-
Notifications
You must be signed in to change notification settings - Fork 273
/
azure_loadbalancer.go
3240 lines (2864 loc) · 122 KB
/
azure_loadbalancer.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
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
/*
Copyright 2020 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 provider
import (
"context"
"encoding/json"
"fmt"
"math"
"reflect"
"sort"
"strconv"
"strings"
"github.com/Azure/azure-sdk-for-go/services/network/mgmt/2020-08-01/network"
"github.com/Azure/go-autorest/autorest/to"
v1 "k8s.io/api/core/v1"
utilerrors "k8s.io/apimachinery/pkg/util/errors"
"k8s.io/apimachinery/pkg/util/sets"
cloudprovider "k8s.io/cloud-provider"
servicehelpers "k8s.io/cloud-provider/service/helpers"
"k8s.io/klog/v2"
utilnet "k8s.io/utils/net"
azcache "sigs.k8s.io/cloud-provider-azure/pkg/cache"
"sigs.k8s.io/cloud-provider-azure/pkg/consts"
"sigs.k8s.io/cloud-provider-azure/pkg/metrics"
"sigs.k8s.io/cloud-provider-azure/pkg/retry"
)
// GetLoadBalancer returns whether the specified load balancer and its components exist, and
// if so, what its status is.
func (az *Cloud) GetLoadBalancer(ctx context.Context, clusterName string, service *v1.Service) (status *v1.LoadBalancerStatus, exists bool, err error) {
// Since public IP is not a part of the load balancer on Azure,
// there is a chance that we could orphan public IP resources while we delete the load blanacer (kubernetes/kubernetes#80571).
// We need to make sure the existence of the load balancer depends on the load balancer resource and public IP resource on Azure.
existsPip := func() bool {
pipName, _, err := az.determinePublicIPName(clusterName, service)
if err != nil {
return false
}
pipResourceGroup := az.getPublicIPAddressResourceGroup(service)
_, existsPip, err := az.getPublicIPAddress(pipResourceGroup, pipName)
if err != nil {
return false
}
return existsPip
}()
_, status, existsLb, err := az.getServiceLoadBalancer(service, clusterName, nil, false, []network.LoadBalancer{})
if err != nil {
return nil, existsPip, err
}
// Return exists = false only if the load balancer and the public IP are not found on Azure
if !existsLb && !existsPip {
serviceName := getServiceName(service)
klog.V(5).Infof("getloadbalancer (cluster:%s) (service:%s) - doesn't exist", clusterName, serviceName)
return nil, false, nil
}
// Return exists = true if either the load balancer or the public IP (or both) exists
return status, true, nil
}
func getPublicIPDomainNameLabel(service *v1.Service) (string, bool) {
if labelName, found := service.Annotations[consts.ServiceAnnotationDNSLabelName]; found {
return labelName, found
}
return "", false
}
// EnsureLoadBalancer creates a new load balancer 'name', or updates the existing one. Returns the status of the balancer
func (az *Cloud) EnsureLoadBalancer(ctx context.Context, clusterName string, service *v1.Service, nodes []*v1.Node) (*v1.LoadBalancerStatus, error) {
// When a client updates the internal load balancer annotation,
// the service may be switched from an internal LB to a public one, or vise versa.
// Here we'll firstly ensure service do not lie in the opposite LB.
serviceName := getServiceName(service)
klog.V(5).Infof("ensureloadbalancer(%s): START clusterName=%q", serviceName, clusterName)
mc := metrics.NewMetricContext("services", "ensure_loadbalancer", az.ResourceGroup, az.SubscriptionID, serviceName)
isOperationSucceeded := false
defer func() {
mc.ObserveOperationWithResult(isOperationSucceeded)
}()
lb, err := az.reconcileLoadBalancer(clusterName, service, nodes, true /* wantLb */)
if err != nil {
klog.Errorf("reconcileLoadBalancer(%s) failed: %v", serviceName, err)
return nil, err
}
lbStatus, _, err := az.getServiceLoadBalancerStatus(service, lb)
if err != nil {
klog.Errorf("getServiceLoadBalancerStatus(%s) failed: %v", serviceName, err)
return nil, err
}
var serviceIP *string
if lbStatus != nil && len(lbStatus.Ingress) > 0 {
serviceIP = &lbStatus.Ingress[0].IP
}
klog.V(2).Infof("EnsureLoadBalancer: reconciling security group for service %q with IP %q, wantLb = true", serviceName, logSafe(serviceIP))
if _, err := az.reconcileSecurityGroup(clusterName, service, serviceIP, true /* wantLb */); err != nil {
klog.Errorf("reconcileSecurityGroup(%s) failed: %#v", serviceName, err)
return nil, err
}
updateService := updateServiceLoadBalancerIP(service, to.String(serviceIP))
flippedService := flipServiceInternalAnnotation(updateService)
if _, err := az.reconcileLoadBalancer(clusterName, flippedService, nil, false /* wantLb */); err != nil {
klog.Errorf("reconcileLoadBalancer(%s) failed: %#v", serviceName, err)
return nil, err
}
// lb is not reused here because the ETAG may be changed in above operations, hence reconcilePublicIP() would get lb again from cache.
klog.V(2).Infof("EnsureLoadBalancer: reconciling pip")
if _, err := az.reconcilePublicIP(clusterName, updateService, to.String(lb.Name), true /* wantLb */); err != nil {
klog.Errorf("reconcilePublicIP(%s) failed: %#v", serviceName, err)
return nil, err
}
isOperationSucceeded = true
return lbStatus, nil
}
// UpdateLoadBalancer updates hosts under the specified load balancer.
func (az *Cloud) UpdateLoadBalancer(ctx context.Context, clusterName string, service *v1.Service, nodes []*v1.Node) error {
if !az.shouldUpdateLoadBalancer(clusterName, service, nodes) {
klog.V(2).Infof("UpdateLoadBalancer: skipping service %s because it is either being deleted or does not exist anymore", service.Name)
return nil
}
_, err := az.EnsureLoadBalancer(ctx, clusterName, service, nodes)
return err
}
// EnsureLoadBalancerDeleted deletes the specified load balancer if it
// exists, returning nil if the load balancer specified either didn't exist or
// was successfully deleted.
