Changed the skip healthchecks to a feature flag.

Author: Mhodesty

config/manager/manager.yaml

@@ -48,6 +48,7 @@ spec:
             - --oidc-client-secret=$(OIDC_CLIENT_SECRET)
             - --oidc-issuer-url=$(OIDC_ISSUER_URL)
             - --only-public-clusters=$(ONLY_PUBLIC_CLUSTERS)
+            - --skip-infrastructure-tests=$(SKIP_INFRASTRUCTURE_TESTS)
           env:
             - name: LOG_LEVEL
               # level 1 is debug, so when we want to raise the level we can
@@ -70,6 +71,8 @@ spec:
               value: ""  # Default to empty - can be overridden in app-interface
             - name: ONLY_PUBLIC_CLUSTERS
               value: "false"  # Default to false - can be overridden in app-interface
+            - name: SKIP_INFRASTRUCTURE_TESTS
+              value: "false"  # Default to false - can be overridden in app-interface
           image: controller:latest
           name: manager
           securityContext:

controllers/hostedcontrolplane/healthcheck.go

@@ -63,9 +63,9 @@ const (
 // HCP namespace, and this process will be restarted. Additionally, should healthchecking need to be skipped for any reason, the annotation
 // "routemonitor.managed.openshift.io/successful-healthchecks" can be added-to/edited-on the configmap with a large number (ie - 999) to bypass
 // this functionality
-func (r *HostedControlPlaneReconciler) hcpReady(ctx context.Context, hostedcontrolplane *hypershiftv1beta1.HostedControlPlane) (bool, error) {
-	// Skip health check for test HCPs (e.g., osde2e tests without real kube-apiserver)
-	if _, osde2eTesting := hostedcontrolplane.Annotations["routemonitor.openshift.io/osde2e-testing"]; osde2eTesting {
+func (r *HostedControlPlaneReconciler) hcpReady(ctx context.Context, hostedcontrolplane *hypershiftv1beta1.HostedControlPlane, cfg RHOBSConfig) (bool, error) {
+	// Skip health check for test environments (e.g., osde2e tests without real kube-apiserver)
+	if cfg.SkipInfrastructureTests {
 		return true, nil
 	}
 
@@ -232,9 +232,9 @@ func olderThan(obj metav1.Object, age time.Duration) bool {
 }
 
 // isVpcEndpointReady checks if the VPC Endpoint associated with the HostedControlPlane is ready.
-func (r *HostedControlPlaneReconciler) isVpcEndpointReady(ctx context.Context, hostedcontrolplane *hypershiftv1beta1.HostedControlPlane) (bool, error) {
-	// Skip VPC endpoint check for test HCPs (e.g., osde2e tests without real VPC infrastructure)
-	if _, osde2eTesting := hostedcontrolplane.Annotations["routemonitor.openshift.io/osde2e-testing"]; osde2eTesting {
+func (r *HostedControlPlaneReconciler) isVpcEndpointReady(ctx context.Context, hostedcontrolplane *hypershiftv1beta1.HostedControlPlane, cfg RHOBSConfig) (bool, error) {
+	// Skip VPC endpoint check for test environments (e.g., osde2e tests without real VPC infrastructure)
+	if cfg.SkipInfrastructureTests {
 		return true, nil
 	}
 

controllers/hostedcontrolplane/hostedcontrolplane.go

@@ -133,6 +133,9 @@ func (r *HostedControlPlaneReconciler) getRHOBSConfig(ctx context.Context) RHOBS
 	if strings.TrimSpace(configMap.Data["only-public-clusters"]) == "true" {
 		cfg.OnlyPublicClusters = true
 	}
+	if strings.TrimSpace(configMap.Data["skip-infrastructure-tests"]) == "true" {
+		cfg.SkipInfrastructureTests = true
+	}
 
 	return cfg
 }
@@ -260,7 +263,7 @@ func (r *HostedControlPlaneReconciler) Reconcile(ctx context.Context, req ctrl.R
 	}
 
 	// Ensure cluster is ready to be monitored before deploying any probing objects
-	hcpReady, err := r.hcpReady(ctx, hostedcontrolplane)
+	hcpReady, err := r.hcpReady(ctx, hostedcontrolplane, rhobsConfig)
 	if err != nil {
 		log.Error(err, "HCP readiness check failed")
 		return utilreconcile.RequeueWith(fmt.Errorf("HCP readiness check failed: %v", err))
@@ -270,7 +273,7 @@ func (r *HostedControlPlaneReconciler) Reconcile(ctx context.Context, req ctrl.R
 		return utilreconcile.RequeueAfter(healthcheckIntervalSeconds * time.Second), nil
 	}
 
