[kubevirt] Hackish minimal rhcos auto-detection

Discover the rhcos container image from the release payload based on the
checksum. The openstack mirror and the release payload use different
names for the images. The checksum is used to cross-reference the
release payload version with the quay.io/containerdisks/rhcos version.

Check for containerDisks or DataVolumeTemplates which are named
`containervolume` and have in their container source `{rhcos:version}`
set and replace it with the discovered rhcos container version.

As an example: `quay.io/containerdisks/rhcos:{rhcos:version}` would ge
replaced be modified to `quay.io/containerdisks/rhcos:mytag`, if `mytag`
would be discovered in the release payload.

Signed-off-by: Roman Mohr <rmohr@redhat.com>

api/v1alpha1/nodepool_types.go

@@ -10,6 +10,7 @@ import (
 const (
 	NodePoolValidReleaseImageConditionType       = "ValidReleaseImage"
 	NodePoolValidAMIConditionType                = "ValidAMI"
+	NodePoolValidRHCOSImageConditionType         = "ValidRHCOSImage"
 	NodePoolConfigValidConfigConditionType       = "ValidConfig"
 	NodePoolUpdateManagementEnabledConditionType = "UpdateManagementEnabled"
 	NodePoolAutoscalingEnabledConditionType      = "AutoscalingEnabled"

cmd/cluster/core/create.go

@@ -76,6 +76,7 @@ type KubevirtPlatformCreateOptions struct {
 	Memory             string
 	Cores              uint32
 	ContainerDiskImage string
+	ContainerDiskRepo  string
 }
 
 type AWSPlatformOptions struct {

cmd/cluster/kubevirt/create.go

@@ -29,6 +29,7 @@ func NewCreateCommand(opts *core.CreateOptions) *cobra.Command {
 	cmd.Flags().StringVar(&opts.KubevirtPlatform.Memory, "memory", opts.KubevirtPlatform.Memory, "The amount of memory which is visible inside the Guest OS (type BinarySI, e.g. 5Gi, 100Mi)")
 	cmd.Flags().Uint32Var(&opts.KubevirtPlatform.Cores, "cores", opts.KubevirtPlatform.Cores, "The number of cores inside the vmi, Must be a value greater or equal 1")
 	cmd.Flags().StringVar(&opts.KubevirtPlatform.ContainerDiskImage, "containerdisk", opts.KubevirtPlatform.ContainerDiskImage, "A reference to docker image with the embedded disk to be used to create the machines")
+	cmd.Flags().StringVar(&opts.KubevirtPlatform.ContainerDiskRepo, "containerdisk-repository", opts.KubevirtPlatform.ContainerDiskRepo, "A reference to docker image registry with the embedded disk to be used to create the machines, the tag will be auto-discovered")
 
 	cmd.RunE = func(cmd *cobra.Command, args []string) error {
 		ctx := cmd.Context()
@@ -64,8 +65,14 @@ func applyPlatformSpecificsValues(ctx context.Context, exampleOptions *apifixtur
 		// As long as there is no official container image
 		// The image must be provided by user
 		// Otherwise it must fail
-		if opts.KubevirtPlatform.ContainerDiskImage == "" {
-			return errors.New("the container disk image for the Kubevirt machine must be provided by user (\"--containerdisk\" flag)")
+		if opts.KubevirtPlatform.ContainerDiskImage != "" && opts.KubevirtPlatform.ContainerDiskRepo != "" {
+			return errors.New("only one of \"--containerdisk\" and \"--containerdisk-repository\" can be specified")
+		}
+		if opts.KubevirtPlatform.ContainerDiskImage == "" && opts.KubevirtPlatform.ContainerDiskRepo == "" {
+			return errors.New("the container disk image for the Kubevirt machine must be provided by user (\"--containerdisk\" or \"--containerdisk-repository\" flag)")
+		}
+		if opts.KubevirtPlatform.ContainerDiskRepo != "" {
+			opts.KubevirtPlatform.ContainerDiskImage = opts.KubevirtPlatform.ContainerDiskRepo + ":{rhcos:version}"
 		}
 	}
 

cmd/nodepool/kubevirt/create.go

@@ -2,6 +2,7 @@ package kubevirt
 
 import (
 	"context"
+	"errors"
 
 	"github.com/spf13/cobra"
 	apiresource "k8s.io/apimachinery/pkg/api/resource"
@@ -17,6 +18,7 @@ type KubevirtPlatformCreateOptions struct {
 	Memory             string
 	Cores              uint32
 	ContainerDiskImage string
+	ContainerDiskRepo  string
 }
 
 func NewCreateCommand(coreOpts *core.CreateNodePoolOptions) *cobra.Command {
@@ -34,6 +36,7 @@ func NewCreateCommand(coreOpts *core.CreateNodePoolOptions) *cobra.Command {
 	cmd.Flags().StringVar(&platformOpts.Memory, "memory", platformOpts.Memory, "The amount of memory which is visible inside the Guest OS (type BinarySI, e.g. 5Gi, 100Mi)")
 	cmd.Flags().Uint32Var(&platformOpts.Cores, "cores", platformOpts.Cores, "The number of cores inside the vmi, Must be a value greater or equal 1")
 	cmd.Flags().StringVar(&platformOpts.ContainerDiskImage, "containerdisk", platformOpts.ContainerDiskImage, "A reference to docker image with the embedded disk to be used to create the machines")
+	cmd.Flags().StringVar(&platformOpts.ContainerDiskRepo, "containerdisk-repository", platformOpts.ContainerDiskRepo, "A reference to docker image registry with the embedded disk to be used to create the machines, the tag will be auto-discovered")
 
