/
mon.go
1445 lines (1261 loc) · 52.9 KB
/
mon.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 2016 The Rook Authors. All rights reserved.
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 mon provides methods for creating clusters of Ceph mons in Kubernetes, for monitoring the
// cluster's status, for taking corrective actions if the status is non-ideal, and for reporting
// mon cluster failures.
package mon
import (
"encoding/json"
"fmt"
"reflect"
"strconv"
"strings"
"sync"
"time"
"github.com/banzaicloud/k8s-objectmatcher/patch"
"github.com/coreos/pkg/capnslog"
"github.com/pkg/errors"
cephv1 "github.com/rook/rook/pkg/apis/ceph.rook.io/v1"
"github.com/rook/rook/pkg/clusterd"
cephclient "github.com/rook/rook/pkg/daemon/ceph/client"
cephutil "github.com/rook/rook/pkg/daemon/ceph/util"
"github.com/rook/rook/pkg/operator/ceph/config"
"github.com/rook/rook/pkg/operator/ceph/config/keyring"
"github.com/rook/rook/pkg/operator/ceph/controller"
"github.com/rook/rook/pkg/operator/ceph/csi"
cephver "github.com/rook/rook/pkg/operator/ceph/version"
"github.com/rook/rook/pkg/operator/k8sutil"
apps "k8s.io/api/apps/v1"
v1 "k8s.io/api/core/v1"
kerrors "k8s.io/apimachinery/pkg/api/errors"
metav1 "k8s.io/apimachinery/pkg/apis/meta/v1"
"k8s.io/apimachinery/pkg/labels"
"k8s.io/apimachinery/pkg/util/wait"
)
const (
// EndpointConfigMapName is the name of the configmap with mon endpoints
EndpointConfigMapName = "rook-ceph-mon-endpoints"
// EndpointDataKey is the name of the key inside the mon configmap to get the endpoints
EndpointDataKey = "data"
// MaxMonIDKey is the name of the max mon id used
MaxMonIDKey = "maxMonId"
// MappingKey is the name of the mapping for the mon->node and node->port
MappingKey = "mapping"
// AppName is the name of the secret storing cluster mon.admin key, fsid and name
AppName = "rook-ceph-mon"
// OperatorCreds is the name of the secret
OperatorCreds = "rook-ceph-operator-creds"
monClusterAttr = "mon_cluster"
fsidSecretNameKey = "fsid"
monSecretNameKey = "mon-secret"
// AdminSecretName is the name of the admin secret
adminSecretNameKey = "admin-secret"
cephUsernameKey = "ceph-username"
cephUserSecretKey = "ceph-secret"
// DefaultMonCount Default mon count for a cluster
DefaultMonCount = 3
// MaxMonCount Maximum allowed mon count for a cluster
MaxMonCount = 9
// DefaultMsgr1Port is the default port Ceph mons use to communicate amongst themselves prior
// to Ceph Nautilus.
DefaultMsgr1Port int32 = 6789
// DefaultMsgr2Port is the listening port of the messenger v2 protocol introduced in Ceph
// Nautilus. In Nautilus and a few Ceph releases after, Ceph can use both v1 and v2 protocols.
DefaultMsgr2Port int32 = 3300
// minimum amount of memory in MB to run the pod
cephMonPodMinimumMemory uint64 = 1024
// default storage request size for ceph monitor pvc
// https://docs.ceph.com/docs/master/start/hardware-recommendations/#monitors-and-managers-ceph-mon-and-ceph-mgr
cephMonDefaultStorageRequest = "10Gi"
// canary pod scheduling uses retry loops when cleaning up previous canary
// pods and waiting for kubernetes scheduling to complete.
canaryRetries = 30
canaryRetryDelaySeconds = 5
DisasterProtectionFinalizerName = cephv1.CustomResourceGroup + "/disaster-protection"
)
var (
logger = capnslog.NewPackageLogger("github.com/rook/rook", "op-mon")
// hook for tests to override
waitForMonitorScheduling = realWaitForMonitorScheduling
)
// Cluster represents the Rook and environment configuration settings needed to set up Ceph mons.
type Cluster struct {
ClusterInfo *cephclient.ClusterInfo
context *clusterd.Context
spec cephv1.ClusterSpec
Namespace string
Keyring string
rookVersion string
orchestrationMutex sync.Mutex
Port int32
maxMonID int
waitForStart bool
monTimeoutList map[string]time.Time
mapping *Mapping
ownerInfo *k8sutil.OwnerInfo
csiConfigMutex *sync.Mutex
isUpgrade bool
arbiterMon string
}
// monConfig for a single monitor
type monConfig struct {
// ResourceName is the name given to the mon's Kubernetes resources in metadata
ResourceName string
// DaemonName is the name given the mon daemon ("a", "b", "c,", etc.)
