[decomission-controller] switch to maintenance-operator

api/v1/hypervisor_types.go

@@ -57,10 +57,11 @@ const (
 	ConditionReasonReadyEvicted     = "Evicted"
 
 	// ConditionTypeOnboarding reasons
-	ConditionReasonInitial    = "Initial"
-	ConditionReasonOnboarding = "Onboarding"
-	ConditionReasonTesting    = "Testing"
-	ConditionReasonAborted    = "Aborted"
+	ConditionReasonInitial         = "Initial"
+	ConditionReasonOnboarding      = "Onboarding"
+	ConditionReasonTesting         = "Testing"
+	ConditionReasonAborted         = "Aborted"
+	ConditionReasonDecommissioning = "Decommissioning"
 )
 
 // HypervisorSpec defines the desired state of Hypervisor

charts/openstack-hypervisor-operator/crds/hypervisor-crd.yaml

@@ -315,6 +315,14 @@ spec:
                   firmwareVersion:
                     description: FirmwareVersion
                     type: string
+                  gardenLinuxCommitID:
+                    description: Represents the Garden Linux build commit id
+                    type: string
+                  gardenLinuxFeatures:
+                    description: Represents the Garden Linux Feature Set
+                    items:
+                      type: string
+                    type: array
                   hardwareModel:
                     description: HardwareModel
                     type: string
@@ -336,6 +344,10 @@ spec:
                   prettyVersion:
                     description: PrettyVersion
                     type: string
+                  variantID:
+                    description: Identifying a specific variant or edition of the
+                      operating system
+                    type: string
                   version:
                     description: Represents the Operating System version.
                     type: string

internal/controller/aggregates_controller.go

@@ -70,7 +70,7 @@ func (ac *AggregatesController) Reconcile(ctx context.Context, req ctrl.Request)
 		return ctrl.Result{}, nil
 	}
 
-	aggs, err := aggregatesByName(ctx, ac.computeClient)
+	aggs, err := openstack.GetAggregatesByName(ctx, ac.computeClient)
 	if err != nil {
 		err = fmt.Errorf("failed listing aggregates: %w", err)
 		if err2 := ac.setErrorCondition(ctx, hv, err.Error()); err2 != nil {
@@ -163,24 +163,6 @@ func (ac *AggregatesController) SetupWithManager(mgr ctrl.Manager) error {
 		Complete(ac)
 }
 
-func aggregatesByName(ctx context.Context, serviceClient *gophercloud.ServiceClient) (map[string]*aggregates.Aggregate, error) {
-	pages, err := aggregates.List(serviceClient).AllPages(ctx)
-	if err != nil {
-		return nil, fmt.Errorf("cannot list aggregates due to %w", err)
-	}
-
-	aggs, err := aggregates.ExtractAggregates(pages)
-	if err != nil {
-		return nil, fmt.Errorf("cannot list aggregates due to %w", err)
-	}
-
-	aggregateMap := make(map[string]*aggregates.Aggregate, len(aggs))
-	for _, aggregate := range aggs {
-		aggregateMap[aggregate.Name] = &aggregate
-	}
-	return aggregateMap, nil
-}
-
 func addToAggregate(ctx context.Context, serviceClient *gophercloud.ServiceClient, aggs map[string]*aggregates.Aggregate, host, name, zone string) (err error) {
 	aggregate, found := aggs[name]
 	log := logger.FromContext(ctx)

internal/controller/decomission_controller.go

@@ -26,25 +26,24 @@ import (
 
 	"github.com/gophercloud/gophercloud/v2"
 	"github.com/gophercloud/gophercloud/v2/openstack/compute/v2/aggregates"
+	"github.com/gophercloud/gophercloud/v2/openstack/compute/v2/hypervisors"
 	"github.com/gophercloud/gophercloud/v2/openstack/compute/v2/services"
 	"github.com/gophercloud/gophercloud/v2/openstack/placement/v1/resourceproviders"
-	corev1 "k8s.io/api/core/v1"
 	"k8s.io/apimachinery/pkg/api/meta"
 	metav1 "k8s.io/apimachinery/pkg/apis/meta/v1"
 	"k8s.io/apimachinery/pkg/runtime"
 	"k8s.io/client-go/util/retry"
 	ctrl "sigs.k8s.io/controller-runtime"
 	k8sclient "sigs.k8s.io/controller-runtime/pkg/client"
-	"sigs.k8s.io/controller-runtime/pkg/controller/controllerutil"
 	logger "sigs.k8s.io/controller-runtime/pkg/log"
 