// This construction is useful because many cloud providers' load balancers
// have multiple underlying components, meaning a Get could say that the LB
// doesn't exist even if some part of it is still laying around.
func (az *Cloud) EnsureLoadBalancerDeleted(ctx context.Context, clusterName string, service *v1.Service) error {
isInternal := requiresInternalLoadBalancer(service)
serviceName := getServiceName(service)
klog.V(5).Infof("Delete service (%s): START clusterName=%q", serviceName, clusterName)
mc := metrics.NewMetricContext("services", "ensure_loadbalancer_deleted", az.ResourceGroup, az.SubscriptionID, serviceName)
isOperationSucceeded := false
defer func() {
mc.ObserveOperationWithResult(isOperationSucceeded)
}()
serviceIPToCleanup, err := az.findServiceIPAddress(ctx, clusterName, service, isInternal)
if err != nil && !retry.HasStatusForbiddenOrIgnoredError(err) {
return err
}
klog.V(2).Infof("EnsureLoadBalancerDeleted: reconciling security group for service %q with IP %q, wantLb = false", serviceName, serviceIPToCleanup)
if _, err := az.reconcileSecurityGroup(clusterName, service, &serviceIPToCleanup, false /* wantLb */); err != nil {
return err
}
if _, err := az.reconcileLoadBalancer(clusterName, service, nil, false /* wantLb */); err != nil && !retry.HasStatusForbiddenOrIgnoredError(err) {
return err
}
if _, err := az.reconcilePublicIP(clusterName, service, "", false /* wantLb */); err != nil {
return err
}
klog.V(2).Infof("Delete service (%s): FINISH", serviceName)
isOperationSucceeded = true
return nil
}
// GetLoadBalancerName returns the LoadBalancer name.
func (az *Cloud) GetLoadBalancerName(ctx context.Context, clusterName string, service *v1.Service) string {
return cloudprovider.DefaultLoadBalancerName(service)
}
func (az *Cloud) getLoadBalancerResourceGroup() string {
if az.LoadBalancerResourceGroup != "" {
return az.LoadBalancerResourceGroup
}
return az.ResourceGroup
}
// cleanupVMSetFromBackendPoolByCondition removes nodes of the unwanted vmSet from the lb backend pool.
// This is needed in two scenarios:
// 1. When migrating from single SLB to multiple SLBs, the existing
// SLB's backend pool contains nodes from different agent pools, while we only want the
// nodes from the primary agent pool to join the backend pool.
// 2. When migrating from dedicated SLB to shared SLB (or vice versa), we should move the vmSet from
// one SLB to another one.
func (az *Cloud) cleanupVMSetFromBackendPoolByCondition(slb *network.LoadBalancer, service *v1.Service, clusterName string, shouldRemoveVMSetFromSLB func(string) bool) (*network.LoadBalancer, error) {
lbBackendPoolName := getBackendPoolName(clusterName, service)
lbResourceGroup := az.getLoadBalancerResourceGroup()
lbBackendPoolID := az.getBackendPoolID(to.String(slb.Name), lbResourceGroup, lbBackendPoolName)
newBackendPools := make([]network.BackendAddressPool, 0)
if slb.LoadBalancerPropertiesFormat != nil && slb.BackendAddressPools != nil {
newBackendPools = *slb.BackendAddressPools
}
vmSetNameToBackendIPConfigurationsToBeDeleted := make(map[string][]network.InterfaceIPConfiguration)
for j, bp := range newBackendPools {
if strings.EqualFold(to.String(bp.Name), lbBackendPoolName) {
klog.V(2).Infof("cleanupVMSetFromBackendPoolByCondition: checking the backend pool %s from standard load balancer %s", to.String(bp.Name), to.String(slb.Name))
if bp.BackendAddressPoolPropertiesFormat != nil && bp.BackendIPConfigurations != nil {
for i := len(*bp.BackendIPConfigurations) - 1; i >= 0; i-- {
ipConf := (*bp.BackendIPConfigurations)[i]
ipConfigID := to.String(ipConf.ID)
_, vmSetName, err := az.VMSet.GetNodeNameByIPConfigurationID(ipConfigID)
if err != nil {
return nil, err
}
if shouldRemoveVMSetFromSLB(vmSetName) {
klog.V(2).Infof("cleanupVMSetFromBackendPoolByCondition: found unwanted vmSet %s, decouple it from the LB", vmSetName)
// construct a backendPool that only contains the IP config of the node to be deleted
interfaceIPConfigToBeDeleted := network.InterfaceIPConfiguration{
ID: to.StringPtr(ipConfigID),
}
vmSetNameToBackendIPConfigurationsToBeDeleted[vmSetName] = append(vmSetNameToBackendIPConfigurationsToBeDeleted[vmSetName], interfaceIPConfigToBeDeleted)
*bp.BackendIPConfigurations = append((*bp.BackendIPConfigurations)[:i], (*bp.BackendIPConfigurations)[i+1:]...)
