatus.count
}
gen := fedHpa.Generation
newFedHpaStatus := autoscalingv1.HorizontalPodAutoscalerStatus{ObservedGeneration: &gen}
needUpdate := false
if (fedHpa.Status.CurrentCPUUtilizationPercentage == nil &&
averageCPUUtilizationPercentage != 0) ||
(fedHpa.Status.CurrentCPUUtilizationPercentage != nil &&
averageCPUUtilizationPercentage !=
*fedHpa.Status.CurrentCPUUtilizationPercentage) {
needUpdate = true
newFedHpaStatus.CurrentCPUUtilizationPercentage = &averageCPUUtilizationPercentage
}
if (fedHpa.Status.LastScaleTime == nil && newStatus.lastScaleTime != nil) ||
(fedHpa.Status.LastScaleTime != nil && newStatus.lastScaleTime == nil) ||
((fedHpa.Status.LastScaleTime != nil && newStatus.lastScaleTime != nil) &&
newStatus.lastScaleTime.After(fedHpa.Status.LastScaleTime.Time)) {
needUpdate = true
newFedHpaStatus.LastScaleTime = newStatus.lastScaleTime
}
if fedHpa.Status.DesiredReplicas != newStatus.desiredReplicas {
needUpdate = true
newFedHpaStatus.CurrentReplicas = newStatus.currentReplicas
}
if fedHpa.Status.CurrentReplicas != newStatus.currentReplicas {
needUpdate = true
newFedHpaStatus.DesiredReplicas = newStatus.desiredReplicas
}
return needUpdate, newFedHpaStatus
}
// prepareForScheduling prepares the lists and totals from the
// existing objs.
// currentObjs has the list of all clusters, with obj as nil
// for those clusters which do not have hpa yet.
func (a *HpaAdapter) prepareForScheduling(currentObjs map[string]pkgruntime.Object) (hpaLists, replicaNums, map[string]*replicaNums) {
lists := hpaLists{
availableMax: sets.NewString(),
availableMin: sets.NewString(),
noHpa: sets.NewString(),
}
existingTotal := replicaNums{
min: int32(0),
max: int32(0),
}
scheduleState := make(map[string]*replicaNums)
for cluster, obj := range currentObjs {
if obj == nil {
lists.noHpa.Insert(cluster)
scheduleState[cluster] = nil
continue
}
if a.maxReplicasReducible(obj) {
lists.availableMax.Insert(cluster)
}
if a.minReplicasReducible(obj) {
lists.availableMin.Insert(cluster)
}
replicas := replicaNums{min: 0, max: 0}
scheduleState[cluster] = &replicas
if obj.(*autoscalingv1.HorizontalPodAutoscaler).Spec.MinReplicas != nil {
existingTotal.min += *obj.(*autoscalingv1.HorizontalPodAutoscaler).Spec.MinReplicas
replicas.min = *obj.(*autoscalingv1.HorizontalPodAutoscaler).Spec.MinReplicas
}
existingTotal.max += obj.(*autoscalingv1.HorizontalPodAutoscaler).Spec.MaxReplicas
replicas.max = obj.(*autoscalingv1.HorizontalPodAutoscaler).Spec.MaxReplicas
}
return lists, existingTotal, scheduleState
}
// Note: reduceMinReplicas and reduceMaxReplicas, look quite similar in flow
// and code, however there are subtle differences. They together can be made
// into 1 function with an arg governing the functionality difference and
// additional args (superset of args in both) as needed. Doing so however
// makes the logical flow quite less readable. They are thus left as 2 for
// readability.
// reduceMinReplicas reduces the min replicas from existing clusters.
// At the end of the function excessMin should be 0 and the MinList
// and the scheduledReplicas properly updated in place.
func reduceMinReplicas(excessMin int32, availableMinList sets.String, scheduled map[string]*replicaNums) int32 {
if excessMin > 0 {
// first we try reducing from those clusters which already offer min
if availableMinList.Len() > 0 {
for _, cluster := range availableMinList.List() {
replicas := scheduled[cluster]
if replicas.min > 1 {
replicas.min--
availableMinList.Delete(cluster)
excessMin--
if excessMin <= 0 {
break
}
}
}
}
}
// If we could not get needed replicas from already offered min above
// we abruptly start removing replicas from some/all clusters.
