-
Notifications
You must be signed in to change notification settings - Fork 299
/
reflector.go
325 lines (283 loc) · 11.9 KB
/
reflector.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
/*
Copyright 2019 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 readiness
import (
"fmt"
"reflect"
"sync"
"time"
"github.com/GoogleCloudPlatform/k8s-cloud-provider/pkg/cloud/meta"
v1 "k8s.io/api/core/v1"
"k8s.io/apimachinery/pkg/util/wait"
"k8s.io/client-go/kubernetes"
"k8s.io/client-go/kubernetes/scheme"
unversionedcore "k8s.io/client-go/kubernetes/typed/core/v1"
"k8s.io/client-go/tools/cache"
"k8s.io/client-go/tools/record"
"k8s.io/client-go/util/workqueue"
"k8s.io/ingress-gce/pkg/neg/metrics"
negtypes "k8s.io/ingress-gce/pkg/neg/types"
"k8s.io/ingress-gce/pkg/neg/types/shared"
"k8s.io/ingress-gce/pkg/utils/zonegetter"
"k8s.io/klog/v2"
"k8s.io/utils/clock"
)
const (
maxRetries = 15
// negReadyReason is the pod condition reason when pod becomes Healthy in NEG or pod no longer belongs to any NEG
negReadyReason = "LoadBalancerNegReady"
// negReadyTimedOutReason is the pod condition reason when timeout is reached but pod is still not healthy in NEG
negReadyTimedOutReason = "LoadBalancerNegTimeout"
// negReadyUnhealthCheckedReason is the pod condition reason when pod is in a NEG without associated health checking
negReadyUnhealthCheckedReason = "LoadBalancerNegWithoutHealthCheck"
// negNotReadyReason is the pod condition reason when pod is not healthy in NEG
negNotReadyReason = "LoadBalancerNegNotReady"
// unreadyTimeout is the timeout for health status feedback for pod readiness. If load balancer health
// check is still not showing as Healthy for long than the time out since the pod is created. Skip waiting and mark
// the pod as load balancer ready.
// This is a fail-safe in case that should be longer than any reasonable amount of time for the healthy infrastructure catch up.
unreadyTimeout = 10 * time.Minute
)
// readinessReflector implements the Reflector interface
type readinessReflector struct {
// podUpdateLock ensures that at any time there is only one
podUpdateLock sync.Mutex
client kubernetes.Interface
clock clock.Clock
// pollerLock ensures there is only poll
pollerLock sync.Mutex
poller *poller
podLister cache.Indexer
lookup NegLookup
eventRecorder record.EventRecorder
queue workqueue.RateLimitingInterface
zoneGetter *zonegetter.ZoneGetter
// If enabled, the reflector will mark pods that are from the non-default
// subnet as ready without processing.
enableMultiSubnetCluster bool
logger klog.Logger
}
func NewReadinessReflector(kubeClient, eventRecorderClient kubernetes.Interface, podLister cache.Indexer, negCloud negtypes.NetworkEndpointGroupCloud, lookup NegLookup, zoneGetter *zonegetter.ZoneGetter, enableDualStackNEG, enableMultiSubnetCluster bool, logger klog.Logger) Reflector {
broadcaster := record.NewBroadcaster()
broadcaster.StartLogging(klog.Infof)
broadcaster.StartRecordingToSink(&unversionedcore.EventSinkImpl{
Interface: eventRecorderClient.CoreV1().Events(""),
})
recorder := broadcaster.NewRecorder(scheme.Scheme, v1.EventSource{Component: "neg-readiness-reflector"})
logger = logger.WithName("ReadinessReflector")
reflector := &readinessReflector{
client: kubeClient,
podLister: podLister,
clock: clock.RealClock{},
lookup: lookup,
eventRecorder: recorder,
queue: workqueue.NewRateLimitingQueue(workqueue.DefaultControllerRateLimiter()),
zoneGetter: zoneGetter,
enableMultiSubnetCluster: enableMultiSubnetCluster,
logger: logger,
}
poller := NewPoller(podLister, lookup, reflector, negCloud, enableDualStackNEG, logger)
reflector.poller = poller
return reflector
}
func (r *readinessReflector) Run(stopCh <-chan struct{}) {
defer r.queue.ShutDown()
r.logger.V(2).Info("Starting NEG readiness reflector")
defer r.logger.V(2).Info("Shutting down NEG readiness reflector")
go wait.Until(r.worker, time.Second, stopCh)
<-stopCh
}
func (r *readinessReflector) worker() {
for r.processNextWorkItem() {
}
}
func (r *readinessReflector) processNextWorkItem() bool {
key, quit := r.queue.Get()
if quit {
return false
}
defer r.queue.Done(key)
err := r.syncPod(key.(string), nil, nil)
r.handleErr(err, key)
return true
}
// handleErr handles errors from syncPod
func (r *readinessReflector) handleErr(err error, key interface{}) {
if err == nil {
r.queue.Forget(key)
return
}
if r.queue.NumRequeues(key) < maxRetries {
r.logger.V(2).Info("Error syncing pod. Retrying.", "pod", key, "err", err)
r.queue.AddRateLimited(key)
return
}
r.logger.Info("Dropping pod out of the queue", "pod", key, "err", err)
r.queue.Forget(key)
}
// syncPod process pod and patch the NEG readiness condition if needed
// if neg and backendService is specified, it means pod is Healthy in the NEG attached to backendService.