}
}
}
newBackendPools[j] = bp
break
}
}
for vmSetName := range vmSetNameToBackendIPConfigurationsToBeDeleted {
backendIPConfigurationsToBeDeleted := vmSetNameToBackendIPConfigurationsToBeDeleted[vmSetName]
backendpoolToBeDeleted := &[]network.BackendAddressPool{
{
ID: to.StringPtr(lbBackendPoolID),
BackendAddressPoolPropertiesFormat: &network.BackendAddressPoolPropertiesFormat{
BackendIPConfigurations: &backendIPConfigurationsToBeDeleted,
},
},
}
// decouple the backendPool from the node
err := az.VMSet.EnsureBackendPoolDeleted(service, lbBackendPoolID, vmSetName, backendpoolToBeDeleted)
if err != nil {
return nil, err
}
slb.BackendAddressPools = &newBackendPools
// Proactively disable the etag to prevent etag mismatch error when putting lb later.
// This could be happen because when we remove the hosts from the lb, the nrp
// would put the lb to remove the backend references as well.
slb.Etag = nil
}
return slb, nil
}
// shouldChangeLoadBalancer determines if the load balancer of the service should be switched to another one
// according to the mode annotation on the service. This could be happened when the LB selection mode of an
// existing service is changed to another VMSS/VMAS.
func (az *Cloud) shouldChangeLoadBalancer(service *v1.Service, currLBName, clusterName string) bool {
hasMode, isAuto, vmSetName := az.getServiceLoadBalancerMode(service)
// if no mode is given or the mode is `__auto__`, the current LB should be kept
if !hasMode || isAuto {
return false
}
// if using the single standard load balancer, the current LB should be kept
useSingleSLB := az.useStandardLoadBalancer() && !az.EnableMultipleStandardLoadBalancers
if useSingleSLB {
return false
}
// if the current LB is what we want, keep it
lbName := strings.TrimSuffix(currLBName, consts.InternalLoadBalancerNameSuffix)
if strings.EqualFold(lbName, vmSetName) {
return false
}
if strings.EqualFold(vmSetName, az.VMSet.GetPrimaryVMSetName()) && strings.EqualFold(clusterName, lbName) {
return false
}
// if the vmSet selected by the annotation is sharing the primary slb, and the service
// has been associated to the primary slb, keep it
useMultipleSLBs := az.useStandardLoadBalancer() && az.EnableMultipleStandardLoadBalancers
if useMultipleSLBs &&
az.getVMSetNamesSharingPrimarySLB().Has(strings.ToLower(vmSetName)) &&
strings.EqualFold(lbName, clusterName) {
return false
}
// if the VMSS/VMAS of the current LB is different from the mode, change the LB
// to another one
klog.V(2).Infof("shouldChangeLoadBalancer(%s, %s, %s): change the LB to another one", service.Name, currLBName, clusterName)
return true
}
func (az *Cloud) removeFrontendIPConfigurationFromLoadBalancer(lb *network.LoadBalancer, fip *network.FrontendIPConfiguration, clusterName string, service *v1.Service) error {
if lb == nil || lb.LoadBalancerPropertiesFormat == nil || lb.FrontendIPConfigurations == nil {
return nil
}
fipConfigs := *lb.FrontendIPConfigurations
for i, fipConfig := range fipConfigs {
if strings.EqualFold(to.String(fipConfig.Name), to.String(fip.Name)) {
fipConfigs = append(fipConfigs[:i], fipConfigs[i+1:]...)
break
}
}
lb.FrontendIPConfigurations = &fipConfigs
// also remove the corresponding rules/probes
if lb.LoadBalancingRules != nil {
lbRules := *lb.LoadBalancingRules
for i := len(lbRules) - 1; i >= 0; i-- {
if strings.Contains(to.String(lbRules[i].Name), to.String(fip.Name)) {
lbRules = append(lbRules[:i], lbRules[i+1:]...)
}
}
lb.LoadBalancingRules = &lbRules
}
if lb.Probes != nil {
lbProbes := *lb.Probes
for i := len(lbProbes) - 1; i >= 0; i-- {
if strings.Contains(to.String(lbProbes[i].Name), to.String(fip.Name)) {
lbProbes = append(lbProbes[:i], lbProbes[i+1:]...)
}
}
lb.Probes = &lbProbes
}
if len(fipConfigs) == 0 {
klog.V(2).Infof("removeFrontendIPConfigurationFromLoadBalancer(%s, %s, %s, %s): deleting load balancer because there is no remaining frontend IP configurations", to.String(lb.Name), to.String(fip.Name), clusterName, service.Name)
err := az.cleanOrphanedLoadBalancer(lb, service, clusterName)
if err != nil {
klog.Errorf("removeFrontendIPConfigurationFromLoadBalancer(%s, %s, %s, %s): failed to cleanupOrphanedLoadBalancer: %v", to.String(lb.Name), to.String(fip.Name), clusterName, service.Name, err)
return err
}
} else {
klog.V(2).Infof("removeFrontendIPConfigurationFromLoadBalancer(%s, %s, %s, %s): updating the load balancer", to.String(lb.Name), to.String(fip.Name), clusterName, service.Name)
err := az.CreateOrUpdateLB(service, *lb)
if err != nil {
klog.Errorf("removeFrontendIPConfigurationFromLoadBalancer(%s, %s, %s, %s): failed to CreateOrUpdateLB: %v", to.String(lb.Name), to.String(fip.Name), clusterName, service.Name, err)
return err
}
_ = az.lbCache.Delete(to.String(lb.Name))
}
return nil
}
func (az *Cloud) cleanOrphanedLoadBalancer(lb *network.LoadBalancer, service *v1.Service, clusterName string) error {
lbName := to.String(lb.Name)
serviceName := getServiceName(service)
isBackendPoolPreConfigured := az.isBackendPoolPreConfigured(service)
lbResourceGroup := az.getLoadBalancerResourceGroup()
lbBackendPoolName := getBackendPoolName(clusterName, service)
lbBackendPoolID := az.getBackendPoolID(lbName, lbResourceGroup, lbBackendPoolName)
if isBackendPoolPreConfigured {
klog.V(2).Infof("cleanOrphanedLoadBalancer(%s, %s, %s): ignore cleanup of dirty lb because the lb is pre-configured", lbName, serviceName, clusterName)
} else {
// When FrontendIPConfigurations is empty, we need to delete the Azure load balancer resource itself,
// because an Azure load balancer cannot have an empty FrontendIPConfigurations collection
klog.V(2).Infof("cleanOrphanedLoadBalancer(%s, %s, %s): deleting the LB since there are no remaining frontendIPConfigurations", lbName, serviceName, clusterName)
// Remove backend pools from vmSets. This is required for virtual machine scale sets before removing the LB.