DaemonName string
// PublicIP is the IP of the mon's service that the mon will receive connections on
PublicIP string
// Port is the port on which the mon will listen for connections
Port int32
// The zone used for a stretch cluster
Zone string
// DataPathMap is the mapping relationship between mon data stored on the host and mon data
// stored in containers.
DataPathMap *config.DataPathMap
}
// Mapping is mon node and port mapping
type Mapping struct {
// This isn't really node info since it could also be for zones, but we leave it as "node" for backward compatibility.
Schedule map[string]*MonScheduleInfo `json:"node"`
}
// MonScheduleInfo contains name and address of a node.
type MonScheduleInfo struct {
// Name of the node. **json names are capitalized for backwards compat**
Name string `json:"Name,omitempty"`
Hostname string `json:"Hostname,omitempty"`
Address string `json:"Address,omitempty"`
Zone string `json:"zone,omitempty"`
}
type SchedulingResult struct {
Node *v1.Node
CanaryDeployment *apps.Deployment
CanaryPVC string
}
// New creates an instance of a mon cluster
func New(context *clusterd.Context, namespace string, spec cephv1.ClusterSpec, ownerInfo *k8sutil.OwnerInfo, csiConfigMutex *sync.Mutex) *Cluster {
return &Cluster{
context: context,
spec: spec,
Namespace: namespace,
maxMonID: -1,
waitForStart: true,
monTimeoutList: map[string]time.Time{},
mapping: &Mapping{
Schedule: map[string]*MonScheduleInfo{},
},
ownerInfo: ownerInfo,
csiConfigMutex: csiConfigMutex,
}
}
// Start begins the process of running a cluster of Ceph mons.
func (c *Cluster) Start(clusterInfo *cephclient.ClusterInfo, rookVersion string, cephVersion cephver.CephVersion, spec cephv1.ClusterSpec) (*cephclient.ClusterInfo, error) {
// Only one goroutine can orchestrate the mons at a time
c.acquireOrchestrationLock()
defer c.releaseOrchestrationLock()
clusterInfo.OwnerInfo = c.ownerInfo
c.ClusterInfo = clusterInfo
if c.ClusterInfo.Context == nil {
panic("nil context")
}
c.rookVersion = rookVersion
c.spec = spec
// fail if we were instructed to deploy more than one mon on the same machine with host networking
if c.spec.Network.IsHost() && c.spec.Mon.AllowMultiplePerNode && c.spec.Mon.Count > 1 {
return nil, errors.Errorf("refusing to deploy %d monitors on the same host with host networking and allowMultiplePerNode is %t. only one monitor per node is allowed", c.spec.Mon.Count, c.spec.Mon.AllowMultiplePerNode)
}
// Validate pod's memory if specified
err := controller.CheckPodMemory(cephv1.ResourcesKeyMon, cephv1.GetMonResources(c.spec.Resources), cephMonPodMinimumMemory)
if err != nil {
return nil, errors.Wrap(err, "failed to check pod memory")
}
logger.Infof("start running mons")
logger.Debugf("establishing ceph cluster info")
if err := c.initClusterInfo(cephVersion, c.ClusterInfo.NamespacedName().Name); err != nil {
return nil, errors.Wrap(err, "failed to initialize ceph cluster info")
}
logger.Infof("targeting the mon count %d", c.spec.Mon.Count)
// create the mons for a new cluster or ensure mons are running in an existing cluster
return c.ClusterInfo, c.startMons(c.spec.Mon.Count)
}
func (c *Cluster) startMons(targetCount int) error {
// init the mon config
existingCount, mons, err := c.initMonConfig(targetCount)
if err != nil {
return errors.Wrap(err, "failed to init mon config")
}
// Assign the mons to nodes
if err := c.assignMons(mons); err != nil {
return errors.Wrap(err, "failed to assign pods to mons")
}
// The centralized mon config database can only be used if there is at least one mon
// operational. If we are starting mons, and one is already up, then there is a cluster already
// created, and we can immediately set values in the config database. The goal is to set configs
// only once and do it as early as possible in the mon orchestration.
setConfigsNeedsRetry := false
if existingCount > 0 {
err := config.SetOrRemoveDefaultConfigs(c.context, c.ClusterInfo, c.spec)
if err != nil {
// If we fail here, it could be because the mons are not healthy, and this might be
// fixed by updating the mon deployments. Instead of returning error here, log a
// warning, and retry setting this later.
setConfigsNeedsRetry = true
logger.Warningf("failed to set Rook and/or user-defined Ceph config options before starting mons; will retry after starting mons. %v", err)
}
}
if existingCount < len(mons) {
// Start the new mons one at a time
for i := existingCount; i < targetCount; i++ {
if c.ClusterInfo.Context.Err() != nil {
return c.ClusterInfo.Context.Err()
}
if err := c.ensureMonsRunning(mons, i, targetCount, true); err != nil {
return err
}
// If this is the first mon being created, we have to wait until it is created to set
// values in the config database. Do this only when the existing count is zero so that
// this is only done once when the cluster is created.