 	kvmv1 "github.com/cobaltcore-dev/openstack-hypervisor-operator/api/v1"
 	"github.com/cobaltcore-dev/openstack-hypervisor-operator/internal/openstack"
+	"github.com/cobaltcore-dev/openstack-hypervisor-operator/internal/utils"
 )
 
 const (
-	decommissionFinalizerName  = "cobaltcore.cloud.sap/decommission-hypervisor"
-	DecommissionControllerName = "nodeDecommission"
+	DecommissionControllerName = "decommission"
 )
 
 type NodeDecommissionReconciler struct {
@@ -57,145 +56,108 @@ type NodeDecommissionReconciler struct {
 // The counter-side in gardener is here:
 // https://github.com/gardener/machine-controller-manager/blob/rel-v0.56/pkg/util/provider/machinecontroller/machine.go#L646
 
-// +kubebuilder:rbac:groups="",resources=nodes,verbs=get;list;watch;patch;update
-// +kubebuilder:rbac:groups="",resources=nodes/finalizers,verbs=update
 // +kubebuilder:rbac:groups=kvm.cloud.sap,resources=hypervisors,verbs=get;list;watch
 // +kubebuilder:rbac:groups=kvm.cloud.sap,resources=hypervisors/status,verbs=get;list;watch;update;patch
+
 func (r *NodeDecommissionReconciler) Reconcile(ctx context.Context, req ctrl.Request) (ctrl.Result, error) {
-	hostname := req.Name
-	log := logger.FromContext(ctx).WithName(req.Name).WithValues("hostname", hostname)
-	ctx = logger.IntoContext(ctx, log)
+	log := logger.FromContext(ctx).WithName(req.Name)
+	hv := &kvmv1.Hypervisor{}
+	if err := retry.RetryOnConflict(retry.DefaultRetry, func() error {
+		if err := r.Get(ctx, req.NamespacedName, hv); err != nil {
+			// ignore not found errors, could be deleted
+			return k8sclient.IgnoreNotFound(err)
+		}
 
-	node := &corev1.Node{}
-	if err := r.Get(ctx, req.NamespacedName, node); err != nil {
-		return ctrl.Result{}, k8sclient.IgnoreNotFound(err)
-	}
+		setDecommissioningCondition := func(msg string) {
+			meta.SetStatusCondition(&hv.Status.Conditions, metav1.Condition{
+				Type:    kvmv1.ConditionTypeReady,
+				Status:  metav1.ConditionFalse,
+				Reason:  kvmv1.ConditionReasonDecommissioning,
+				Message: msg,
+			})
+		}
 
-	// Fetch HV to check if lifecycle management is enabled
-	hv := &kvmv1.Hypervisor{}
-	if err := r.Get(ctx, k8sclient.ObjectKey{Name: hostname}, hv); err != nil {
-		// ignore not found errors, could be deleted
-		return ctrl.Result{}, k8sclient.IgnoreNotFound(err)
-	}
-	if !hv.Spec.LifecycleEnabled {
-		// Get out of the way
-		return r.removeFinalizer(ctx, node)
-	}
+		if meta.IsStatusConditionTrue(hv.Status.Conditions, kvmv1.ConditionTypeReady) {
+			setDecommissioningCondition("Node is being decommissioned, removing host from nova")
+			return r.Status().Update(ctx, hv)
+		}
 
-	if !controllerutil.ContainsFinalizer(node, decommissionFinalizerName) {
-		return ctrl.Result{}, retry.RetryOnConflict(retry.DefaultRetry, func() error {
-			patch := k8sclient.MergeFrom(node.DeepCopy())
-			controllerutil.AddFinalizer(node, decommissionFinalizerName)
-			if err := r.Patch(ctx, node, patch); err != nil {
-				return fmt.Errorf("failed to add finalizer due to %w", err)
+		hypervisor, err := openstack.GetHypervisorByName(ctx, r.computeClient, hv.Name, true)
+		if err != nil {
+			if errors.Is(err, openstack.ErrNoHypervisor) {
+				// We are (hopefully) done
+				setDecommissioningCondition("Node not registered in nova anymore, proceeding with deletion")
+				hv.Status.Evicted = true
+				return r.Status().Update(ctx, hv)
 			}
-			log.Info("Added finalizer")
-			return nil
-		})
-	}
 