vmSetName := az.mapLoadBalancerNameToVMSet(lbName, clusterName)
if _, ok := az.VMSet.(*availabilitySet); ok {
// do nothing for availability set
lb.BackendAddressPools = nil
}
err := az.VMSet.EnsureBackendPoolDeleted(service, lbBackendPoolID, vmSetName, lb.BackendAddressPools)
if err != nil {
klog.Errorf("cleanOrphanedLoadBalancer(%s, %s, %s): failed to EnsureBackendPoolDeleted: %v", lbName, serviceName, clusterName, err)
return err
}
klog.V(10).Infof("cleanOrphanedLoadBalancer(%s, %s, %s): EnsureBackendPoolDeleted finished", lbName, serviceName, clusterName)
// Remove the LB.
klog.V(10).Infof("cleanOrphanedLoadBalancer(%s, %s, %s): az.DeleteLB starts", lbName, serviceName, clusterName)
deleteErr := az.DeleteLB(service, lbName)
if deleteErr != nil {
klog.Warningf("cleanOrphanedLoadBalancer(%s, %s, %s): failed to DeleteLB: %v", lbName, serviceName, clusterName, deleteErr)
rgName, vmssName, parseErr := retry.GetVMSSMetadataByRawError(deleteErr.Error())
if parseErr != nil {
klog.Warningf("cleanOrphanedLoadBalancer(%s, %s, %s): failed to parse error: %v", lbName, serviceName, clusterName, parseErr)
return deleteErr
}
if rgName == "" || vmssName == "" {
klog.Warningf("cleanOrphanedLoadBalancer(%s, %s, %s): empty rgName or vmssName", lbName, serviceName, clusterName)
return deleteErr
}
// if we reach here, it means the VM couldn't be deleted because it is being referenced by a VMSS
if _, ok := az.VMSet.(*ScaleSet); !ok {
klog.Warningf("cleanOrphanedLoadBalancer(%s, %s, %s): unexpected VMSet type, expected VMSS", lbName, serviceName, clusterName)
return deleteErr
}
if !strings.EqualFold(rgName, az.ResourceGroup) {
return fmt.Errorf("cleanOrphanedLoadBalancer(%s, %s, %s): the VMSS %s is in the resource group %s, but is referencing the LB in %s", lbName, serviceName, clusterName, vmssName, rgName, az.ResourceGroup)
}
vmssNamesMap := map[string]bool{vmssName: true}
err := az.VMSet.EnsureBackendPoolDeletedFromVMSets(vmssNamesMap, lbBackendPoolID)
if err != nil {
klog.Errorf("cleanOrphanedLoadBalancer(%s, %s, %s): failed to EnsureBackendPoolDeletedFromVMSets: %v", lbName, serviceName, clusterName, err)
return err
}
deleteErr := az.DeleteLB(service, lbName)
if deleteErr != nil {
klog.Errorf("cleanOrphanedLoadBalancer(%s, %s, %s): failed delete lb for the second time, stop retrying: %v", lbName, serviceName, clusterName, deleteErr)
return deleteErr
}
}
klog.V(10).Infof("cleanOrphanedLoadBalancer(%s, %s, %s): az.DeleteLB finished", lbName, serviceName, clusterName)
}
return nil
}
// safeDeleteLoadBalancer deletes the load balancer after decoupling it from the vmSet
func (az *Cloud) safeDeleteLoadBalancer(lb network.LoadBalancer, clusterName, vmSetName string, service *v1.Service) error {
lbBackendPoolID := az.getBackendPoolID(to.String(lb.Name), az.getLoadBalancerResourceGroup(), getBackendPoolName(clusterName, service))
err := az.VMSet.EnsureBackendPoolDeleted(service, lbBackendPoolID, vmSetName, lb.BackendAddressPools)
if err != nil {
return fmt.Errorf("deleteDedicatedLoadBalancer: failed to EnsureBackendPoolDeleted: %w", err)
}
klog.V(2).Infof("deleteDedicatedLoadBalancer: deleting LB %s because the corresponding vmSet is supposed to be in the primary SLB", to.String(lb.Name))
err = az.DeleteLB(service, to.String(lb.Name))
if err != nil {
return fmt.Errorf("deleteDedicatedLoadBalancer : failed to DeleteLB: %w", err)
}
_ = az.lbCache.Delete(to.String(lb.Name))
return nil
}
func extractBackendIPConfigurationIDsFromLB(lb network.LoadBalancer, lbBackendPoolName string) []string {
result := make([]string, 0)
if lb.LoadBalancerPropertiesFormat != nil &&
lb.BackendAddressPools != nil {
for i := 0; i < len(*lb.BackendAddressPools); i++ {
backendPool := (*lb.BackendAddressPools)[i]
if strings.EqualFold(to.String(backendPool.Name), lbBackendPoolName) {
if backendPool.BackendAddressPoolPropertiesFormat != nil &&
backendPool.BackendIPConfigurations != nil {
for _, ipConfiguration := range *backendPool.BackendIPConfigurations {
if ipConfiguration.ID != nil {
result = append(result, to.String(ipConfiguration.ID))
}
}
}
}
}
}
return result
}
// reconcileSharedLoadBalancer deletes the dedicated SLBs of the non-primary vmSets. There are
// two scenarios where this operation is needed:
// 1. Using multiple slbs and the vmSet is supposed to share the primary slb.
