feat: add cube and gpu support via templates

Author: lucasl0st

api/v1alpha1/ionoscloudmachine_types.go

@@ -98,6 +98,10 @@ const (
 	ServerTypeEnterprise ServerType = "ENTERPRISE"
 	// ServerTypeVCPU server of type VCPU.
 	ServerTypeVCPU ServerType = "VCPU"
+	// ServerTypeCUBE server of type CUBE.
+	ServerTypeCUBE ServerType = "CUBE"
+	// ServerTypeGPU server of type GPU.
+	ServerTypeGPU ServerType = "GPU"
 )
 
 // String returns the string representation of the ServerType.
@@ -107,6 +111,8 @@ func (a ServerType) String() string {
 
 // IonosCloudMachineSpec defines the desired state of IonosCloudMachine.
 // +kubebuilder:validation:XValidation:rule="!has(oldSelf.networkID) || has(self.networkID)", message="networkID is required once set"
+// +kubebuilder:validation:XValidation:rule="(self.type != 'CUBE' && self.type != 'GPU') || self.templateID != ''",message="templateID is required when type is CUBE or GPU"
+// +kubebuilder:validation:XValidation:rule="(self.type == 'CUBE' || self.type == 'GPU') || !has(self.templateID) || self.templateID == ''",message="templateID must only be used with CUBE or GPU type"
 type IonosCloudMachineSpec struct {
 	// ProviderID is the IONOS Cloud provider ID
 	// will be in the format ionos://ee090ff2-1eef-48ec-a246-a51a33aa4f3a
@@ -118,7 +124,17 @@ type IonosCloudMachineSpec struct {
 	//+kubebuilder:validation:Format=uuid
 	DatacenterID string `json:"datacenterID"`
 
+	// TemplateID is the ID of the template for creating CUBE or GPU servers.
+	// If a template has GPU cards assigned, then it can only be used to create GPU servers,
+	// otherwise it can only be used for CUBE servers.
+	// NumCores, MemoryMB, CPUFamily and the size parameter of the first volume (boot Disk) are ignored when this is set.
+	// Available TemplateIDs can be listed using the cloud api.
+	//+kubebuilder:validation:Format=uuid
+	//+optional
+	TemplateID string `json:"templateID,omitempty"`
+
 	// NumCores defines the number of cores for the VM.
+	// Ignored when TemplateID is specified.
 	//+kubebuilder:validation:Minimum=1
 	//+kubebuilder:default=1
 	//+optional
@@ -133,6 +149,7 @@ type IonosCloudMachineSpec struct {
 	// MemoryMB is the memory size for the VM in MB.
 	// Size must be specified in multiples of 256 MB with a minimum of 1024 MB
 	// which is required as we are using hot-pluggable RAM by default.
+	// Ignored when TemplateID is specified.
 	//+kubebuilder:validation:MultipleOf=1024
 	//+kubebuilder:validation:Minimum=2048
 	//+kubebuilder:default=3072
@@ -144,6 +161,7 @@ type IonosCloudMachineSpec struct {
 	//
 	// If not specified, the cloud will select a suitable CPU family
 	// based on the availability in the data center.
+	// Ignored when TemplateID is specified.
 	//+kubebuilder:example=AMD_OPTERON
 	//+optional
 	CPUFamily *string `json:"cpuFamily,omitempty"`
@@ -180,9 +198,10 @@ type IonosCloudMachineSpec struct {
 	// +kubebuilder:validation:MinLength=1
 	NetworkID *string `json:"networkID,omitempty"`
 
-	// Type is the server type of the VM. Can be either ENTERPRISE or VCPU.
+	// Type is the server type of the VM. Can be either ENTERPRISE, VCPU, CUBE or GPU.
+	// Use the types CUBE or GPU together with TemplateID.
 	//+kubebuilder:validation:XValidation:rule="self == oldSelf",message="type is immutable"
-	//+kubebuilder:validation:Enum=ENTERPRISE;VCPU
+	//+kubebuilder:validation:Enum=ENTERPRISE;VCPU;CUBE;GPU
 	//+kubebuilder:default=ENTERPRISE
 	//+optional
 	Type ServerType `json:"type,omitempty"`
@@ -216,12 +235,14 @@ type Volume struct {
 	Name string `json:"name,omitempty"`
 