-	// Not yet deleting hv, nothing more to do
-	if node.DeletionTimestamp.IsZero() {
-		return ctrl.Result{}, nil
-	}
+			setDecommissioningCondition(fmt.Sprintf("Failed to get %q from openstack: %v", hv.Name, err))
+			return r.Status().Update(ctx, hv)
+		}
 
-	// Someone is just deleting the hv, without going through termination
-	// See: https://github.com/gardener/machine-controller-manager/blob/rel-v0.56/pkg/util/provider/machinecontroller/machine.go#L658-L659
-	if !IsNodeConditionTrue(node.Status.Conditions, "Terminating") {
-		log.Info("removing finalizer since not terminating")
-		// So we just get out of the way for now
-		return r.removeFinalizer(ctx, node)
-	}
+		if err = r.doDecomission(ctx, hv, hypervisor); err != nil {
+			log.Error(err, "Failed to decomission node", "node", hv.Name)
+			setDecommissioningCondition(err.Error())
+			return r.Status().Update(ctx, hv)
+		}
 
-	if meta.IsStatusConditionTrue(hv.Status.Conditions, kvmv1.ConditionTypeReady) {
-		return r.setDecommissioningCondition(ctx, hv, "Node is being decommissioned, removing host from nova")
+		// Decommissioning succeeded, proceed with deletion
+		hv.Status.Evicted = true
+		return r.Status().Update(ctx, hv)
+	}); err != nil {
+		return ctrl.Result{}, err
 	}
 
-	log.Info("removing host from nova")
-
-	hypervisor, err := openstack.GetHypervisorByName(ctx, r.computeClient, hostname, true)
-	if errors.Is(err, openstack.ErrNoHypervisor) {
-		// We are (hopefully) done
-		return r.removeFinalizer(ctx, node)
-	}
+	return ctrl.Result{RequeueAfter: utils.ShortRetryTime}, nil
+}
 
+func (r *NodeDecommissionReconciler) doDecomission(ctx context.Context, hv *kvmv1.Hypervisor, hypervisor *hypervisors.Hypervisor) error {
 	// TODO: remove since RunningVMs is only available until micro-version 2.87, and also is updated asynchronously
 	// so it might be not accurate
 	if hypervisor.RunningVMs > 0 {
 		// Still running VMs, cannot delete the service
-		msg := fmt.Sprintf("Node is being decommissioned, but still has %d running VMs", hypervisor.RunningVMs)
-		return r.setDecommissioningCondition(ctx, hv, msg)
+		return fmt.Errorf("node is being decommissioned, but still has %d running VMs", hypervisor.RunningVMs)
 	}
 
 	if hypervisor.Servers != nil && len(*hypervisor.Servers) > 0 {
 		// Still VMs assigned to the host, cannot delete the service
-		msg := fmt.Sprintf("Node is being decommissioned, but still has %d assigned VMs, "+
-			"check with `openstack server list --all-projects --host %s`", len(*hypervisor.Servers), hostname)
-		return r.setDecommissioningCondition(ctx, hv, msg)
+		return fmt.Errorf("node is being decommissioned, but still has %d assigned VMs, "+
+			"check with `openstack server list --all-projects --host %s`", len(*hypervisor.Servers), hv.Name)
 	}
 
-	// Before removing the service, first take the node out of the aggregates,
-	// so when the node comes back, it doesn't up with the old associations
-	aggs, err := aggregatesByName(ctx, r.computeClient)
+	// Before removing the service, first take the hypervisor out of the aggregates,
+	// so when the hypervisor comes back, it doesn't up with the old associations
+	aggs, err := openstack.GetAggregatesByName(ctx, r.computeClient)
 	if err != nil {
-		return r.setDecommissioningCondition(ctx, hv, fmt.Sprintf("cannot list aggregates due to %v", err))
+		return fmt.Errorf("cannot list aggregates due to: %w", err)
 	}
 