// 2. When migrating from multiple slbs to single slb mode.
// It also ensures those vmSets are joint the backend pools of the primary SLBs.
// It runs only once everytime the cloud controller manager restarts.
func (az *Cloud) reconcileSharedLoadBalancer(service *v1.Service, clusterName string, nodes []*v1.Node) ([]network.LoadBalancer, error) {
var (
primarySLBs, existingLBs []network.LoadBalancer
changed bool
err error
)
// skip this operation when wantLb=false
if nodes == nil {
return nil, nil
}
// only run once since the controller manager rebooted
if az.isSharedLoadBalancerSynced {
return nil, nil
}
defer func() {
if err == nil {
az.isSharedLoadBalancerSynced = true
}
}()
// skip if the cluster is using basic LB
if !az.useStandardLoadBalancer() {
return nil, nil
}
existingLBs, err = az.ListAgentPoolLBs(service, nodes, clusterName)
if err != nil {
return nil, fmt.Errorf("reconcileSharedLoadBalancer: failed to list LB: %w", err)
}
lbBackendPoolName := getBackendPoolName(clusterName, service)
lbNamesToBeDeleted := sets.NewString()
// loop 1: delete unwanted LBs
for _, lb := range existingLBs {
lbNamePrefix := strings.TrimSuffix(to.String(lb.Name), consts.InternalLoadBalancerNameSuffix)
// skip the internal or external primary load balancer
if strings.EqualFold(lbNamePrefix, clusterName) {
primarySLBs = append(primarySLBs, lb)
continue
}
// skip if the multiple slbs mode is enabled and
// the vmSet is supposed to have dedicated SLBs
vmSetName := strings.ToLower(az.mapLoadBalancerNameToVMSet(to.String(lb.Name), clusterName))
if az.EnableMultipleStandardLoadBalancers && !az.getVMSetNamesSharingPrimarySLB().Has(vmSetName) {
continue
}
// For non-primary load balancer, the lb name is the name of the VMSet.
// If the VMSet name is in az.NodePoolsWithoutDedicatedSLB, we should
// decouple the VMSet from the lb and delete the lb. Then adding the VMSet
// to the backend pool of the primary slb.
err = az.safeDeleteLoadBalancer(lb, clusterName, vmSetName, service)
if err != nil {
return nil, err
}
// remove the deleted lb from the list and construct a new primary
// lb, so that getServiceLoadBalancer doesn't have to call list api again
lbNamesToBeDeleted.Insert(to.String(lb.Name))
changed = true
}
if !changed {
klog.V(4).Infof("reconcileSharedLoadBalancer: no changes made, return now")
return existingLBs, nil
}
ipConfigIDsToBeAddedToPrimarySLB := sets.NewString()
// loop2: add nodes to the backend pool of the primary SLBs
for i := len(existingLBs) - 1; i >= 0; i-- {
lb := existingLBs[i]
if !lbNamesToBeDeleted.Has(to.String(lb.Name)) {
continue
}
vmSetName := strings.ToLower(az.mapLoadBalancerNameToVMSet(to.String(lb.Name), clusterName))
isInternalLB := strings.HasSuffix(to.String(lb.Name), consts.InternalLoadBalancerNameSuffix)
primarySLBName := clusterName
if isInternalLB {
primarySLBName = fmt.Sprintf("%s%s", clusterName, consts.InternalLoadBalancerNameSuffix)
}
primaryLBBackendPoolID := az.getBackendPoolID(primarySLBName, az.getLoadBalancerResourceGroup(), getBackendPoolName(clusterName, service))
klog.V(2).Infof("reconcileSharedLoadBalancer: binding the vmSet %s to the backend pool %s", vmSetName, primaryLBBackendPoolID)
err = az.VMSet.EnsureHostsInPool(service, nodes, primaryLBBackendPoolID, vmSetName, false)
if err != nil {
return nil, fmt.Errorf("reconcileSharedLoadBalancer: failed to EnsureHostsInPool: %w", err)
}
for _, id := range extractBackendIPConfigurationIDsFromLB(lb, lbBackendPoolName) {
ipConfigIDsToBeAddedToPrimarySLB.Insert(id)
}
// remove the deleted LB from the list
existingLBs = append(existingLBs[:i], existingLBs[i+1:]...)
}
for _, primarySLB := range primarySLBs {
if primarySLB.LoadBalancerPropertiesFormat != nil &&
primarySLB.BackendAddressPools != nil {
for i := 0; i < len(*primarySLB.BackendAddressPools); i++ {
if strings.EqualFold(to.String((*primarySLB.BackendAddressPools)[i].Name), lbBackendPoolName) {
backendPoolIPConfigs := (*primarySLB.BackendAddressPools)[i].BackendIPConfigurations
for _, id := range ipConfigIDsToBeAddedToPrimarySLB.List() {
*backendPoolIPConfigs = append(*backendPoolIPConfigs, network.InterfaceIPConfiguration{
ID: to.StringPtr(id),
})
}
break
}
}
}
}
for i, existingLB := range existingLBs {
for _, primarySLB := range primarySLBs {
if strings.EqualFold(to.String(existingLB.Name), to.String(primarySLB.Name)) {
// Proactively disable the etag to prevent etag mismatch error when putting lb later.