-	vpcEndpointReady, err := r.isVpcEndpointReady(ctx, hostedcontrolplane)
+	vpcEndpointReady, err := r.isVpcEndpointReady(ctx, hostedcontrolplane, rhobsConfig)
 	if err != nil {
 		log.Error(err, "VPC Endpoint check failed")
 		return utilreconcile.RequeueWith(err)
@@ -282,7 +285,7 @@ func (r *HostedControlPlaneReconciler) Reconcile(ctx context.Context, req ctrl.R
 
 	// Cluster ready - deploy kube-apiserver monitoring objects
 	log.Info("Deploying internal monitoring objects")
-	err = r.deployInternalMonitoringObjects(ctx, log, hostedcontrolplane)
+	err = r.deployInternalMonitoringObjects(ctx, log, hostedcontrolplane, rhobsConfig)
 	if err != nil {
 		log.Error(err, "failed to deploy internal monitoring components")
 		return utilreconcile.RequeueWith(err)
@@ -321,9 +324,9 @@ func (r *HostedControlPlaneReconciler) Reconcile(ctx context.Context, req ctrl.R
 }
 
 // deployInternalMonitoringObjects creates or updates the objects needed to monitor the kube-apiserver using cluster-internal routes
-func (r *HostedControlPlaneReconciler) deployInternalMonitoringObjects(ctx context.Context, log logr.Logger, hostedcontrolplane *hypershiftv1beta1.HostedControlPlane) error {
-	// Skip internal monitoring for test HCPs (e.g., osde2e tests without real kube-apiserver infrastructure)
-	if _, osde2eTesting := hostedcontrolplane.Annotations["routemonitor.openshift.io/osde2e-testing"]; osde2eTesting {
+func (r *HostedControlPlaneReconciler) deployInternalMonitoringObjects(ctx context.Context, log logr.Logger, hostedcontrolplane *hypershiftv1beta1.HostedControlPlane, cfg RHOBSConfig) error {
+	// Skip internal monitoring for test environments (e.g., osde2e tests without real kube-apiserver infrastructure)
+	if cfg.SkipInfrastructureTests {
 		return nil
 	}
 

controllers/hostedcontrolplane/hostedcontrolplane_test.go

@@ -571,7 +571,7 @@ func TestHostedControlPlaneReconciler_deployInternalMonitoringObjects(t *testing
 	for _, tt := range tests {
 		t.Run(tt.name, func(t *testing.T) {
 			r := newTestReconciler(t, tt.objs...)
-			err := r.deployInternalMonitoringObjects(tt.args.ctx, tt.args.log, tt.args.hostedcontrolplane)
+			err := r.deployInternalMonitoringObjects(tt.args.ctx, tt.args.log, tt.args.hostedcontrolplane, RHOBSConfig{})
 			passed, reason := tt.eval(err, r)
 			if !passed {
 				t.Errorf("HostedControlPlaneReconciler.deployInternalMonitoringObjects() did not pass due to = %s", reason)
@@ -827,7 +827,7 @@ func TestIsVpcEndpointReady(t *testing.T) {
 			}
 
 			// Test the function
-			result, err := r.isVpcEndpointReady(context.Background(), hcp)
+			result, err := r.isVpcEndpointReady(context.Background(), hcp, RHOBSConfig{})
 
 			// Validate the results
 			if result != tt.expectedResult {

go.mod

@@ -14,7 +14,7 @@ require (
 	github.com/openshift/api v0.0.0-20250409155250-8fcc4e71758a
 	github.com/openshift/aws-vpce-operator v0.0.0-20241211114942-1daecf2e4364
 	github.com/openshift/hypershift/api v0.0.0-20241204143212-857ccab4fd7c
-	github.com/openshift/osde2e v0.0.0-20260202205050-a418c410e9bc
+	github.com/openshift/osde2e v0.0.0-20260209164730-1a4541c8e2b1
 	github.com/openshift/osde2e-common v0.0.0-20260126213857-862206787cf9
 	github.com/prometheus-operator/prometheus-operator/pkg/apis/monitoring v0.64.0
 	github.com/prometheus/common v0.67.5

