UPSTREAM: <carry>: skip posting failures to aggregated APIs to avoid getting false positives until the server becomes ready

the availability checks depend on fully initialized SDN
OpenShift carries a few reachability checks that affect /readyz protocol
we skip posting failures to avoid getting false positives until the server becomes ready

UPSTREAM: <carry>: skip posting failures to aggregated APIs to avoid getting false positives until the server becomes ready

marks availability of the server before checking the aggregate APIs
as it can change as we are running the checks.
in that case, skip posting failures to avoid false positives.

note on the next rebase please squash with the previous commit

UPSTREAM: <carry>: expose HasBeenReady lifecycle signal

OpenShift-Rebase-Source: 8558e88

Author: p0lyn0mial

staging/src/k8s.io/apiserver/pkg/server/config.go

@@ -701,6 +701,11 @@ func (c *Config) ShutdownInitiatedNotify() <-chan struct{} {
 	return c.lifecycleSignals.ShutdownInitiated.Signaled()
 }
 
+// HasBeenReadySignal exposes a server's lifecycle signal which is signaled when the readyz endpoint succeeds for the first time.
+func (c *Config) HasBeenReadySignal() <-chan struct{} {
+	return c.lifecycleSignals.HasBeenReady.Signaled()
+}
+
 // Complete fills in any fields not set that are required to have valid data and can be derived
 // from other fields. If you're going to `ApplyOptions`, do that first. It's mutating the receiver.
 func (c *Config) Complete(informers informers.SharedInformerFactory) CompletedConfig {

staging/src/k8s.io/kube-aggregator/pkg/apiserver/apiserver.go

@@ -348,6 +348,7 @@ func (c completedConfig) NewWithDelegate(delegationTarget genericapiserver.Deleg
 			(func() ([]byte, []byte))(s.proxyCurrentCertKeyContent),
 			s.serviceResolver,
 			metrics,
+			c.GenericConfig.HasBeenReadySignal(),
 		)
 		if err != nil {
 			return nil, err

staging/src/k8s.io/kube-aggregator/pkg/controllers/status/remote/remote_available_controller.go

@@ -87,6 +87,9 @@ type AvailableConditionController struct {
 
 	// metrics registered into legacy registry
 	metrics *availabilitymetrics.Metrics
+
+	// hasBeenReady is signaled when the readyz endpoint succeeds for the first time.
+	hasBeenReady <-chan struct{}
 }
 
 // New returns a new remote APIService AvailableConditionController.
@@ -99,6 +102,7 @@ func New(
 	proxyCurrentCertKeyContent certKeyFunc,
 	serviceResolver ServiceResolver,
 	metrics *availabilitymetrics.Metrics,
+	hasBeenReady <-chan struct{},
 ) (*AvailableConditionController, error) {
 
 	endpointSliceGetter, err := proxy.NewEndpointSliceIndexerGetter(endpointSliceInformer)
@@ -122,6 +126,7 @@ func New(
 		proxyTransportDial:         proxyTransportDial,
 		proxyCurrentCertKeyContent: proxyCurrentCertKeyContent,
 		metrics:                    metrics,
+		hasBeenReady:               hasBeenReady,
 	}
 
 	// resync on this one because it is low cardinality and rechecking the actual discovery
@@ -171,6 +176,18 @@ func (c *AvailableConditionController) sync(key string) error {
 		return nil
 	}
 
+	// the availability checks depend on fully initialized SDN
+	// OpenShift carries a few reachability checks that affect /readyz protocol
+	// record availability of the server so that we can
+	// skip posting failures to avoid getting false positives until the server becomes ready
+	hasBeenReady := false
+	select {
+	case <-c.hasBeenReady:
+		hasBeenReady = true
+	default:
+		// continue, we will skip posting only potential failures
+	}
+
 	apiService := originalAPIService.DeepCopy()
 
 	// if a particular transport was specified, use that otherwise build one
@@ -361,6 +378,11 @@ func (c *AvailableConditionController) sync(key string) error {
 		}
 
 		if lastError != nil {
+			if !hasBeenReady {
+				// returning an error will requeue the item in an exponential fashion
+				return fmt.Errorf("the server hasn't been ready yet, skipping updating availability of the aggreaged API until the server becomes ready to avoid false positives, lastError = %v", lastError)
+			}
+
 			availableCondition.Status = apiregistrationv1.ConditionFalse
 			availableCondition.Reason = "FailedDiscoveryCheck"
 			availableCondition.Message = lastError.Error()

staging/src/k8s.io/kube-aggregator/pkg/controllers/status/remote/remote_available_controller_test.go

@@ -148,6 +148,9 @@ func setupAPIServices(t T, apiServices []runtime.Object) (*AvailableConditionCon
 		}
 	}
 
+	alwaysReadyChan := make(chan struct{})
+	close(alwaysReadyChan)
+
 	c := AvailableConditionController{
 		apiServiceClient:    fakeClient.ApiregistrationV1(),
 		apiServiceLister:    listers.NewAPIServiceLister(apiServiceIndexer),
@@ -161,7 +164,8 @@ func setupAPIServices(t T, apiServices []runtime.Object) (*AvailableConditionCon
 			workqueue.NewTypedItemExponentialFailureRateLimiter[string](5*time.Millisecond, 30*time.Second),
 			workqueue.TypedRateLimitingQueueConfig[string]{Name: "AvailableConditionController"},
 		),
-		metrics: availabilitymetrics.New(),
+		metrics:      availabilitymetrics.New(),
+		hasBeenReady: alwaysReadyChan,
 	}
 	for _, svc := range apiServices {
 		c.addAPIService(svc)
@@ -439,6 +443,8 @@ func TestSync(t *testing.T) {
 				w.WriteHeader(tc.backendStatus)
 			}))
 			defer testServer.Close()
+			alwaysReadyChan := make(chan struct{})
+			close(alwaysReadyChan)
 
 			c := AvailableConditionController{
 				apiServiceClient:           fakeClient.ApiregistrationV1(),
@@ -448,6 +454,7 @@ func TestSync(t *testing.T) {
 				serviceResolver:            &fakeServiceResolver{url: testServer.URL},
 				proxyCurrentCertKeyContent: func() ([]byte, []byte) { return emptyCert(), emptyCert() },
 				metrics:                    availabilitymetrics.New(),
+				hasBeenReady:               alwaysReadyChan,
 			}
 			err = c.sync(tc.apiServiceName)
 			if tc.expectedSyncError != "" {