// This could be happen because when we remove the hosts from the lb, the nrp
// would put the lb to remove the backend references as well.
primarySLB.Etag = nil
existingLBs[i] = primarySLB
}
}
}
return existingLBs, nil
}
// getServiceLoadBalancer gets the loadbalancer for the service if it already exists.
// If wantLb is TRUE then -it selects a new load balancer.
// In case the selected load balancer does not exist it returns network.LoadBalancer struct
// with added metadata (such as name, location) and existsLB set to FALSE.
// By default - cluster default LB is returned.
func (az *Cloud) getServiceLoadBalancer(service *v1.Service, clusterName string, nodes []*v1.Node, wantLb bool, existingLBs []network.LoadBalancer) (lb *network.LoadBalancer, status *v1.LoadBalancerStatus, exists bool, err error) {
isInternal := requiresInternalLoadBalancer(service)
var defaultLB *network.LoadBalancer
primaryVMSetName := az.VMSet.GetPrimaryVMSetName()
defaultLBName := az.getAzureLoadBalancerName(clusterName, primaryVMSetName, isInternal)
useMultipleSLBs := az.useStandardLoadBalancer() && az.EnableMultipleStandardLoadBalancers
// reuse the lb list from reconcileSharedLoadBalancer to reduce the api call
if len(existingLBs) == 0 {
existingLBs, err = az.ListLB(service)
if err != nil {
return nil, nil, false, err
}
}
// check if the service already has a load balancer
for i := range existingLBs {
existingLB := existingLBs[i]
existingLBNamePrefix := strings.TrimSuffix(to.String(existingLB.Name), consts.InternalLoadBalancerNameSuffix)
// for the primary standard load balancer (internal or external), when enabled multiple slbs
if strings.EqualFold(existingLBNamePrefix, clusterName) && useMultipleSLBs {
shouldRemoveVMSetFromSLB := func(vmSetName string) bool {
// not removing the vmSet from the primary SLB
// if it is supposed to share the primary SLB.
if az.getVMSetNamesSharingPrimarySLB().Has(strings.ToLower(vmSetName)) {
return false
}
// removing the vmSet from the primary SLB if
// it is not the primary vmSet. There are two situations:
// 1. when migrating from single SLB to multiple SLBs, we
// need to remove all non-primary vmSets from the primary SLB;
// 2. when migrating from shared mode to dedicated SLB, we
// need to remove the specific vmSet from the primary SLB.
return !strings.EqualFold(vmSetName, primaryVMSetName) && vmSetName != ""
}
cleanedLB, err := az.cleanupVMSetFromBackendPoolByCondition(&existingLB, service, clusterName, shouldRemoveVMSetFromSLB)
if err != nil {
return nil, nil, false, err
}
existingLB = *cleanedLB
existingLBs[i] = *cleanedLB
}
if strings.EqualFold(*existingLB.Name, defaultLBName) {
defaultLB = &existingLB
}
if isInternalLoadBalancer(&existingLB) != isInternal {
continue
}
status, fipConfig, err := az.getServiceLoadBalancerStatus(service, &existingLB)
if err != nil {
return nil, nil, false, err
}
if status == nil {
// service is not on this load balancer
continue
}
// select another load balancer instead of returning
// the current one if the change is needed
if wantLb && az.shouldChangeLoadBalancer(service, to.String(existingLB.Name), clusterName) {
if err := az.removeFrontendIPConfigurationFromLoadBalancer(&existingLB, fipConfig, clusterName, service); err != nil {
klog.Errorf("getServiceLoadBalancer(%s, %s, %v): failed to remove frontend IP configuration from load balancer: %v", service.Name, clusterName, wantLb, err)
return nil, nil, false, err
}
break
}
return &existingLB, status, true, nil
}
// Service does not have a load balancer, select one.
// Single standard load balancer doesn't need this because
// all backends nodes should be added to same LB.
useSingleSLB := az.useStandardLoadBalancer() && !az.EnableMultipleStandardLoadBalancers
if wantLb && !useSingleSLB {
// select new load balancer for service
selectedLB, exists, err := az.selectLoadBalancer(clusterName, service, &existingLBs, nodes)
if err != nil {
return nil, nil, false, err
}
return selectedLB, nil, exists, err
}
// create a default LB with meta data if not present
if defaultLB == nil {
defaultLB = &network.LoadBalancer{
Name: &defaultLBName,
Location: &az.Location,
LoadBalancerPropertiesFormat: &network.LoadBalancerPropertiesFormat{},
}
if az.useStandardLoadBalancer() {
defaultLB.Sku = &network.LoadBalancerSku{
Name: network.LoadBalancerSkuNameStandard,
}
}
if az.HasExtendedLocation() {
defaultLB.ExtendedLocation = &network.ExtendedLocation{
Name: &az.ExtendedLocationName,
Type: &az.ExtendedLocationType,
}
}
}
return defaultLB, nil, false, nil
}
// selectLoadBalancer selects load balancer for the service in the cluster.