go.sum

@@ -230,8 +230,8 @@ github.com/openshift/aws-vpce-operator v0.0.0-20241211114942-1daecf2e4364 h1:wrR
 github.com/openshift/aws-vpce-operator v0.0.0-20241211114942-1daecf2e4364/go.mod h1:+/t18O2NwRYEzWbi76OImNMivJXi6bbFVG3ODGeQjMk=
 github.com/openshift/hypershift/api v0.0.0-20241204143212-857ccab4fd7c h1:NBH3no8kaW7Hem/mf/0lIfZZEIeaNTwo7XfTqUm2cJc=
 github.com/openshift/hypershift/api v0.0.0-20241204143212-857ccab4fd7c/go.mod h1:K1e56UPfFQaO8BYXbVS1TbNqb6rmPo+fc6Jxjv+w0Xo=
-github.com/openshift/osde2e v0.0.0-20260202205050-a418c410e9bc h1:W6qCZPI/2f64vRbDbUH0wrVzXTcjV5UBBTrIfD/giO8=
-github.com/openshift/osde2e v0.0.0-20260202205050-a418c410e9bc/go.mod h1:VW4cnPpR92sfzwuhzPWatXYAo7/ZgyjYhCDPPONjUfk=
+github.com/openshift/osde2e v0.0.0-20260209164730-1a4541c8e2b1 h1:2tqxJRv06Psmb1r9CYHEhjK0FW/uZmH6ZorQn9wJHT4=
+github.com/openshift/osde2e v0.0.0-20260209164730-1a4541c8e2b1/go.mod h1:VW4cnPpR92sfzwuhzPWatXYAo7/ZgyjYhCDPPONjUfk=
 github.com/openshift/osde2e-common v0.0.0-20260126213857-862206787cf9 h1:oUYQpPED774xWW41aRfrhvPSEcIN3FR6XeePOVqcljk=
 github.com/openshift/osde2e-common v0.0.0-20260126213857-862206787cf9/go.mod h1:nMeOffePy+qdcCXuwTw5YcMMn/cTfjYQf7KRLjhKweo=
 github.com/pelletier/go-toml/v2 v2.2.2 h1:aYUidT7k73Pcl9nb2gScu7NSrKCSHIDE89b3+6Wq+LM=

main.go

@@ -94,6 +94,7 @@ func main() {
 	var oidcClientSecret string
 	var oidcIssuerURL string
 	var onlyPublicClusters bool
+	var skipInfrastructureTests bool
 
 	flag.StringVar(&blackboxExporterImage, "blackbox-image", "quay.io/prometheus/blackbox-exporter@sha256:b04a9fef4fa086a02fc7fcd8dcdbc4b7b35cc30cdee860fdc6a19dd8b208d63e", "The image that will be used for the blackbox-exporter deployment")
 	flag.StringVar(&blackboxExporterNamespace, "blackbox-namespace", config.OperatorNamespace, "Blackbox-exporter deployment will reside on this Namespace")
@@ -103,6 +104,7 @@ func main() {
 	flag.StringVar(&oidcClientSecret, "oidc-client-secret", "", "OIDC client secret for RHOBS API authentication. When empty, no OIDC authentication is used.")
 	flag.StringVar(&oidcIssuerURL, "oidc-issuer-url", "", "OIDC issuer URL for RHOBS API authentication. When empty, no OIDC authentication is used.")
 	flag.BoolVar(&onlyPublicClusters, "only-public-clusters", false, "When true, only create RHOBS probes for public (non-private) HostedClusters. Defaults to false (process all clusters).")
+	flag.BoolVar(&skipInfrastructureTests, "skip-infrastructure-tests", false, "When true, skip infrastructure health checks (HCP ready, VPC endpoint ready) for test environments. Defaults to false.")
 
 	opts := zap.Options{}
 	opts.BindFlags(flag.CommandLine)
@@ -167,6 +169,14 @@ func main() {
 			flagParams = append(flagParams, "only-public-clusters")
 		}
 
