feat: add check on 2 MultiKueue CRD if v1alpha1 version in the cluste…

…r, error out

- DSC condition on kueueReady:
	reason: MultiKueueCRDV1Alpah1Exist
	message: MultiKueue CRDs MultiKueueConfig v1alpha1 and MultiKueueCluster v1alpah1 exist, please remove them before proceed
- reconcile gets triggered by user's deletion on these 2 CRD

Signed-off-by: Wen Zhou <[email protected]>

apis/components/v1alpha1/kserve_types.go

@@ -76,8 +76,8 @@ type KserveSpec struct {
 type KserveCommonStatus struct {
 	// DefaultDeploymentMode is the value of the defaultDeploymentMode field
 	// as read from the "deploy" JSON in the inferenceservice-config ConfigMap
-	DefaultDeploymentMode string                     `json:"defaultDeploymentMode,omitempty"`
-	ServerlessMode        operatorv1.ManagementState `json:"serverlessMode,omitempty"`
+	DefaultDeploymentMode         string                     `json:"defaultDeploymentMode,omitempty"`
+	ServerlessMode                operatorv1.ManagementState `json:"serverlessMode,omitempty"`
 	common.ComponentReleaseStatus `json:",inline"`
 }
 

bundle/manifests/opendatahub-operator.clusterserviceversion.yaml

@@ -114,16 +114,16 @@ metadata:
     categories: AI/Machine Learning, Big Data
     certified: "False"
     containerImage: quay.io/opendatahub/opendatahub-operator:v2.23.1
-    createdAt: "2025-02-04T09:17:15Z"
+    createdAt: "2025-02-05T09:18:14Z"
     olm.skipRange: '>=1.0.0 <2.23.1'
     operators.operatorframework.io/builder: operator-sdk-v1.31.0
     operators.operatorframework.io/internal-objects: '["featuretrackers.features.opendatahub.io",
       "codeflares.components.platform.opendatahub.io", "dashboards.components.platform.opendatahub.io",
       "datasciencepipelines.components.platform.opendatahub.io", "kserves.components.platform.opendatahub.io",
       "kueues.components.platform.opendatahub.io", "modelmeshservings.components.platform.opendatahub.io",
       "modelregistries.components.platform.opendatahub.io", "rays.components.platform.opendatahub.io",
-      "trainingoperators.components.platform.opendatahub.io", "trustyais.components.platform.opendatahub.io", "workbenches.components.platform.opendatahub.io",
-      "monitorings.services.platform.opendatahub.io","modelcontrollers.components.platform.opendatahub.io",
+      "trainingoperators.components.platform.opendatahub.io", "trustyais.components.platform.opendatahub.io",
+      "workbenches.components.platform.opendatahub.io", "monitorings.services.platform.opendatahub.io","modelcontrollers.components.platform.opendatahub.io",
       "feastoperators.components.platform.opendatahub.io"]'
     operators.operatorframework.io/project_layout: go.kubebuilder.io/v3
     repository: https://github.com/opendatahub-io/opendatahub-operator
@@ -1317,7 +1317,7 @@ spec:
                   value: /opt/manifests
                 - name: ODH_PLATFORM_TYPE
                   value: OpenDataHub
-                image: quay.io/ugiordan/opendatahub-operator:v2.18.0-versions-mapping
+                image: REPLACE_IMAGE:latest
                 imagePullPolicy: Always
                 livenessProbe:
                   httpGet:

controllers/components/kueue/kueue_controller.go

@@ -28,6 +28,8 @@ import (
 	rbacv1 "k8s.io/api/rbac/v1"
 	extv1 "k8s.io/apiextensions-apiserver/pkg/apis/apiextensions/v1"
 	ctrl "sigs.k8s.io/controller-runtime"
+	"sigs.k8s.io/controller-runtime/pkg/event"
+	"sigs.k8s.io/controller-runtime/pkg/predicate"
 
 	componentApi "github.com/opendatahub-io/opendatahub-operator/v2/apis/components/v1alpha1"
 	"github.com/opendatahub-io/opendatahub-operator/v2/pkg/cluster"
@@ -67,6 +69,20 @@ func (s *componentHandler) NewComponentReconciler(ctx context.Context, mgr ctrl.
 		Owns(&admissionregistrationv1.MutatingWebhookConfiguration{}).
 		Owns(&admissionregistrationv1.ValidatingWebhookConfiguration{}).
 		Owns(&appsv1.Deployment{}, reconciler.WithPredicates(resources.NewDeploymentPredicate())).
+		Watches(
+			&extv1.CustomResourceDefinition{},
+			reconciler.WithEventHandler(
+				handlers.ToNamed(componentApi.KueueInstanceName)),
+			reconciler.WithPredicates(predicate.Or(
+				component.ForLabel(labels.ODH.Component(LegacyComponentName), labels.True),
+				// For the case when user manually delete old MultiKueueConfig/MultiKueueCluster CRDs v1alpha1
+				predicate.Funcs{
+					DeleteFunc: func(e event.DeleteEvent) bool {
+						return e.Object.GetName() == gvk.MultiKueueConfigV1Alpha1.Kind || e.Object.GetName() == gvk.MultikueueClusterV1Alpha1.Kind
+					},
+				},
+			)),
+		).
 		Watches(
 			&extv1.CustomResourceDefinition{},
 			reconciler.WithEventHandler(
@@ -75,6 +91,7 @@ func (s *componentHandler) NewComponentReconciler(ctx context.Context, mgr ctrl.
 				component.ForLabel(labels.ODH.Component(LegacyComponentName), labels.True)),
 		).
 		// Add Kueue-specific actions
+		WithAction(checkPreConditions). // check if CRD multikueueconfigs/multikueueclusters with v1alpha1 exist in cluster and not in termination
 		WithAction(initialize).
 		WithAction(devFlags).
 		WithAction(releases.NewAction()).