 	// TODO (nargaman): replace with official container image, after RFE-2501 is completed
 	// As long as there is no official container image
@@ -47,6 +50,19 @@ func NewCreateCommand(coreOpts *core.CreateNodePoolOptions) *cobra.Command {
 }
 
 func (o *KubevirtPlatformCreateOptions) UpdateNodePool(_ context.Context, nodePool *hyperv1.NodePool, _ *hyperv1.HostedCluster, _ crclient.Client) error {
+	// TODO (nargaman): replace with official container image, after RFE-2501 is completed
+	// As long as there is no official container image
+	// The image must be provided by user
+	// Otherwise it must fail
+	if o.ContainerDiskImage != "" && o.ContainerDiskRepo != "" {
+		return errors.New("only one of \"--containerdisk\" and \"--containerdisk-repository\" can be specified")
+	}
+	if o.ContainerDiskImage == "" && o.ContainerDiskRepo == "" {
+		return errors.New("the container disk image for the Kubevirt machine must be provided by user (\"--containerdisk\" or \"containerdisk-repository\" flag)")
+	}
+	if o.ContainerDiskRepo != "" {
+		o.ContainerDiskImage = o.ContainerDiskRepo + ":{rhcos:version}"
+	}
 	runAlways := kubevirtv1.RunStrategyAlways
 	guestQuantity := apiresource.MustParse(o.Memory)
 	nodePool.Spec.Platform.Kubevirt = &hyperv1.KubevirtNodePoolPlatform{

hypershift-operator/controllers/nodepool/kubevirt.go

@@ -1,15 +1,36 @@
 package nodepool
 
 import (
+	"strings"
+
 	hyperv1 "github.com/openshift/hypershift/api/v1alpha1"
 	capikubevirt "sigs.k8s.io/cluster-api-provider-kubevirt/api/v1alpha1"
 )
 
-func kubevirtMachineTemplateSpec(nodePool *hyperv1.NodePool) *capikubevirt.KubevirtMachineTemplateSpec {
+func kubevirtMachineTemplateSpec(nodePool *hyperv1.NodePool, containerDiskChecksum string) *capikubevirt.KubevirtMachineTemplateSpec {
+	template := nodePool.Spec.Platform.Kubevirt.NodeTemplate.DeepCopy()
+	dataVolumeTemplates := template.Spec.DataVolumeTemplates
+	for i, dv := range dataVolumeTemplates {
+		if dv.Name == "containervolume" {
+			if dv.Spec.Source != nil && dv.Spec.Source.Registry != nil &&
+				dv.Spec.Source.Registry.URL != nil && strings.Contains(*dv.Spec.Source.Registry.URL, "{rhcos:version}") {
+				image := strings.ReplaceAll(*dv.Spec.Source.Registry.URL, "{rhcos:version}", containerDiskChecksum)
+				dataVolumeTemplates[i].Spec.Source.Registry.URL = &image
+			}
+		}
+	}
+	volumes := template.Spec.Template.Spec.Volumes
+	for i, volume := range volumes {
+		if volume.Name == "containervolume" {
+			if volume.ContainerDisk != nil && strings.Contains(volume.ContainerDisk.Image, "{rhcos:version}") {
+				volumes[i].ContainerDisk.Image = strings.ReplaceAll(volume.ContainerDisk.Image, "{rhcos:version}", containerDiskChecksum)
+			}
+		}
+	}
 	return &capikubevirt.KubevirtMachineTemplateSpec{
 		Template: capikubevirt.KubevirtMachineTemplateResource{
 			Spec: capikubevirt.KubevirtMachineSpec{
-				VirtualMachineTemplate: *nodePool.Spec.Platform.Kubevirt.NodeTemplate,
+				VirtualMachineTemplate: *template,
 			},
 		},
 	}

hypershift-operator/controllers/nodepool/nodepool_controller.go

@@ -343,7 +343,8 @@ func (r *NodePoolReconciler) reconcile(ctx context.Context, hcluster *hyperv1.Ho
 