// The selection algorithm selects the load balancer which currently has
// the minimum lb rules. If there are multiple LBs with same number of rules,
// then selects the first one (sorted based on name).
func (az *Cloud) selectLoadBalancer(clusterName string, service *v1.Service, existingLBs *[]network.LoadBalancer, nodes []*v1.Node) (selectedLB *network.LoadBalancer, existsLb bool, err error) {
isInternal := requiresInternalLoadBalancer(service)
serviceName := getServiceName(service)
klog.V(2).Infof("selectLoadBalancer for service (%s): isInternal(%v) - start", serviceName, isInternal)
vmSetNames, err := az.VMSet.GetVMSetNames(service, nodes)
if err != nil {
klog.Errorf("az.selectLoadBalancer: cluster(%s) service(%s) isInternal(%t) - az.GetVMSetNames failed, err=(%v)", clusterName, serviceName, isInternal, err)
return nil, false, err
}
klog.V(2).Infof("selectLoadBalancer: cluster(%s) service(%s) isInternal(%t) - vmSetNames %v", clusterName, serviceName, isInternal, *vmSetNames)
mapExistingLBs := map[string]network.LoadBalancer{}
for _, lb := range *existingLBs {
mapExistingLBs[*lb.Name] = lb
}
selectedLBRuleCount := math.MaxInt32
for _, currVMSetName := range *vmSetNames {
currLBName := az.getAzureLoadBalancerName(clusterName, currVMSetName, isInternal)
lb, exists := mapExistingLBs[currLBName]
if !exists {
// select this LB as this is a new LB and will have minimum rules
// create tmp lb struct to hold metadata for the new load-balancer
var loadBalancerSKU network.LoadBalancerSkuName
if az.useStandardLoadBalancer() {
loadBalancerSKU = network.LoadBalancerSkuNameStandard
} else {
loadBalancerSKU = network.LoadBalancerSkuNameBasic
}
selectedLB = &network.LoadBalancer{
Name: &currLBName,
Location: &az.Location,
Sku: &network.LoadBalancerSku{Name: loadBalancerSKU},
LoadBalancerPropertiesFormat: &network.LoadBalancerPropertiesFormat{},
}
if az.HasExtendedLocation() {
selectedLB.ExtendedLocation = &network.ExtendedLocation{
Name: &az.ExtendedLocationName,
Type: &az.ExtendedLocationType,
}
}
return selectedLB, false, nil
}
lbRules := *lb.LoadBalancingRules
currLBRuleCount := 0
if lbRules != nil {
currLBRuleCount = len(lbRules)
}
if currLBRuleCount < selectedLBRuleCount {
selectedLBRuleCount = currLBRuleCount
selectedLB = &lb
}
}
if selectedLB == nil {
err = fmt.Errorf("selectLoadBalancer: cluster(%s) service(%s) isInternal(%t) - unable to find load balancer for selected VM sets %v", clusterName, serviceName, isInternal, *vmSetNames)
klog.Error(err)
return nil, false, err
}
// validate if the selected LB has not exceeded the MaximumLoadBalancerRuleCount
if az.Config.MaximumLoadBalancerRuleCount != 0 && selectedLBRuleCount >= az.Config.MaximumLoadBalancerRuleCount {
err = fmt.Errorf("selectLoadBalancer: cluster(%s) service(%s) isInternal(%t) - all available load balancers have exceeded maximum rule limit %d, vmSetNames (%v)", clusterName, serviceName, isInternal, selectedLBRuleCount, *vmSetNames)
klog.Error(err)
return selectedLB, existsLb, err
}
return selectedLB, existsLb, nil
}
func (az *Cloud) getServiceLoadBalancerStatus(service *v1.Service, lb *network.LoadBalancer) (status *v1.LoadBalancerStatus, fipConfig *network.FrontendIPConfiguration, err error) {
if lb == nil {
klog.V(10).Info("getServiceLoadBalancerStatus: lb is nil")
return nil, nil, nil
}
if lb.FrontendIPConfigurations == nil || *lb.FrontendIPConfigurations == nil {
klog.V(10).Info("getServiceLoadBalancerStatus: lb.FrontendIPConfigurations is nil")
return nil, nil, nil
}
isInternal := requiresInternalLoadBalancer(service)
serviceName := getServiceName(service)
for _, ipConfiguration := range *lb.FrontendIPConfigurations {
owns, isPrimaryService, err := az.serviceOwnsFrontendIP(ipConfiguration, service)
if err != nil {
return nil, nil, fmt.Errorf("get(%s): lb(%s) - failed to filter frontend IP configs with error: %w", serviceName, to.String(lb.Name), err)
}
if owns {
klog.V(2).Infof("get(%s): lb(%s) - found frontend IP config, primary service: %v", serviceName, to.String(lb.Name), isPrimaryService)
var lbIP *string
if isInternal {
lbIP = ipConfiguration.PrivateIPAddress
} else {
if ipConfiguration.PublicIPAddress == nil {
return nil, nil, fmt.Errorf("get(%s): lb(%s) - failed to get LB PublicIPAddress is Nil", serviceName, *lb.Name)
}
pipID := ipConfiguration.PublicIPAddress.ID
if pipID == nil {
return nil, nil, fmt.Errorf("get(%s): lb(%s) - failed to get LB PublicIPAddress ID is Nil", serviceName, *lb.Name)
}
pipName, err := getLastSegment(*pipID, "/")
if err != nil {
return nil, nil, fmt.Errorf("get(%s): lb(%s) - failed to get LB PublicIPAddress Name from ID(%s)", serviceName, *lb.Name, *pipID)
}
pip, existsPip, err := az.getPublicIPAddress(az.getPublicIPAddressResourceGroup(service), pipName)
if err != nil {
return nil, nil, err
}
if existsPip {
lbIP = pip.IPAddress
}
}
klog.V(2).Infof("getServiceLoadBalancerStatus gets ingress IP %q from frontendIPConfiguration %q for service %q", to.String(lbIP), to.String(ipConfiguration.Name), serviceName)
// set additional public IPs to LoadBalancerStatus, so that kube-proxy would create their iptables rules.
lbIngress := []v1.LoadBalancerIngress{{IP: to.String(lbIP)}}
additionalIPs, err := getServiceAdditionalPublicIPs(service)
if err != nil {
return &v1.LoadBalancerStatus{Ingress: lbIngress}, &ipConfiguration, err
}
if len(additionalIPs) > 0 {
for _, pip := range additionalIPs {
lbIngress = append(lbIngress, v1.LoadBalancerIngress{
IP: pip,
})
}
}
return &v1.LoadBalancerStatus{Ingress: lbIngress}, &ipConfiguration, nil
}
}
return nil, nil, nil
}
func (az *Cloud) determinePublicIPName(clusterName string, service *v1.Service) (string, bool, error) {
var shouldPIPExisted bool
if name, found := service.Annotations[consts.ServiceAnnotationPIPName]; found && name != "" {
shouldPIPExisted = true
return name, shouldPIPExisted, nil
}
pipResourceGroup := az.getPublicIPAddressResourceGroup(service)
loadBalancerIP := service.Spec.LoadBalancerIP
// Assume that the service without loadBalancerIP set is a primary service.