+		if configData.SkipInfrastructureTests {
+			setupLog.V(1).Info("Using skip-infrastructure-tests from ConfigMap", "skipInfrastructureTests", configData.SkipInfrastructureTests)
+			skipInfrastructureTests = configData.SkipInfrastructureTests
+			configMapParams = append(configMapParams, "skip-infrastructure-tests")
+		} else {
+			flagParams = append(flagParams, "skip-infrastructure-tests")
+		}
+
 		// Summarize configuration sources
 		if len(configMapParams) > 0 && len(flagParams) > 0 {
 			setupLog.Info("Using mixed configuration sources",
@@ -269,12 +279,13 @@ func main() {
 
 	if enableHCP {
 		rhobsConfig := hostedcontrolplane.RHOBSConfig{
-			ProbeAPIURL:        probeAPIURL,
-			Tenant:             probeTenant,
-			OIDCClientID:       oidcClientID,
-			OIDCClientSecret:   oidcClientSecret,
-			OIDCIssuerURL:      oidcIssuerURL,
-			OnlyPublicClusters: onlyPublicClusters,
+			ProbeAPIURL:             probeAPIURL,
+			Tenant:                  probeTenant,
+			OIDCClientID:            oidcClientID,
+			OIDCClientSecret:        oidcClientSecret,
+			OIDCIssuerURL:           oidcIssuerURL,
+			OnlyPublicClusters:      onlyPublicClusters,
+			SkipInfrastructureTests: skipInfrastructureTests,
 		}
 		hostedControlPlaneReconciler := hostedcontrolplane.NewHostedControlPlaneReconciler(mgr, rhobsConfig)
 		if err = hostedControlPlaneReconciler.SetupWithManager(mgr); err != nil {
@@ -326,12 +337,13 @@ func shouldEnableHCP() (bool, error) {
 
 // OperatorConfig holds configuration values from ConfigMap
 type OperatorConfig struct {
-	ProbeAPIURL        string
-	ProbeTenant        string
-	OIDCClientID       string
-	OIDCClientSecret   string
-	OIDCIssuerURL      string
-	OnlyPublicClusters bool
+	ProbeAPIURL             string
+	ProbeTenant             string
+	OIDCClientID            string
+	OIDCClientSecret        string
+	OIDCIssuerURL           string
+	OnlyPublicClusters      bool
+	SkipInfrastructureTests bool
 }
 
 // getConfigFromConfigMap reads configuration from the route-monitor-operator-config ConfigMap
@@ -363,12 +375,13 @@ func getConfigFromConfigMap() (*OperatorConfig, error) {
 
 	// Extract configuration values, trimming whitespace
 	cfg := &OperatorConfig{
-		ProbeAPIURL:        strings.TrimSpace(configMap.Data["probe-api-url"]),
-		ProbeTenant:        strings.TrimSpace(configMap.Data["probe-tenant"]),
-		OIDCClientID:       strings.TrimSpace(configMap.Data["oidc-client-id"]),
-		OIDCClientSecret:   strings.TrimSpace(configMap.Data["oidc-client-secret"]),
-		OIDCIssuerURL:      strings.TrimSpace(configMap.Data["oidc-issuer-url"]),
-		OnlyPublicClusters: strings.TrimSpace(configMap.Data["only-public-clusters"]) == "true",
+		ProbeAPIURL:             strings.TrimSpace(configMap.Data["probe-api-url"]),
+		ProbeTenant:             strings.TrimSpace(configMap.Data["probe-tenant"]),
+		OIDCClientID:            strings.TrimSpace(configMap.Data["oidc-client-id"]),
+		OIDCClientSecret:        strings.TrimSpace(configMap.Data["oidc-client-secret"]),
+		OIDCIssuerURL:           strings.TrimSpace(configMap.Data["oidc-issuer-url"]),
+		OnlyPublicClusters:      strings.TrimSpace(configMap.Data["only-public-clusters"]) == "true",
+		SkipInfrastructureTests: strings.TrimSpace(configMap.Data["skip-infrastructure-tests"]) == "true",
 	}
 
 	// Log detailed information about what was found in the ConfigMap
@@ -411,6 +424,12 @@ func getConfigFromConfigMap() (*OperatorConfig, error) {
 		missingParams = append(missingParams, "only-public-clusters")
 	}
 
+	if configMap.Data["skip-infrastructure-tests"] != "" {
+		foundParams = append(foundParams, "skip-infrastructure-tests")
+	} else {
+		missingParams = append(missingParams, "skip-infrastructure-tests")
+	}
+
 	setupLog.Info("ConfigMap found and processed",
 		"configmap", configMapName,
 		"namespace", config.OperatorNamespace,