func (r *readinessReflector) syncPod(podKey string, neg, backendService *meta.Key) (err error) {
// podUpdateLock to ensure there is no race in pod status update
r.podUpdateLock.Lock()
defer r.podUpdateLock.Unlock()
namespace, name, err := cache.SplitMetaNamespaceKey(podKey)
if err != nil {
return err
}
pod, exists, err := getPodFromStore(r.podLister, namespace, name)
if err != nil {
return err
}
if !exists {
r.logger.V(3).Info("Pod no longer exists. Skipping", "pod", podKey)
return nil
}
// This is to prevent if the pod got updated after being added to the queue
if !needToProcess(pod) {
return nil
}
r.logger.V(3).Info("Syncing pod", "pod", podKey, "neg", neg, "backendService", backendService)
expectedCondition := r.getExpectedNegCondition(pod, neg, backendService)
return r.ensurePodNegCondition(pod, expectedCondition)
}
// getExpectedCondition returns the expected NEG readiness condition for the given pod
func (r *readinessReflector) getExpectedNegCondition(pod *v1.Pod, neg, backendService *meta.Key) v1.PodCondition {
expectedCondition := v1.PodCondition{Type: shared.NegReadinessGate}
if pod == nil {
expectedCondition.Message = "Unknown status for unknown pod."
return expectedCondition
}
if neg != nil {
if backendService != nil {
expectedCondition.Status = v1.ConditionTrue
expectedCondition.Reason = negReadyReason
expectedCondition.Message = fmt.Sprintf("Pod has become Healthy in NEG %q attached to BackendService %q. Marking condition %q to True.", neg.String(), backendService.String(), shared.NegReadinessGate)
} else {
expectedCondition.Status = v1.ConditionTrue
expectedCondition.Reason = negReadyUnhealthCheckedReason
expectedCondition.Message = fmt.Sprintf("Pod is in NEG %q. NEG is not attached to any BackendService with health checking. Marking condition %q to True.", neg.String(), shared.NegReadinessGate)
}
return expectedCondition
}
negs := r.lookup.ReadinessGateEnabledNegs(pod.Namespace, pod.Labels)
// mark pod as ready if it belongs to no NEGs
if len(negs) == 0 {
expectedCondition.Status = v1.ConditionTrue
expectedCondition.Reason = negReadyReason
expectedCondition.Message = fmt.Sprintf("Pod does not belong to any NEG. Marking condition %q to True.", shared.NegReadinessGate)
return expectedCondition
}
// check if the pod has been waiting for the endpoint to show up as Healthy in NEG for too long
if r.clock.Now().After(pod.CreationTimestamp.Add(unreadyTimeout)) {
r.logger.Info(pod.Name, "now", r.clock.Now(), "pod time", pod.CreationTimestamp)
expectedCondition.Status = v1.ConditionTrue
expectedCondition.Reason = negReadyTimedOutReason
expectedCondition.Message = fmt.Sprintf("Timeout waiting for pod to become healthy in at least one of the NEG(s): %v. Marking condition %q to True.", negs, shared.NegReadinessGate)
return expectedCondition
}
if r.enableMultiSubnetCluster {
if pod.Spec.NodeName == "" {
r.logger.Error(nil, "Unable to determine the pod's node name.", "podNamespace", pod.Namespace, "podName", pod.Name)
expectedCondition.Reason = negNotReadyReason
expectedCondition.Message = "Unable to determine the pod's node name."