if existingCount == 0 {
err := config.SetOrRemoveDefaultConfigs(c.context, c.ClusterInfo, c.spec)
if err != nil {
return errors.Wrap(err, "failed to set Rook and/or user-defined Ceph config options after creating the first mon")
}
} else if setConfigsNeedsRetry && i == existingCount {
// Or if we need to retry, only do this when we are on the first iteration of the
// loop. This could be in the same if statement as above, but separate it to get a
// different error message.
err := config.SetOrRemoveDefaultConfigs(c.context, c.ClusterInfo, c.spec)
if err != nil {
return errors.Wrap(err, "failed to set Rook and/or user-defined Ceph config options after updating the existing mons")
}
}
}
} else {
// Ensure all the expected mon deployments exist, but don't require full quorum to continue
lastMonIndex := len(mons) - 1
if err := c.ensureMonsRunning(mons, lastMonIndex, targetCount, false); err != nil {
return err
}
if setConfigsNeedsRetry {
err := config.SetOrRemoveDefaultConfigs(c.context, c.ClusterInfo, c.spec)
if err != nil {
return errors.Wrap(err, "failed to set Rook and/or user-defined Ceph config options after forcefully updating the existing mons")
}
}
}
if c.spec.IsStretchCluster() {
if err := c.configureStretchCluster(mons); err != nil {
return errors.Wrap(err, "failed to configure stretch mons")
}
}
logger.Debugf("mon endpoints used are: %s", FlattenMonEndpoints(c.ClusterInfo.Monitors))
// reconcile mon PDB
if err := c.reconcileMonPDB(); err != nil {
return errors.Wrap(err, "failed to reconcile mon PDB")
}
// Check if there are orphaned mon resources that should be cleaned up at the end of a reconcile.
// There may be orphaned resources if a mon failover was aborted.
c.removeOrphanMonResources()
return nil
}
func (c *Cluster) configureStretchCluster(mons []*monConfig) error {
// Enable the mon connectivity strategy
if err := cephclient.EnableStretchElectionStrategy(c.context, c.ClusterInfo); err != nil {
return errors.Wrap(err, "failed to enable stretch cluster")
}
// Create the default crush rule for stretch clusters, that by default will also apply to all pools
if err := cephclient.CreateDefaultStretchCrushRule(c.context, c.ClusterInfo, &c.spec, c.stretchFailureDomainName()); err != nil {
return errors.Wrap(err, "failed to create default stretch rule")
}
return nil
}
func (c *Cluster) getArbiterZone() string {
for _, zone := range c.spec.Mon.StretchCluster.Zones {
if zone.Arbiter {
return zone.Name
}
}
return ""
}
func (c *Cluster) isArbiterZone(zone string) bool {
if !c.spec.IsStretchCluster() {
return false
}
return c.getArbiterZone() == zone
}
func (c *Cluster) ConfigureArbiter() error {
if c.arbiterMon == "" {
return errors.New("arbiter not specified for the stretch cluster")
}
monDump, err := cephclient.GetMonDump(c.context, c.ClusterInfo)
if err != nil {
logger.Warningf("attempting to enable arbiter after failed to detect if already enabled. %v", err)
} else if monDump.StretchMode {
// only support arbiter failover if at least v16.2.7
if !c.ClusterInfo.CephVersion.IsAtLeast(arbiterFailoverSupportedCephVersion) {
logger.Info("stretch mode is already enabled")
return nil
}
if monDump.TiebreakerMon == c.arbiterMon {
logger.Infof("stretch mode is already enabled with tiebreaker %q", c.arbiterMon)
return nil
}
// Set the new mon tiebreaker
logger.Infof("updating tiebreaker mon from %q to %q", monDump.TiebreakerMon, c.arbiterMon)
if err := cephclient.SetNewTiebreaker(c.context, c.ClusterInfo, c.arbiterMon); err != nil {
return errors.Wrap(err, "failed to set new mon tiebreaker")
}
return nil
}
// Wait for the CRUSH map to have at least two zones
// The timeout is relatively short since the operator will requeue the reconcile
// and try again at a higher level if not yet found
failureDomain := c.stretchFailureDomainName()
logger.Infof("enabling stretch mode... waiting for two failure domains of type %q to be found in the CRUSH map after OSD initialization", failureDomain)
pollInterval := 5 * time.Second