 	// Validate platform specific input.
 	var ami string
-	if nodePool.Spec.Platform.Type == hyperv1.AWSPlatform {
+	switch nodePool.Spec.Platform.Type {
+	case hyperv1.AWSPlatform:
 		if hcluster.Spec.Platform.AWS == nil {
 			return ctrl.Result{}, fmt.Errorf("the HostedCluster for this NodePool has no .Spec.Platform.AWS, this is unsupported")
 		}
@@ -359,14 +360,33 @@ func (r *NodePoolReconciler) reconcile(ctx context.Context, hcluster *hyperv1.Ho
 			})
 			return ctrl.Result{}, fmt.Errorf("couldn't discover an AMI for release image: %w", err)
 		}
+		meta.SetStatusCondition(&nodePool.Status.Conditions, metav1.Condition{
+			Type:               hyperv1.NodePoolValidAMIConditionType,
+			Status:             metav1.ConditionTrue,
+			Reason:             hyperv1.NodePoolAsExpectedConditionReason,
+			Message:            fmt.Sprintf("Bootstrap AMI is %q", ami),
+			ObservedGeneration: nodePool.Generation,
+		})
+	case hyperv1.KubevirtPlatform:
+		ami, err = getContainerDiskChecksum(releaseImage)
+		if err != nil {
+			meta.SetStatusCondition(&nodePool.Status.Conditions, metav1.Condition{
+				Type:               hyperv1.NodePoolValidRHCOSImageConditionType,
+				Status:             metav1.ConditionFalse,
+				Reason:             hyperv1.NodePoolValidationFailedConditionReason,
+				Message:            fmt.Sprintf("Couldn't discover any openstack rhcos checksum for release image %q: %s", nodePool.Spec.Release.Image, err.Error()),
+				ObservedGeneration: nodePool.Generation,
+			})
+			return ctrl.Result{}, fmt.Errorf("couldn't discover any openstack rhcos checksum for release image: %w", err)
+		}
+		meta.SetStatusCondition(&nodePool.Status.Conditions, metav1.Condition{
+			Type:               hyperv1.NodePoolValidRHCOSImageConditionType,
+			Status:             metav1.ConditionTrue,
+			Reason:             hyperv1.NodePoolAsExpectedConditionReason,
+			Message:            fmt.Sprintf("Bootstrap openstack RHCOS image checksum is %q", ami),
+			ObservedGeneration: nodePool.Generation,
+		})
 	}
-	meta.SetStatusCondition(&nodePool.Status.Conditions, metav1.Condition{
-		Type:               hyperv1.NodePoolValidAMIConditionType,
-		Status:             metav1.ConditionTrue,
-		Reason:             hyperv1.NodePoolAsExpectedConditionReason,
-		Message:            fmt.Sprintf("Bootstrap AMI is %q", ami),
-		ObservedGeneration: nodePool.Generation,
-	})
 
 	// Validate config input.
 	// 3 generic core config resoures: fips, ssh and haproxy.
@@ -807,6 +827,10 @@ func getAMI(nodePool *hyperv1.NodePool, region string, releaseImage *releaseinfo
 	return defaultNodePoolAMI(region, releaseImage)
 }
 
+func getContainerDiskChecksum(releaseImage *releaseinfo.ReleaseImage) (string, error) {
+	return defaultContainerDiskChecksum(releaseImage)
+}
+
 func ignConfig(encodedCACert, encodedToken, endpoint string) ignitionapi.Config {
 	return ignitionapi.Config{
 		Ignition: ignitionapi.Ignition{
@@ -990,6 +1014,28 @@ func validateAutoscaling(nodePool *hyperv1.NodePool) error {
 	return nil
 }
 
+func defaultContainerDiskChecksum(releaseImage *releaseinfo.ReleaseImage) (string, error) {
+	// TODO: The architecture should be specified from the API
+	arch, foundArch := releaseImage.StreamMetadata.Architectures["x86_64"]
+	if !foundArch {
+		return "", fmt.Errorf("couldn't find OS metadata for architecture %q", "x64_64")
+	}
+	openStack, exists := arch.Artifacts["openstack"]
+	if !exists {
+		return "", fmt.Errorf("couldn't find OS metadata for openstack")
+	}
+	artifact, exists := openStack.Formats["qcow2.gz"]
+	if !exists {
+		return "", fmt.Errorf("couldn't find OS metadata for openstack artifact %v", "qcow2.gz")
+	}
+	disk, exists := artifact["disk"]
+	if !exists {
+		return "", fmt.Errorf("couldn't find OS metadata for the openstack disk")
+	}
+
+	return disk.SHA256, nil
+}
+
 func defaultNodePoolAMI(region string, releaseImage *releaseinfo.ReleaseImage) (string, error) {
 	// TODO: The architecture should be specified from the API
 	arch, foundArch := releaseImage.StreamMetadata.Architectures["x86_64"]
@@ -1319,7 +1365,7 @@ func machineTemplateBuilders(hcluster *hyperv1.HostedCluster, nodePool *hyperv1.
 		}
 	case hyperv1.KubevirtPlatform:
 		template = &capikubevirt.KubevirtMachineTemplate{}
-		machineTemplateSpec = kubevirtMachineTemplateSpec(nodePool)
+		machineTemplateSpec = kubevirtMachineTemplateSpec(nodePool, ami)
 		mutateTemplate = func(object client.Object) error {
 			o, _ := object.(*capikubevirt.KubevirtMachineTemplate)
 			o.Spec = *machineTemplateSpec.(*capikubevirt.KubevirtMachineTemplateSpec)