return expectedCondition
}
isInDefaultSubnet, err := r.zoneGetter.IsDefaultSubnetNode(pod.Spec.NodeName, r.logger)
if err != nil {
r.logger.Error(err, "Unable to determine if the pod is in the default subnet.", "podNamespace", pod.Namespace, "podName", pod.Name)
expectedCondition.Reason = negNotReadyReason
expectedCondition.Message = "Unable to determine if the pod is in the default subnet."
return expectedCondition
}
if !isInDefaultSubnet {
expectedCondition.Status = v1.ConditionTrue
expectedCondition.Reason = negReadyReason
expectedCondition.Message = fmt.Sprintf("Pod belongs to a node in non-default subnet. Marking condition %q to True.", shared.NegReadinessGate)
return expectedCondition
}
}
// do not patch condition status in this case to prevent race condition:
// 1. poller marks a pod ready
// 2. syncPod gets call and does not retrieve the updated pod spec with true neg readiness condition
// 3. syncPod patches the neg readiness condition to be false
expectedCondition.Reason = negNotReadyReason
expectedCondition.Message = fmt.Sprintf("Waiting for pod to become healthy in at least one of the NEG(s): %v", negs)
return expectedCondition
}
// SyncPod filter the pods that needed to be processed and put it into queue
func (r *readinessReflector) SyncPod(pod *v1.Pod) {
key, err := cache.DeletionHandlingMetaNamespaceKeyFunc(pod)
if err != nil {
r.logger.Error(err, "Failed to generate pod key")
metrics.PublishNegControllerErrorCountMetrics(err, true)
return
}
if !needToProcess(pod) {
r.logger.V(3).Info("Skip processing pod", "pod", key)
}
r.queue.Add(key)
}
// CommitPods registers the current network endpoints in a NEG and starts polling them if needed
func (r *readinessReflector) CommitPods(syncerKey negtypes.NegSyncerKey, negName string, zone string, endpointMap negtypes.EndpointPodMap) {
key := negMeta{
SyncerKey: syncerKey,
Name: negName,
Zone: zone,
}
r.poller.RegisterNegEndpoints(key, endpointMap)
r.poll()
}
// poll spins off go routines to poll NEGs
func (r *readinessReflector) poll() {
r.pollerLock.Lock()
defer r.pollerLock.Unlock()
for _, key := range r.poller.ScanForWork() {
go r.pollNeg(key)
}
}
// pollNeg polls a NEG
func (r *readinessReflector) pollNeg(key negMeta) {
r.logger.V(3).Info("Polling NEG", "neg", key.String())
retry, err := r.poller.Poll(key)
if err != nil {
r.logger.Error(err, "Failed to poll neg", "neg", key)
metrics.PublishNegControllerErrorCountMetrics(err, true)
}
if retry {
r.poll()
}
}
// ensurePodNegCondition ensures the pod neg condition is as expected
// TODO(freehan): also populate lastTransitionTime in the condition
func (r *readinessReflector) ensurePodNegCondition(pod *v1.Pod, expectedCondition v1.PodCondition) error {
if pod == nil {
return nil
}
// check if it is necessary to patch
condition, ok := NegReadinessConditionStatus(pod)
if ok && reflect.DeepEqual(expectedCondition, condition) {
r.logger.V(3).Info("NEG condition for pod is expected, skip patching", "pod", klog.KRef(pod.Namespace, pod.Name))
return nil
}
// calculate patch bytes, send patch and record event
oldStatus := pod.Status.DeepCopy()
SetNegReadinessConditionStatus(pod, expectedCondition)
patchBytes, err := preparePatchBytesforPodStatus(*oldStatus, pod.Status)
if err != nil {
return fmt.Errorf("failed to prepare patch bytes for pod %v: %v", pod, err)
}
r.eventRecorder.Eventf(pod, v1.EventTypeNormal, expectedCondition.Reason, expectedCondition.Message)
_, _, err = patchPodStatus(r.client, pod.Namespace, pod.Name, patchBytes)
return err
}