 	// DiskType defines the type of the hard drive.
+	// Ignored when TemplateID of IonosCloudMachineSpec is specified (for the boot volume).
 	//+kubebuilder:validation:Enum=HDD;SSD Standard;SSD Premium
 	//+kubebuilder:default=HDD
 	//+optional
 	DiskType VolumeDiskType `json:"diskType,omitempty"`
 
-	// SizeGB defines the size of the volume in GB
+	// SizeGB defines the size of the volume in GB.
+	// Ignored when TemplateID of IonosCloudMachineSpec is specified (for the boot volume).
 	//+kubebuilder:validation:Minimum=10
 	//+kubebuilder:default=20
 	//+optional
@@ -392,7 +413,7 @@ type IonosCloudMachine struct {
 	metav1.TypeMeta   `json:",inline"`
 	metav1.ObjectMeta `json:"metadata,omitempty"`
 
-	//+kubebuilder:validation:XValidation:rule="self.type != 'VCPU' || !has(self.cpuFamily)",message="cpuFamily must not be specified when using VCPU"
+	//+kubebuilder:validation:XValidation:rule="(self.type != 'VCPU' && self.type != 'CUBE' && self.type != 'GPU') || !has(self.cpuFamily)",message="cpuFamily must not be specified when using VCPU, CUBE or GPU"
 	Spec   IonosCloudMachineSpec   `json:"spec,omitempty"`
 	Status IonosCloudMachineStatus `json:"status,omitempty"`
 }

config/crd/bases/infrastructure.cluster.x-k8s.io_ionoscloudmachines.yaml

@@ -164,6 +164,7 @@ spec:
 
                   If not specified, the cloud will select a suitable CPU family
                   based on the availability in the data center.
+                  Ignored when TemplateID is specified.
                 example: AMD_OPTERON
                 type: string
               datacenterID:
@@ -189,7 +190,9 @@ spec:
                     type: string
                   diskType:
                     default: HDD
-                    description: DiskType defines the type of the hard drive.
+                    description: |-
+                      DiskType defines the type of the hard drive.
+                      Ignored when TemplateID of IonosCloudMachineSpec is specified (for the boot volume).
                     enum:
                     - HDD
                     - SSD Standard
@@ -242,7 +245,9 @@ spec:
                     type: string
                   sizeGB:
                     default: 20
-                    description: SizeGB defines the size of the volume in GB
+                    description: |-
+                      SizeGB defines the size of the volume in GB.
+                      Ignored when TemplateID of IonosCloudMachineSpec is specified (for the boot volume).
                     minimum: 10
                     type: integer
                 required:
@@ -325,6 +330,7 @@ spec:
                   MemoryMB is the memory size for the VM in MB.
                   Size must be specified in multiples of 256 MB with a minimum of 1024 MB
                   which is required as we are using hot-pluggable RAM by default.
+                  Ignored when TemplateID is specified.
                 format: int32
                 minimum: 2048
                 multipleOf: 1024
@@ -343,7 +349,9 @@ spec:
                   rule: self == oldSelf
               numCores:
                 default: 1
-                description: NumCores defines the number of cores for the VM.
+                description: |-
+                  NumCores defines the number of cores for the VM.
+                  Ignored when TemplateID is specified.
                 format: int32
                 minimum: 1
                 type: integer
@@ -352,13 +360,25 @@ spec:
                   ProviderID is the IONOS Cloud provider ID
                   will be in the format ionos://ee090ff2-1eef-48ec-a246-a51a33aa4f3a
                 type: string
+              templateID:
+                description: |-
+                  TemplateID is the ID of the template for creating CUBE or GPU servers.
+                  If a template has GPU cards assigned, then it can only be used to create GPU servers,
+                  otherwise it can only be used for CUBE servers.
+                  NumCores, MemoryMB, CPUFamily and the size parameter of the first volume (boot Disk) are ignored when this is set.
+                  Available TemplateIDs can be listed using the cloud api.
+                format: uuid
+                type: string
               type:
                 default: ENTERPRISE
-                description: Type is the server type of the VM. Can be either ENTERPRISE
-                  or VCPU.
+                description: |-
+                  Type is the server type of the VM. Can be either ENTERPRISE, VCPU, CUBE or GPU.
+                  Use the types CUBE or GPU together with TemplateID.
                 enum:
                 - ENTERPRISE
                 - VCPU
+                - CUBE
+                - GPU
                 type: string
                 x-kubernetes-validations:
                 - message: type is immutable
@@ -368,10 +388,17 @@ spec:
             - disk
             type: object
             x-kubernetes-validations:
-            - message: cpuFamily must not be specified when using VCPU
-              rule: self.type != 'VCPU' || !has(self.cpuFamily)
+            - message: cpuFamily must not be specified when using VCPU, CUBE or GPU
+              rule: (self.type != 'VCPU' && self.type != 'CUBE' && self.type != 'GPU')
+                || !has(self.cpuFamily)
             - message: networkID is required once set
               rule: '!has(oldSelf.networkID) || has(self.networkID)'
+            - message: templateID is required when type is CUBE or GPU
+              rule: (self.type != 'CUBE' && self.type != 'GPU') || self.templateID
+                != ''
+            - message: templateID must only be used with CUBE or GPU type
+              rule: (self.type == 'CUBE' || self.type == 'GPU') || !has(self.templateID)
+                || self.templateID == ''
           status:
             description: IonosCloudMachineStatus defines the observed state of IonosCloudMachine.
             properties:

config/crd/bases/infrastructure.cluster.x-k8s.io_ionoscloudmachinetemplates.yaml

@@ -184,6 +184,7 @@ spec:
 
                           If not specified, the cloud will select a suitable CPU family
                           based on the availability in the data center.
+                          Ignored when TemplateID is specified.
                         example: AMD_OPTERON
                         type: string
                       datacenterID:
@@ -209,7 +210,9 @@ spec:
                             type: string
                           diskType:
                             default: HDD
-                            description: DiskType defines the type of the hard drive.
+                            description: |-
+                              DiskType defines the type of the hard drive.
+                              Ignored when TemplateID of IonosCloudMachineSpec is specified (for the boot volume).
                             enum:
                             - HDD
                             - SSD Standard
@@ -264,8 +267,9 @@ spec:
                             type: string
                           sizeGB:
                             default: 20
-                            description: SizeGB defines the size of the volume in
-                              GB
+                            description: |-
+                              SizeGB defines the size of the volume in GB.
+                              Ignored when TemplateID of IonosCloudMachineSpec is specified (for the boot volume).
                             minimum: 10
                             type: integer
                         required:
@@ -349,6 +353,7 @@ spec:
                           MemoryMB is the memory size for the VM in MB.
                           Size must be specified in multiples of 256 MB with a minimum of 1024 MB
                           which is required as we are using hot-pluggable RAM by default.
+                          Ignored when TemplateID is specified.
                         format: int32
                         minimum: 2048
                         multipleOf: 1024
@@ -367,8 +372,9 @@ spec:
                           rule: self == oldSelf
                       numCores:
                         default: 1
-                        description: NumCores defines the number of cores for the
-                          VM.
+                        description: |-
+                          NumCores defines the number of cores for the VM.
+                          Ignored when TemplateID is specified.
                         format: int32
                         minimum: 1
                         type: integer
@@ -377,13 +383,25 @@ spec:
                           ProviderID is the IONOS Cloud provider ID
                           will be in the format ionos://ee090ff2-1eef-48ec-a246-a51a33aa4f3a
                         type: string
+                      templateID:
+                        description: |-
+                          TemplateID is the ID of the template for creating CUBE or GPU servers.
+                          If a template has GPU cards assigned, then it can only be used to create GPU servers,
+                          otherwise it can only be used for CUBE servers.
+                          NumCores, MemoryMB, CPUFamily and the size parameter of the first volume (boot Disk) are ignored when this is set.
+                          Available TemplateIDs can be listed using the cloud api.
+                        format: uuid
+                        type: string
                       type:
                         default: ENTERPRISE
-                        description: Type is the server type of the VM. Can be either
-                          ENTERPRISE or VCPU.
+                        description: |-
+                          Type is the server type of the VM. Can be either ENTERPRISE, VCPU, CUBE or GPU.
+                          Use the types CUBE or GPU together with TemplateID.
                         enum:
                         - ENTERPRISE
                         - VCPU
+                        - CUBE
+                        - GPU
                         type: string
                         x-kubernetes-validations:
                         - message: type is immutable
@@ -395,6 +413,12 @@ spec:
                     x-kubernetes-validations:
                     - message: networkID is required once set
                       rule: '!has(oldSelf.networkID) || has(self.networkID)'
+                    - message: templateID is required when type is CUBE or GPU
+                      rule: (self.type != 'CUBE' && self.type != 'GPU') || self.templateID
+                        != ''
+                    - message: templateID must only be used with CUBE or GPU type
+                      rule: (self.type == 'CUBE' || self.type == 'GPU') || !has(self.templateID)
+                        || self.templateID == ''
                 required:
                 - spec
                 type: object