totalWaitTime := 2 * time.Minute
err = wait.Poll(pollInterval, totalWaitTime, func() (bool, error) {
return c.readyToConfigureArbiter(true)
})
if err != nil {
return errors.Wrapf(err, "failed to find two failure domains %q in the CRUSH map", failureDomain)
}
// Set the mon tiebreaker
if err := cephclient.SetMonStretchTiebreaker(c.context, c.ClusterInfo, c.arbiterMon, failureDomain); err != nil {
return errors.Wrap(err, "failed to set mon tiebreaker")
}
return nil
}
func (c *Cluster) readyToConfigureArbiter(checkOSDPods bool) (bool, error) {
failureDomain := c.stretchFailureDomainName()
if checkOSDPods {
// Wait for the OSD pods to be running
// can't use osd.AppName due to a circular dependency
allRunning, err := k8sutil.PodsWithLabelAreAllRunning(c.ClusterInfo.Context, c.context.Clientset, c.Namespace, fmt.Sprintf("%s=rook-ceph-osd", k8sutil.AppAttr))
if err != nil {
return false, errors.Wrap(err, "failed to check whether all osds are running before enabling the arbiter")
}
if !allRunning {
logger.Infof("waiting for all OSD pods to be in running state")
return false, nil
}
}
crushMap, err := cephclient.GetCrushMap(c.context, c.ClusterInfo)
if err != nil {
return false, errors.Wrap(err, "failed to get crush map")
}
// Check if the crush rule already exists
zoneCount := 0
zoneWeight := -1
for _, bucket := range crushMap.Buckets {
if bucket.TypeName == failureDomain {
// skip zones specific to device classes
if strings.Index(bucket.Name, "~") > 0 {
logger.Debugf("skipping device class bucket %q", bucket.Name)
continue
}
logger.Infof("found %s %q in CRUSH map with weight %d", failureDomain, bucket.Name, bucket.Weight)
zoneCount++
// check that the weights of the failure domains are all the same
if zoneWeight == -1 {
// found the first matching bucket
zoneWeight = bucket.Weight
} else if zoneWeight != bucket.Weight {
logger.Infof("found failure domains that have different weights")
return false, nil
}
}
}
if zoneCount < 2 {
// keep waiting to see if more zones will be created
return false, nil
}
if zoneCount > 2 {
return false, fmt.Errorf("cannot configure stretch cluster with more than 2 failure domains, and found %d of type %q", zoneCount, failureDomain)
}
logger.Infof("found two expected failure domains %q for the stretch cluster", failureDomain)
return true, nil
}
// ensureMonsRunning is called in two scenarios:
// 1. To create a new mon and wait for it to join quorum (requireAllInQuorum = true). This method will be called multiple times
// to add a mon until we have reached the desired number of mons.
// 2. To check that the majority of existing mons are in quorum. It is ok if not all mons are in quorum. (requireAllInQuorum = false)
// This is needed when the operator is restarted and all mons may not be up or in quorum.
func (c *Cluster) ensureMonsRunning(mons []*monConfig, i, targetCount int, requireAllInQuorum bool) error {
if requireAllInQuorum {
logger.Infof("creating mon %s", mons[i].DaemonName)
} else {
logger.Info("checking for basic quorum with existing mons")
}
// Calculate how many mons we expected to exist after this method is completed.
// If we are adding a new mon, we expect one more than currently exist.
// If we haven't created all the desired mons already, we will be adding a new one with this iteration
expectedMonCount := len(c.ClusterInfo.Monitors)
if expectedMonCount < targetCount {
expectedMonCount++
}
// Init the mon IPs
if err := c.initMonIPs(mons[0:expectedMonCount]); err != nil {
return errors.Wrap(err, "failed to init mon services")
}
// save the mon config after we have "initiated the IPs"
if err := c.saveMonConfig(); err != nil {
return errors.Wrap(err, "failed to save mons")
}
// Start the deployment
if err := c.startDeployments(mons[0:expectedMonCount], requireAllInQuorum); err != nil {
return errors.Wrap(err, "failed to start mon pods")
}
return nil
}
// initClusterInfo retrieves the ceph cluster info if it already exists.
// If a new cluster, create new keys.