controllers/components/kueue/kueue_controller_actions.go

@@ -4,14 +4,46 @@ import (
 	"context"
 	"fmt"
 
+	"k8s.io/apimachinery/pkg/api/meta"
+	metav1 "k8s.io/apimachinery/pkg/apis/meta/v1"
+
 	componentApi "github.com/opendatahub-io/opendatahub-operator/v2/apis/components/v1alpha1"
+	"github.com/opendatahub-io/opendatahub-operator/v2/controllers/status"
+	"github.com/opendatahub-io/opendatahub-operator/v2/pkg/cluster"
 	"github.com/opendatahub-io/opendatahub-operator/v2/pkg/cluster/gvk"
+	odherrors "github.com/opendatahub-io/opendatahub-operator/v2/pkg/controller/actions/errors"
 	odhtypes "github.com/opendatahub-io/opendatahub-operator/v2/pkg/controller/types"
 	odhdeploy "github.com/opendatahub-io/opendatahub-operator/v2/pkg/deploy"
 	"github.com/opendatahub-io/opendatahub-operator/v2/pkg/metadata/annotations"
 	"github.com/opendatahub-io/opendatahub-operator/v2/pkg/resources"
 )
 
+func checkPreConditions(ctx context.Context, rr *odhtypes.ReconciliationRequest) error {
+	k, ok := rr.Instance.(*componentApi.Kueue)
+	if !ok {
+		return fmt.Errorf("resource instance %v is not a componentApi.Kueue)", rr.Instance)
+	}
+
+	rConfig, eConfig := cluster.CRDVersioNotExists(ctx, rr.Client, gvk.MultiKueueConfigV1Alpha1.GroupKind(), gvk.MultiKueueConfigV1Alpha1.Version)
+	rCluster, eCluster := cluster.CRDVersioNotExists(ctx, rr.Client, gvk.MultikueueClusterV1Alpha1.GroupKind(), gvk.MultikueueClusterV1Alpha1.Version)
+	if eConfig != nil || eCluster != nil {
+		return odherrors.NewStopError("failed to check CRDs version: %v, %v", eConfig, eCluster)
+	}
+	if !rConfig || !rCluster {
+		s := k.GetStatus()
+		s.Phase = status.PhaseNotReady
+		meta.SetStatusCondition(&s.Conditions, metav1.Condition{
+			Type:               status.ConditionTypeReady,
+			Status:             metav1.ConditionFalse,
+			Reason:             status.MultiKueueCRDReason,
+			Message:            status.MultiKueueCRDMessage,
+			ObservedGeneration: s.ObservedGeneration,
+		})
+		return odherrors.NewStopError(status.MultiKueueCRDMessage)
+	}
+	return nil
+}
+
 func initialize(_ context.Context, rr *odhtypes.ReconciliationRequest) error {
 	rr.Manifests = append(rr.Manifests, manifestsPath())
 	return nil

controllers/status/status.go

@@ -102,6 +102,12 @@ const (
 		"remove existing Argo workflows or set `spec.components.datasciencepipelines.managementState` to Removed to proceed"
 )
 
+// For Kueue MultiKueue CRD.
+const (
+	MultiKueueCRDReason  = "MultiKueueCRDV1Alpha1Exist"
+	MultiKueueCRDMessage = "MultiKueue CRDs MultiKueueConfig v1alpha1 and MultiKueueCluster v1Alpha1 exist, please remove them before proceed"
+)
+
 // SetProgressingCondition sets the ProgressingCondition to True and other conditions to false or
 // Unknown. Used when we are just starting to reconcile, and there are no existing conditions.
 func SetProgressingCondition(conditions *[]conditionsv1.Condition, reason string, message string) {

pkg/cluster/gvk/gvk.go

@@ -250,4 +250,16 @@ var (
 		Version: "v1",
 		Kind:    "ValidatingAdmissionPolicyBinding",
 	}
+
+	MultiKueueConfigV1Alpha1 = schema.GroupVersionKind{
+		Group:   "kueue.x-k8s.io",
+		Version: "v1alpha1",
+		Kind:    "MultiKueueConfig",
+	}
+
+	MultikueueClusterV1Alpha1 = schema.GroupVersionKind{
+		Group:   "kueue.x-k8s.io",
+		Version: "v1alpha1",
+		Kind:    "MultiKueueCluster",
+	}
 )

pkg/cluster/operator.go

@@ -100,3 +100,28 @@ func CustomResourceDefinitionExists(ctx context.Context, cli client.Client, crdG
 
 	return err
 }
+
+// return true if not found, return false if not found.
+// checks on both CRD API version also if it is under deletion.
+func CRDVersioNotExists(ctx context.Context, cli client.Client, crdGK schema.GroupKind, version string) (bool, error) {
+	crd := &apiextv1.CustomResourceDefinition{}
+	name := strings.ToLower(fmt.Sprintf("%ss.%s", crdGK.Kind, crdGK.Group))
+	err := cli.Get(ctx, client.ObjectKey{Name: name}, crd)
+	if err != nil {
+		if errors.IsNotFound(err) {
+			return true, nil
+		}
+		return true, err
+	}
+	for _, v := range crd.Status.StoredVersions {
+		if v == version {
+			for _, condition := range crd.Status.Conditions {
+				if condition.Type == apiextv1.Terminating && condition.Status == apiextv1.ConditionTrue {
+					return true, nil
+				}
+			}
+			return false, nil
+		}
+	}
+	return true, nil
+}