// If a secondary service doesn't set the loadBalancerIP, it is not allowed to share the IP.
if len(loadBalancerIP) == 0 {
return az.getPublicIPName(clusterName, service), shouldPIPExisted, nil
}
// For the services with loadBalancerIP set, an existing public IP is required, primary
// or secondary, or a public IP not found error would be reported.
pip, err := az.findMatchedPIPByLoadBalancerIP(service, loadBalancerIP, pipResourceGroup)
if err != nil {
return "", shouldPIPExisted, err
}
if pip != nil && pip.Name != nil {
return *pip.Name, shouldPIPExisted, nil
}
return "", shouldPIPExisted, fmt.Errorf("user supplied IP Address %s was not found in resource group %s", loadBalancerIP, pipResourceGroup)
}
func (az *Cloud) findMatchedPIPByLoadBalancerIP(service *v1.Service, loadBalancerIP, pipResourceGroup string) (*network.PublicIPAddress, error) {
pips, err := az.ListPIP(service, pipResourceGroup)
if err != nil {
return nil, err
}
for _, pip := range pips {
if pip.PublicIPAddressPropertiesFormat.IPAddress != nil &&
*pip.PublicIPAddressPropertiesFormat.IPAddress == loadBalancerIP {
return &pip, nil
}
}
return nil, fmt.Errorf("findMatchedPIPByLoadBalancerIP: cannot find public IP with IP address %s in resource group %s", loadBalancerIP, pipResourceGroup)
}
func flipServiceInternalAnnotation(service *v1.Service) *v1.Service {
copyService := service.DeepCopy()
if copyService.Annotations == nil {
copyService.Annotations = map[string]string{}
}
if v, ok := copyService.Annotations[consts.ServiceAnnotationLoadBalancerInternal]; ok && v == consts.TrueAnnotationValue {
// If it is internal now, we make it external by remove the annotation
delete(copyService.Annotations, consts.ServiceAnnotationLoadBalancerInternal)
} else {
// If it is external now, we make it internal
copyService.Annotations[consts.ServiceAnnotationLoadBalancerInternal] = consts.TrueAnnotationValue
}
return copyService
}
func updateServiceLoadBalancerIP(service *v1.Service, serviceIP string) *v1.Service {
copyService := service.DeepCopy()
if len(serviceIP) > 0 && copyService != nil {
copyService.Spec.LoadBalancerIP = serviceIP
}
return copyService
}
func (az *Cloud) findServiceIPAddress(ctx context.Context, clusterName string, service *v1.Service, isInternalLb bool) (string, error) {
if len(service.Spec.LoadBalancerIP) > 0 {
return service.Spec.LoadBalancerIP, nil
}
if len(service.Status.LoadBalancer.Ingress) > 0 && len(service.Status.LoadBalancer.Ingress[0].IP) > 0 {
return service.Status.LoadBalancer.Ingress[0].IP, nil
}
_, lbStatus, existsLb, err := az.getServiceLoadBalancer(service, clusterName, nil, false, []network.LoadBalancer{})
if err != nil {
return "", err
}
if !existsLb {
klog.V(2).Infof("Expected to find an IP address for service %s but did not. Assuming it has been removed", service.Name)
return "", nil
}
if len(lbStatus.Ingress) < 1 {
klog.V(2).Infof("Expected to find an IP address for service %s but it had no ingresses. Assuming it has been removed", service.Name)
return "", nil
}
return lbStatus.Ingress[0].IP, nil
}
func (az *Cloud) ensurePublicIPExists(service *v1.Service, pipName string, domainNameLabel, clusterName string, shouldPIPExisted, foundDNSLabelAnnotation bool) (*network.PublicIPAddress, error) {
pipResourceGroup := az.getPublicIPAddressResourceGroup(service)
pip, existsPip, err := az.getPublicIPAddress(pipResourceGroup, pipName)
if err != nil {
return nil, err
}
serviceName := getServiceName(service)
var changed bool
if existsPip {
// ensure that the service tag is good for managed pips
owns, isUserAssignedPIP := serviceOwnsPublicIP(service, &pip, clusterName)
if owns && !isUserAssignedPIP {
changed, err = bindServicesToPIP(&pip, []string{serviceName}, false)
if err != nil {
return nil, err
}
}
if pip.Tags == nil {
pip.Tags = make(map[string]*string)
}
// return if pip exist and dns label is the same
if strings.EqualFold(getDomainNameLabel(&pip), domainNameLabel) {
if existingServiceName, ok := pip.Tags[consts.ServiceUsingDNSKey]; ok &&
strings.EqualFold(*existingServiceName, serviceName) {
klog.V(6).Infof("ensurePublicIPExists for service(%s): pip(%s) - "+
"the service is using the DNS label on the public IP", serviceName, pipName)
var rerr *retry.Error
if changed {
klog.V(2).Infof("ensurePublicIPExists: updating the PIP %s for the incoming service %s", pipName, serviceName)
err = az.CreateOrUpdatePIP(service, pipResourceGroup, pip)