func (c *Cluster) initClusterInfo(cephVersion cephver.CephVersion, clusterName string) error {
var err error
context := c.ClusterInfo.Context
// get the cluster info from secret
c.ClusterInfo, c.maxMonID, c.mapping, err = CreateOrLoadClusterInfo(c.context, context, c.Namespace, c.ownerInfo)
if err != nil {
return errors.Wrap(err, "failed to get cluster info")
}
c.ClusterInfo.CephVersion = cephVersion
c.ClusterInfo.OwnerInfo = c.ownerInfo
c.ClusterInfo.Context = context
c.ClusterInfo.SetName(clusterName)
// save cluster monitor config
if err = c.saveMonConfig(); err != nil {
return errors.Wrap(err, "failed to save mons")
}
k := keyring.GetSecretStore(c.context, c.ClusterInfo, c.ownerInfo)
// store the keyring which all mons share
if err := k.CreateOrUpdate(keyringStoreName, c.genMonSharedKeyring()); err != nil {
return errors.Wrap(err, "failed to save mon keyring secret")
}
// also store the admin keyring for other daemons that might need it during init
if err := k.Admin().CreateOrUpdate(c.ClusterInfo); err != nil {
return errors.Wrap(err, "failed to save admin keyring secret")
}
return nil
}
func (c *Cluster) initMonConfig(size int) (int, []*monConfig, error) {
// initialize the mon pod info for mons that have been previously created
mons := c.clusterInfoToMonConfig("")
// initialize mon info if we don't have enough mons (at first startup)
existingCount := len(c.ClusterInfo.Monitors)
for i := len(c.ClusterInfo.Monitors); i < size; i++ {
c.maxMonID++
zone, err := c.findAvailableZoneIfStretched(mons)
if err != nil {
return existingCount, mons, errors.Wrap(err, "stretch zone not available")
}
mons = append(mons, c.newMonConfig(c.maxMonID, zone))
}
return existingCount, mons, nil
}
func (c *Cluster) clusterInfoToMonConfig(excludedMon string) []*monConfig {
mons := []*monConfig{}
for _, monitor := range c.ClusterInfo.Monitors {
if monitor.Name == excludedMon {
// Skip a mon if it is being failed over
continue
}
var zone string
schedule := c.mapping.Schedule[monitor.Name]
if schedule != nil {
zone = schedule.Zone
}
mons = append(mons, &monConfig{
ResourceName: resourceName(monitor.Name),
DaemonName: monitor.Name,
Port: cephutil.GetPortFromEndpoint(monitor.Endpoint),
PublicIP: cephutil.GetIPFromEndpoint(monitor.Endpoint),
Zone: zone,
DataPathMap: config.NewStatefulDaemonDataPathMap(c.spec.DataDirHostPath, dataDirRelativeHostPath(monitor.Name), config.MonType, monitor.Name, c.Namespace),
})
}
return mons
}
func (c *Cluster) newMonConfig(monID int, zone string) *monConfig {
daemonName := k8sutil.IndexToName(monID)
return &monConfig{
ResourceName: resourceName(daemonName),
DaemonName: daemonName,
Port: DefaultMsgr1Port,
Zone: zone,
DataPathMap: config.NewStatefulDaemonDataPathMap(
c.spec.DataDirHostPath, dataDirRelativeHostPath(daemonName), config.MonType, daemonName, c.Namespace),
}
}
func (c *Cluster) findAvailableZoneIfStretched(mons []*monConfig) (string, error) {
if !c.spec.IsStretchCluster() {
return "", nil
}
// Build the count of current mons per zone
zoneCount := map[string]int{}
for _, m := range mons {
if m.Zone == "" {
return "", errors.Errorf("zone not found on mon %q", m.DaemonName)
}
zoneCount[m.Zone]++
}
// Find a zone in the stretch cluster that still needs an assignment
for _, zone := range c.spec.Mon.StretchCluster.Zones {
count, ok := zoneCount[zone.Name]
if !ok {
// The zone isn't currently assigned to any mon, so return it
return zone.Name, nil
}
if c.spec.Mon.Count == 5 && count == 1 && !zone.Arbiter {
// The zone only has 1 mon assigned, but needs 2 mons since it is not the arbiter
return zone.Name, nil
}
}
return "", errors.New("A zone is not available to assign a new mon")
}
// resourceName ensures the mon name has the rook-ceph-mon prefix
func resourceName(name string) string {
if strings.HasPrefix(name, AppName) {
return name
}
return fmt.Sprintf("%s-%s", AppName, name)
}
// scheduleMonitor selects a node for a monitor deployment.
// see startMon() and design/ceph/ceph-mon-pv.md for additional details.
func scheduleMonitor(c *Cluster, mon *monConfig) (*apps.Deployment, error) {
// build the canary deployment.
d, err := c.makeDeployment(mon, true)
if err != nil {
return nil, err
}
d.Name += "-canary"
d.Spec.Template.ObjectMeta.Name += "-canary"
// the canary and real monitor deployments will mount the same storage. to
// avoid issues with the real deployment, the canary should be careful not
// to modify the storage by instead running an innocuous command.
d.Spec.Template.Spec.InitContainers = []v1.Container{}
d.Spec.Template.Spec.Containers[0].Image = c.rookVersion
d.Spec.Template.Spec.Containers[0].Command = []string{"sleep"} // sleep responds to signals so we don't need to wrap it
d.Spec.Template.Spec.Containers[0].Args = []string{"3600"}
// remove the liveness probe on the canary pod
d.Spec.Template.Spec.Containers[0].LivenessProbe = nil
// setup affinity settings for pod scheduling
p := c.getMonPlacement(mon.Zone)
p.ApplyToPodSpec(&d.Spec.Template.Spec)
k8sutil.SetNodeAntiAffinityForPod(&d.Spec.Template.Spec, requiredDuringScheduling(&c.spec), v1.LabelHostname,
map[string]string{k8sutil.AppAttr: AppName}, nil)
// setup storage on the canary since scheduling will be affected when
// monitors are configured to use persistent volumes. the pvcName is set to
// the non-empty name of the PVC only when the PVC is created as a result of
// this call to the scheduler.
if c.monVolumeClaimTemplate(mon) == nil {
d.Spec.Template.Spec.Volumes = append(d.Spec.Template.Spec.Volumes,
controller.DaemonVolumesDataHostPath(mon.DataPathMap)...)