-	host := node.Name
+	host := hv.Name
 	for name, aggregate := range aggs {
 		if slices.Contains(aggregate.Hosts, host) {
 			opts := aggregates.RemoveHostOpts{Host: host}
 			if err = aggregates.RemoveHost(ctx, r.computeClient, aggregate.ID, opts).Err; err != nil {
-				msg := fmt.Sprintf("failed to remove host %v from aggregate %v due to %v", name, host, err)
-				return r.setDecommissioningCondition(ctx, hv, msg)
+				return fmt.Errorf("failed to remove host %v from aggregate %v due to %w", name, host, err)
 			}
 		}
 	}
 
 	// Deleting and evicted, so better delete the service
 	err = services.Delete(ctx, r.computeClient, hypervisor.Service.ID).ExtractErr()
 	if err != nil && !gophercloud.ResponseCodeIs(err, http.StatusNotFound) {
-		msg := fmt.Sprintf("cannot delete service %s due to %v", hypervisor.Service.ID, err)
-		return r.setDecommissioningCondition(ctx, hv, msg)
+		return fmt.Errorf("cannot delete service %s due to %w", hypervisor.Service.ID, err)
 	}
 
 	rp, err := resourceproviders.Get(ctx, r.placementClient, hypervisor.ID).Extract()
 	if err != nil && !gophercloud.ResponseCodeIs(err, http.StatusNotFound) {
-		return r.setDecommissioningCondition(ctx, hv, fmt.Sprintf("cannot get resource provider: %v", err))
+		return fmt.Errorf("cannot get resource provider %s due to %w", hypervisor.ID, err)
 	}
 
 	if err = openstack.CleanupResourceProvider(ctx, r.placementClient, rp); err != nil {
-		return r.setDecommissioningCondition(ctx, hv, fmt.Sprintf("cannot clean up resource provider: %v", err))
-	}
-
-	return r.removeFinalizer(ctx, node)
-}
-
-func (r *NodeDecommissionReconciler) removeFinalizer(ctx context.Context, node *corev1.Node) (ctrl.Result, error) {
-	if !controllerutil.ContainsFinalizer(node, decommissionFinalizerName) {
-		return ctrl.Result{}, nil
+		return fmt.Errorf("cannot cleanup resource provider: %w", err)
 	}
 
-	nodeBase := node.DeepCopy()
-	controllerutil.RemoveFinalizer(node, decommissionFinalizerName)
-	err := r.Patch(ctx, node, k8sclient.MergeFromWithOptions(nodeBase,
-		k8sclient.MergeFromWithOptimisticLock{}), k8sclient.FieldOwner(DecommissionControllerName))
-	return ctrl.Result{}, err
-}
-
-func (r *NodeDecommissionReconciler) setDecommissioningCondition(ctx context.Context, hv *kvmv1.Hypervisor, message string) (ctrl.Result, error) {
-	base := hv.DeepCopy()
-	meta.SetStatusCondition(&hv.Status.Conditions, metav1.Condition{
-		Type:    kvmv1.ConditionTypeReady,
-		Status:  metav1.ConditionFalse,
-		Reason:  "Decommissioning",
-		Message: message,
-	})
-	if err := r.Status().Patch(ctx, hv, k8sclient.MergeFromWithOptions(base,
-		k8sclient.MergeFromWithOptimisticLock{}), k8sclient.FieldOwner(DecommissionControllerName)); err != nil {
-		return ctrl.Result{}, fmt.Errorf("cannot update hypervisor status due to %w", err)
-	}
-	return ctrl.Result{RequeueAfter: shortRetryTime}, nil
+	return nil
 }
 
 // SetupWithManager sets up the controller with the Manager.
@@ -217,6 +179,8 @@ func (r *NodeDecommissionReconciler) SetupWithManager(mgr ctrl.Manager) error {
 
 	return ctrl.NewControllerManagedBy(mgr).
 		Named(DecommissionControllerName).
-		For(&corev1.Node{}).
+		For(&kvmv1.Hypervisor{}).
+		WithEventFilter(utils.HypervisorTerminationPredicate).
+		WithEventFilter(utils.LifecycleEnabledPredicate).
 		Complete(r)
 }