docs/custom-image.md

@@ -32,9 +32,15 @@ export PACKER_FLAGS="--var 'kubernetes_rpm_version=1.28.3' --var 'kubernetes_sem
 
 Now you can build the image:
 ```sh
-make build-qemu-ubuntu-2204
+make build-qemu-ubuntu-2404
 ```
-In this case, for Ubuntu 22.04. Works also for `1804` or `2004`.   
+
+For efi support (required to use GPUs) build the image like this:
+```sh
+make build-qemu-ubuntu-2404-efi
+```
+
+In this case, for Ubuntu 24.04. Works also for `2204`.   
 The image creation takes quite some time, so be patient.   
 
 The image will be created in the `output` directory.
@@ -71,6 +77,20 @@ You can also do that via Cloud API:
 ionosctl img update --image-id <image-id> --cloudinit V1 --licence-type LINUX
 ```
 
+### Enabling efi support for your image
+
+Important: you **must** build your image with EFI support for this to work!   
+To enable UEFI functionality for your image, you need to make an adjustment in DCD:
+
+1. Go to Management -> Images & Snapshots -> select your image.
+2. Set UEFI Support to "Image is UEFI-compatible" and save your changes.
+
+You can also do that via Cloud API:
+
+```sh
+ionosctl img update --image-id <image-id> --require-legacy-bios=false
+```
+
 ### Enabling disk serial exposure
 
 The disk serial number is required for dynamic volume provisioning plugins like a CSI driver to function properly.

internal/service/cloud/server.go

@@ -246,7 +246,10 @@ func (s *Service) deleteServer(ctx context.Context, ms *scope.Machine, server *s
 
 	deleteVolumes := ms.ClusterScope.IsDeleted()
 	bootVolumeID := server.GetProperties().GetBootVolume().GetId()
-	if !deleteVolumes && bootVolumeID != nil {
+
+	// when using templates (cubes or gpu servers) we cannot delete the boot volume
+	// the whole server must be deleted at once
+	if !deleteVolumes && bootVolumeID != nil && (server.Properties != nil && server.Properties.TemplateUuid == nil) {
 		// We need to make sure to only delete volumes if the cluster is being deleted.
 		// If a node is being replaced, we only delete the boot volume and keep all other volumes.
 		// The CSI will take care of re-attaching the existing volumes to the new node.
@@ -381,6 +384,15 @@ func (*Service) buildServerProperties(
 		Type:             ptr.To(machineSpec.Type.String()),
 	}
 
+	if machineSpec.TemplateID != "" {
+		props.TemplateUuid = &machineSpec.TemplateID
+
+		// cores, ram and cpuFamily cannot be set when using a template
+		props.Cores = nil
+		props.Ram = nil
+		props.CpuFamily = nil
+	}
+
 	return props
 }
 
@@ -404,6 +416,12 @@ func (s *Service) buildServerEntities(ms *scope.Machine, params serverEntityPara
 		},
 	}
 
+	// size and type of the boot volume cannot be set when using a template
+	if machineSpec.TemplateID != "" {
+		bootVolume.Properties.Size = nil
+		bootVolume.Properties.Type = nil
+	}
+
 	if params.imageID != "" {
 		bootVolume.Properties.Image = &params.imageID
 	}