} else {
// the pvc that is created here won't be deleted: it will be reattached
// to the real monitor deployment.
pvc, err := c.makeDeploymentPVC(mon, true)
if err != nil {
return nil, errors.Wrapf(err, "failed to make monitor %s pvc", d.Name)
}
_, err = c.context.Clientset.CoreV1().PersistentVolumeClaims(c.Namespace).Create(c.ClusterInfo.Context, pvc, metav1.CreateOptions{})
if err == nil {
logger.Infof("created canary monitor %s pvc %s", d.Name, pvc.Name)
} else {
if kerrors.IsAlreadyExists(err) {
logger.Debugf("creating mon %s pvc %s: already exists.", d.Name, pvc.Name)
} else {
return nil, errors.Wrapf(err, "failed to create mon %s pvc %s", d.Name, pvc.Name)
}
}
d.Spec.Template.Spec.Volumes = append(d.Spec.Template.Spec.Volumes,
controller.DaemonVolumesDataPVC(mon.ResourceName))
controller.AddVolumeMountSubPath(&d.Spec.Template.Spec, "ceph-daemon-data")
}
// spin up the canary deployment. if it exists, delete it first, since if it
// already exists it may have been scheduled with a different crd config.
createdDeployment := false
for i := 0; i < canaryRetries; i++ {
if c.ClusterInfo.Context.Err() != nil {
return nil, c.ClusterInfo.Context.Err()
}
_, err := c.context.Clientset.AppsV1().Deployments(c.Namespace).Create(c.ClusterInfo.Context, d, metav1.CreateOptions{})
if err == nil {
createdDeployment = true
logger.Infof("created canary deployment %s", d.Name)
break
} else if kerrors.IsAlreadyExists(err) {
if err := k8sutil.DeleteDeployment(c.ClusterInfo.Context, c.context.Clientset, c.Namespace, d.Name); err != nil {
return nil, errors.Wrapf(err, "failed to delete canary deployment %s", d.Name)
}
logger.Infof("deleted existing canary deployment %s", d.Name)
time.Sleep(time.Second * canaryRetryDelaySeconds)
} else {
return nil, errors.Wrapf(err, "failed to create canary monitor deployment %s", d.Name)
}
}
// failed after retrying
if !createdDeployment {
return nil, errors.Errorf("failed to create canary deployment %s", d.Name)
}
// caller should arrange for the deployment to be removed
return d, nil
}
// GetMonPlacement returns the placement for the MON service
func (c *Cluster) getMonPlacement(zone string) cephv1.Placement {
// If the mon is the arbiter in a stretch cluster and its placement is specified, return it
// without merging with the "all" placement so it can be handled separately from all other daemons
if c.isArbiterZone(zone) {
p := cephv1.GetArbiterPlacement(c.spec.Placement)
noPlacement := cephv1.Placement{}
if !reflect.DeepEqual(p, noPlacement) {
// If the arbiter placement was specified, go ahead and use it.
return p
}
}
// If not the arbiter, or the arbiter placement is not specified, fall back to the same placement used for other mons
return cephv1.GetMonPlacement(c.spec.Placement)
}
func realWaitForMonitorScheduling(c *Cluster, d *apps.Deployment) (SchedulingResult, error) {
// target node decision, and deployment/pvc to cleanup
result := SchedulingResult{}
// wait for the scheduler to make a placement decision
for i := 0; i < canaryRetries; i++ {
if c.ClusterInfo.Context.Err() != nil {
return result, c.ClusterInfo.Context.Err()
}
if i > 0 {
time.Sleep(time.Second * canaryRetryDelaySeconds)
}
listOptions := metav1.ListOptions{
LabelSelector: labels.Set(d.Spec.Selector.MatchLabels).String(),
}
pods, err := c.context.Clientset.CoreV1().Pods(c.Namespace).List(c.ClusterInfo.Context, listOptions)
if err != nil {
return result, errors.Wrapf(err, "failed to list canary pods %s", d.Name)
}
if len(pods.Items) == 0 {
logger.Infof("waiting for canary pod creation %s", d.Name)
continue
}
pod := pods.Items[0]
if pod.Spec.NodeName == "" {
logger.Debugf("monitor %s canary pod %s not yet scheduled", d.Name, pod.Name)
continue
}
node, err := c.context.Clientset.CoreV1().Nodes().Get(c.ClusterInfo.Context, pod.Spec.NodeName, metav1.GetOptions{})
if err != nil {
return result, errors.Wrapf(err, "failed to get node %s", pod.Spec.NodeName)
}
result.Node = node