// Here we might make some min to 0 signalling that this hpa might be a
// candidate to be removed from this cluster altogether.
for excessMin > 0 {
for _, replicas := range scheduled {
if replicas != nil &&
replicas.min > 0 {
replicas.min--
excessMin--
if excessMin <= 0 {
break
}
}
}
}
return excessMin
}
// reduceMaxReplicas reduces the max replicas from existing clusters.
// At the end of the function excessMax should be 0 and the MaxList
// and the scheduledReplicas properly updated in place.
func reduceMaxReplicas(excessMax int32, availableMaxList sets.String, scheduled map[string]*replicaNums) int32 {
if excessMax > 0 {
// first we try reducing from those clusters which already offer max
if availableMaxList.Len() > 0 {
for _, cluster := range availableMaxList.List() {
replicas := scheduled[cluster]
if replicas != nil && !((replicas.max - replicas.min) < 0) {
replicas.max--
availableMaxList.Delete(cluster)
excessMax--
if excessMax <= 0 {
break
}
}
}
}
}
// If we could not get needed replicas to reduce from already offered
// max above we abruptly start removing replicas from some/all clusters.
// Here we might make some max and min to 0, signalling that this hpa be
// removed from this cluster altogether
for excessMax > 0 {
for _, replicas := range scheduled {
if replicas != nil &&
!((replicas.max - replicas.min) < 0) {
replicas.max--
excessMax--
if excessMax <= 0 {
break
}
}
}
}
return excessMax
}
// distributeMaxReplicas
// Takes input:
// toDistributeMax: number of replicas to distribute.
// lists: cluster name lists, which have clusters with available max,
// available min and those with no hpas yet.
// rdc: replicadistributioncount for max and min.
// currentObjs: list of current cluster hpas.
// scheduled: schedule state which will be updated in place.
func (a *HpaAdapter) distributeMaxReplicas(toDistributeMax int32, lists hpaLists, rdc replicaNums,
currentObjs map[string]pkgruntime.Object, scheduled map[string]*replicaNums) int32 {
for cluster, replicas := range scheduled {
if toDistributeMax == 0 {
break
}
if replicas == nil {
continue
}
if a.maxReplicasNeeded(currentObjs[cluster]) {
replicas.max++
if lists.availableMax.Len() > 0 {
popped, notEmpty := lists.availableMax.PopAny()
if notEmpty {
// Boundary checks have happened earlier in
// minReplicasReducible().
scheduled[popped].max--
}
}
// Any which ways utilise available map replicas
toDistributeMax--
}
}
// If we have new clusters where we can give our replicas,
// then give away all our replicas to the new clusters first.
if lists.noHpa.Len() > 0 {
for toDistributeMax > 0 {
for _, cluster := range lists.noHpa.UnsortedList() {
if scheduled[cluster] == nil {
scheduled[cluster] = &replicaNums{min: 0, max: 0}
}
replicas := scheduled[cluster]
// first give away max from clusters offering them
// this case especially helps getting hpa into newly joining
// clusters.
if lists.availableMax.Len() > 0 {
popped, notEmpty := lists.availableMax.PopAny()
if notEmpty {
// Boundary checks to reduce max have happened earlier in
// minReplicasReducible().
replicas.max++
scheduled[popped].max--
toDistributeMax--
continue
}
}
if toDistributeMax < rdc.max {
replicas.max += toDistributeMax
toDistributeMax = 0
break
}
replicas.max += rdc.max
toDistributeMax -= rdc.max
}
}
} else { // we have no new clusters but if still have max replicas to distribute;
// just distribute all in current clusters.
for toDistributeMax > 0 {
for cluster, replicas := range scheduled {
if replicas == nil {
replicas = &replicaNums{min: 0, max: 0}
scheduled[cluster] = replicas
}
// First give away max from clusters offering them.