logger.Infof("canary monitor deployment %s scheduled to %s", d.Name, node.Name)
return result, nil
}
return result, errors.New("failed to schedule canary pod(s)")
}
func (c *Cluster) initMonIPs(mons []*monConfig) error {
for _, m := range mons {
if c.ClusterInfo.Context.Err() != nil {
return c.ClusterInfo.Context.Err()
}
if c.spec.Network.IsHost() {
logger.Infof("setting mon endpoints for hostnetwork mode")
node, ok := c.mapping.Schedule[m.DaemonName]
if !ok || node == nil {
return errors.Errorf("failed to found node for mon %q in assignment map", m.DaemonName)
}
m.PublicIP = node.Address
} else {
serviceIP, err := c.createService(m)
if err != nil {
return errors.Wrap(err, "failed to create mon service")
}
m.PublicIP = serviceIP
}
c.ClusterInfo.Monitors[m.DaemonName] = cephclient.NewMonInfo(m.DaemonName, m.PublicIP, m.Port)
}
return nil
}
// Delete mon canary deployments (and associated PVCs) using deployment labels
// to select this kind of temporary deployments
func (c *Cluster) removeCanaryDeployments() {
canaryDeployments, err := k8sutil.GetDeployments(c.ClusterInfo.Context, c.context.Clientset, c.Namespace, "app=rook-ceph-mon,mon_canary=true")
if err != nil {
logger.Warningf("failed to get the list of monitor canary deployments. %v", err)
return
}
// Delete the canary mons, but don't wait for them to exit
for _, canary := range canaryDeployments.Items {
logger.Infof("cleaning up canary monitor deployment %q", canary.Name)
var gracePeriod int64
propagation := metav1.DeletePropagationForeground
options := &metav1.DeleteOptions{GracePeriodSeconds: &gracePeriod, PropagationPolicy: &propagation}
if err := c.context.Clientset.AppsV1().Deployments(c.Namespace).Delete(c.ClusterInfo.Context, canary.Name, *options); err != nil {
logger.Warningf("failed to delete canary monitor deployment %q. %v", canary.Name, err)
}
}
}
func (c *Cluster) assignMons(mons []*monConfig) error {
// when monitors are scheduling below by invoking scheduleMonitor() a canary
// deployment and optional canary PVC are created. In order for the
// anti-affinity rules to be effective, we leave the canary pods in place
// until all of the canaries have been scheduled. Only after the
// monitor/node assignment process is complete are the canary deployments
// and pvcs removed here.
defer c.removeCanaryDeployments()
var monSchedulingWait sync.WaitGroup
var resultLock sync.Mutex
failedMonSchedule := false
// ensure that all monitors have either (1) a node assignment that will be
// enforced using a node selector, or (2) configuration permits k8s to handle
// scheduling for the monitor.
for _, mon := range mons {
if c.ClusterInfo.Context.Err() != nil {
return c.ClusterInfo.Context.Err()
}
// scheduling for this monitor has already been completed
if _, ok := c.mapping.Schedule[mon.DaemonName]; ok {
logger.Debugf("mon %s already scheduled", mon.DaemonName)
continue
}
// determine a placement for the monitor. note that this scheduling is
// performed even when a node selector is not required. this may be
// non-optimal, but it is convenient to catch some failures early,
// before a decision is stored in the node mapping.
deployment, err := scheduleMonitor(c, mon)
if err != nil {
return errors.Wrap(err, "failed to schedule monitor")
}
// start waiting for the deployment
monSchedulingWait.Add(1)
go func(deployment *apps.Deployment, mon *monConfig) {
// signal that the mon is done scheduling
defer monSchedulingWait.Done()
result, err := waitForMonitorScheduling(c, deployment)
if err != nil {
logger.Errorf("failed to schedule mon %q. %v", mon.DaemonName, err)
failedMonSchedule = true
return
}
nodeChoice := result.Node
if nodeChoice == nil {
logger.Errorf("failed to schedule monitor %q", mon.DaemonName)
failedMonSchedule = true
return
}
// store nil in the node mapping to indicate that an explicit node
// placement is not being made. otherwise, the node choice will map
// directly to a node selector on the monitor pod.