// This case especially helps getting hpa into newly joining
// clusters.
if lists.availableMax.Len() > 0 {
popped, notEmpty := lists.availableMax.PopAny()
if notEmpty {
// Boundary checks have happened earlier in
// minReplicasReducible().
replicas.max++
scheduled[popped].max--
toDistributeMax--
continue
}
}
if toDistributeMax < rdc.max {
replicas.max += toDistributeMax
toDistributeMax = 0
break
}
replicas.max += rdc.max
toDistributeMax -= rdc.max
}
}
}
return toDistributeMax
}
// distributeMinReplicas
// Takes input:
// toDistributeMin: number of replicas to distribute.
// lists: cluster name lists, which have clusters with available max,
// available min and those with no hpas yet.
// rdc: replicadistributioncount for max and min.
// currentObjs: list of current cluster hpas.
// scheduled: schedule state which will be updated in place.
func (a *HpaAdapter) distributeMinReplicas(toDistributeMin int32, lists hpaLists, rdc replicaNums,
currentObjs map[string]pkgruntime.Object, scheduled map[string]*replicaNums) int32 {
for cluster, replicas := range scheduled {
if toDistributeMin == 0 {
break
}
// We have distriubted Max and thus scheduled might not be nil
// but probably current (what we got originally) is nil(no hpa)
if replicas == nil || currentObjs[cluster] == nil {
continue
}
if a.minReplicasIncreasable(currentObjs[cluster]) {
if lists.availableMin.Len() > 0 {
popped, notEmpty := lists.availableMin.PopAny()
if notEmpty {
// Boundary checks have happened earlier.
scheduled[popped].min--
replicas.min++
toDistributeMin--
}
}
}
}
if lists.noHpa.Len() > 0 {
// TODO: can this become an infinite loop?
for toDistributeMin > 0 {
for _, cluster := range lists.noHpa.UnsortedList() {
replicas := scheduled[cluster]
if replicas == nil {
// We did not get max here so this cluster
// remains without hpa
continue
}
var replicaNum int32 = 0
if toDistributeMin < rdc.min {
replicaNum = toDistributeMin
} else {
replicaNum = rdc.min
}
if (replicas.max - replicaNum) < replicas.min {
// Cannot increase the min in this cluster
// as it will go beyond max
continue
}
if lists.availableMin.Len() > 0 {
popped, notEmpty := lists.availableMin.PopAny()
if notEmpty {
// Boundary checks have happened earlier.
scheduled[popped].min--
replicas.min++
toDistributeMin--
continue
}
}
replicas.min += replicaNum
toDistributeMin -= replicaNum
}
}
} else { // we have no new clusters but if still have min replicas to distribute;
// just distribute all in current clusters.
for toDistributeMin > 0 {
for _, replicas := range scheduled {
if replicas == nil {
// We did not get max here so this cluster
// remains without hpa
continue
}
var replicaNum int32 = 0
if toDistributeMin < rdc.min {
replicaNum = toDistributeMin
} else {
replicaNum = rdc.min
}
if (replicas.max - replicaNum) < replicas.min {
// Cannot increase the min in this cluster
// as it will go beyond max
continue
}
if lists.availableMin.Len() > 0 {
popped, notEmpty := lists.availableMin.PopAny()
if notEmpty {
// Boundary checks have happened earlier.
scheduled[popped].min--
replicas.min++
toDistributeMin--
continue
}
}
replicas.min += replicaNum
toDistributeMin -= replicaNum
}
}
}
return toDistributeMin
}
// finaliseScheduleState ensures that the minReplica count is made to 1
// for those clusters which got max, but did not get min. This is because
// k8s hpa does not accept hpas with 0 min replicas.
// The replica num distribution can thus have more mins then fedHpa requested
// but its better then having all replicas go into one cluster (if fedHpa
// requested min=1 (which is the most usual case).
func finaliseScheduleState(scheduled map[string]*replicaNums) map[string]*replicaNums {
for _, replicas := range scheduled {
if (replicas != nil) && (replicas.min <= 0) && (replicas.max > 0) {
// Min total does not necessarily meet the federated min limit.
replicas.min = 1
}
}
return scheduled
}
// isPristine is used to determine if so far local controller has been
// able to really determine, what should be the desired replica number for
// this cluster.
// This is used to get hpas into those clusters which might join fresh,
// and so far other cluster hpas haven't really reached anywhere.
// TODO: There is a flaw here, that a just born object would also offer its
// replicas which can also lead to fast thrashing.
// The only better way is to either ensure that object creation time stamp is set
// and can be used authoritatively; or have another field on the local object
// which is mandatorily set on creation and can be used authoritatively.
// Should we abuse annotations again for this, or this can be a proper requirement?
func isPristine(hpa *autoscalingv1.HorizontalPodAutoscaler) bool {
if hpa.Status.LastScaleTime == nil &&
hpa.Status.DesiredReplicas == 0 {
return true
}
return false
}
// isScaleable tells if it already has been a reasonable amount of
// time since this hpa scaled. Its used to avoid fast thrashing.
func (a *HpaAdapter) isScaleable(hpa *autoscalingv1.HorizontalPodAutoscaler) bool {
if hpa.Status.LastScaleTime == nil {
return false
}
t := hpa.Status.LastScaleTime.Add(a.scaleForbiddenWindow)
if t.After(time.Now()) {
return false
}
return true
}
func (a *HpaAdapter) maxReplicasReducible(obj pkgruntime.Object) bool {
hpa := obj.(*autoscalingv1.HorizontalPodAutoscaler)
if (hpa.Spec.MinReplicas != nil) &&
(((hpa.Spec.MaxReplicas - 1) - *hpa.Spec.MinReplicas) < 0) {
return false
}
if isPristine(hpa) {
return true
}
if !a.isScaleable(hpa) {
return false
}
if (hpa.Status.DesiredReplicas < hpa.Status.CurrentReplicas) ||
((hpa.Status.DesiredReplicas == hpa.Status.CurrentReplicas) &&
(hpa.Status.DesiredReplicas < hpa.Spec.MaxReplicas)) {
return true
}
return false
}
// minReplicasReducible checks if this cluster (hpa) can offer replicas which are
// stuck here because of min limit.
// Its noteworthy, that min and max are adjusted separately, but if the replicas
// are not being used here, the max adjustment will lead it to become equal to min,
// but will not be able to scale down further and offer max to some other cluster
// which needs replicas.
func (a *HpaAdapter) minReplicasReducible(obj pkgruntime.Object) bool {
hpa := obj.(*autoscalingv1.HorizontalPodAutoscaler)
if isPristine(hpa) && (hpa.Spec.MinReplicas != nil) &&
(*hpa.Spec.MinReplicas > 1) &&
(*hpa.Spec.MinReplicas <= hpa.Spec.MaxReplicas) {
return true
}
if !a.isScaleable(hpa) {
return false
}
if (hpa.Spec.MinReplicas != nil) &&
(*hpa.Spec.MinReplicas > 1) &&
(hpa.Status.DesiredReplicas == hpa.Status.CurrentReplicas) &&
(hpa.Status.CurrentReplicas == *hpa.Spec.MinReplicas) {
return true
}
return false
}
func (a *HpaAdapter) maxReplicasNeeded(obj pkgruntime.Object) bool {
hpa := obj.(*autoscalingv1.HorizontalPodAutoscaler)
if !a.isScaleable(hpa) {
return false
}
if (hpa.Status.CurrentReplicas == hpa.Status.DesiredReplicas) &&
(hpa.Status.CurrentReplicas == hpa.Spec.MaxReplicas) {
return true
}
return false
}
func (a *HpaAdapter) minReplicasIncreasable(obj pkgruntime.Object) bool {
hpa := obj.(*autoscalingv1.HorizontalPodAutoscaler)
if !a.isScaleable(hpa) ||
((hpa.Spec.MinReplicas != nil) &&
(*hpa.Spec.MinReplicas) >= hpa.Spec.MaxReplicas) {
return false
}
if (hpa.Spec.MinReplicas != nil) &&
(hpa.Status.DesiredReplicas > *hpa.Spec.MinReplicas) {
return true
}
return false
}