var schedule *MonScheduleInfo
if c.spec.Network.IsHost() || c.monVolumeClaimTemplate(mon) == nil {
logger.Infof("mon %s assigned to node %s", mon.DaemonName, nodeChoice.Name)
schedule, err = getNodeInfoFromNode(*nodeChoice)
if err != nil {
logger.Errorf("failed to get node info for node %q. %v", nodeChoice.Name, err)
failedMonSchedule = true
return
}
} else {
logger.Infof("mon %q placement using native scheduler", mon.DaemonName)
}
if c.spec.IsStretchCluster() {
if schedule == nil {
schedule = &MonScheduleInfo{}
}
logger.Infof("mon %q is assigned to zone %q", mon.DaemonName, mon.Zone)
schedule.Zone = mon.Zone
}
// protect against multiple goroutines updating the status at the same time
resultLock.Lock()
c.mapping.Schedule[mon.DaemonName] = schedule
resultLock.Unlock()
}(deployment, mon)
}
monSchedulingWait.Wait()
if failedMonSchedule {
return errors.New("failed to schedule mons")
}
logger.Debug("mons have been scheduled")
return nil
}
func (c *Cluster) monVolumeClaimTemplate(mon *monConfig) *v1.PersistentVolumeClaim {
if !c.spec.IsStretchCluster() {
return c.spec.Mon.VolumeClaimTemplate
}
// If a stretch cluster, a zone can override the template from the default.
for _, zone := range c.spec.Mon.StretchCluster.Zones {
if zone.Name == mon.Zone {
if zone.VolumeClaimTemplate != nil {
// Found an override for the volume claim template in the zone
return zone.VolumeClaimTemplate
}
break
}
}
// Return the default template since one wasn't found in the zone
return c.spec.Mon.VolumeClaimTemplate
}
func (c *Cluster) startDeployments(mons []*monConfig, requireAllInQuorum bool) error {
if len(mons) == 0 {
return errors.New("cannot start 0 mons")
}
// If all the mon deployments don't exist, allow the mon deployments to all be started without checking for quorum.
// This will be the case where:
// 1) New clusters where we are starting one deployment at a time. We only need to check for quorum once when we add a new mon.
// 2) Clusters being restored where no mon deployments are running. We need to start all the deployments before checking quorum.
onlyCheckQuorumOnce := false
deployments, err := c.context.Clientset.AppsV1().Deployments(c.Namespace).List(c.ClusterInfo.Context, metav1.ListOptions{LabelSelector: fmt.Sprintf("app=%s", AppName)})
if err != nil {
if kerrors.IsNotFound(err) {
logger.Infof("0 of %d expected mon deployments exist. creating new deployment(s).", len(mons))
onlyCheckQuorumOnce = true
} else {
logger.Warningf("failed to list mon deployments. attempting to continue. %v", err)
}
}
readyReplicas := 0
// Ensuring the mon deployments should be ready
for _, deploy := range deployments.Items {
if deploy.Status.AvailableReplicas > 0 {
readyReplicas++
}
}
if len(deployments.Items) < len(mons) {
logger.Infof("%d of %d expected mon deployments exist. creating new deployment(s).", len(deployments.Items), len(mons))
onlyCheckQuorumOnce = true
} else if readyReplicas == 0 {
logger.Infof("%d of %d expected mons are ready. creating or updating deployments without checking quorum in attempt to achieve a healthy mon cluster", readyReplicas, len(mons))
onlyCheckQuorumOnce = true
}
// Ensure each of the mons have been created. If already created, it will be a no-op.
for i := 0; i < len(mons); i++ {
schedule := c.mapping.Schedule[mons[i].DaemonName]
err := c.startMon(mons[i], schedule)
if err != nil {
if c.isUpgrade {
// if we're upgrading, we don't want to risk the health of the cluster by continuing to upgrade
// and potentially cause more mons to fail. Therefore, we abort if the mon failed to start after upgrade.
return errors.Wrapf(err, "failed to upgrade mon %q.", mons[i].DaemonName)
}
// We will attempt to start all mons, then check for quorum as needed after this. During an operator restart
// we need to do everything possible to verify the basic health of a cluster, complete the first orchestration,
// and start watching for all the CRs. If mons still have quorum we can continue with the orchestration even
// if they aren't all up.
logger.Errorf("attempting to continue after failing to start mon %q. %v", mons[i].DaemonName, err)
}
// For the initial deployment (first creation) it's expected to not have all the monitors in quorum
// However, in an event of an update, it's crucial to proceed monitors by monitors
// At the end of the method we perform one last check where all the monitors must be in quorum
if !onlyCheckQuorumOnce || (onlyCheckQuorumOnce && i == len(mons)-1) {
requireAllInQuorum := false
err = c.waitForMonsToJoin(mons, requireAllInQuorum)
if err != nil {
return errors.Wrapf(err, "failed to check mon quorum %s", mons[i].DaemonName)
}
}
}
logger.Infof("mons created: %d", len(mons))
// Final verification that **all** mons are in quorum
// Do not proceed if one monitor is still syncing
// Only do this when monitors versions are different so we don't block the orchestration if a mon is down.
versions, err := cephclient.GetAllCephDaemonVersions(c.context, c.ClusterInfo)
if err != nil {
logger.Warningf("failed to get ceph daemons versions; this likely means there is no cluster yet. %v", err)
} else {
if len(versions.Mon) != 1 {
requireAllInQuorum = true
}
}