Add nvidia configuration to the api

k8s/crds/kops.k8s.io_clusters.yaml

@@ -4238,6 +4238,19 @@ spec:
                       to false.
                     type: boolean
                 type: object
+              nvidia:
+                description: NvidiaConfiguration configures the Nvidia GPU runtime.
+                properties:
+                  enabled:
+                    description: Enabled determines if kOps will install the Nvidia
+                      GPU runtime and drivers. They will only be installed on intances
+                      that has an Nvidia GPU.
+                    type: boolean
+                  package:
+                    description: Package is the name of the nvidia driver package
+                      that will be installed. Default is "nvidia-headless-460-server".
+                    type: string
+                type: object
               podCIDR:
                 description: PodCIDR is the CIDR from which we allocate IPs for pods
                 type: string

pkg/apis/kops/cluster.go

@@ -208,12 +208,21 @@ type ClusterSpec struct {
 	ClusterAutoscaler *ClusterAutoscalerConfig `json:"clusterAutoscaler,omitempty"`
 	// WarmPool defines the default warm pool settings for instance groups (AWS only).
 	WarmPool *WarmPoolSpec `json:"warmPool,omitempty"`
-
 	// ServiceAccountIssuerDiscovery configures the OIDC Issuer for ServiceAccounts.
 	ServiceAccountIssuerDiscovery *ServiceAccountIssuerDiscoveryConfig `json:"serviceAccountIssuerDiscovery,omitempty"`
-
 	// SnapshotController defines the CSI Snapshot Controller configuration.
 	SnapshotController *SnapshotControllerConfig `json:"snapshotController,omitempty"`
+	// NvidiaConfiguration configures the Nvidia GPU runtime.
+	Nvidia *NvidiaConfig `json:"nvidia,omitempty"`
+}
+
+type NvidiaConfig struct {
+	// Package is the name of the nvidia driver package that will be installed.
+	// Default is "nvidia-headless-460-server".
+	DriverPackage string `json:"package,omitempty"`
+	// Enabled determines if kOps will install the Nvidia GPU runtime and drivers.
+	// They will only be installed on intances that has an Nvidia GPU.
+	Enabled *bool `json:"enabled,omitempty"`
 }
 
 // ServiceAccountIssuerDiscoveryConfig configures an OIDC Issuer.

pkg/apis/kops/v1alpha2/cluster.go

@@ -206,12 +206,21 @@ type ClusterSpec struct {
 	ClusterAutoscaler *ClusterAutoscalerConfig `json:"clusterAutoscaler,omitempty"`
 	// WarmPool defines the default warm pool settings for instance groups (AWS only).
 	WarmPool *WarmPoolSpec `json:"warmPool,omitempty"`
-
 	// ServiceAccountIssuerDiscovery configures the OIDC Issuer for ServiceAccounts.
 	ServiceAccountIssuerDiscovery *ServiceAccountIssuerDiscoveryConfig `json:"serviceAccountIssuerDiscovery,omitempty"`
-
 	// SnapshotController defines the CSI Snapshot Controller configuration.
 	SnapshotController *SnapshotControllerConfig `json:"snapshotController,omitempty"`
+	// NvidiaConfiguration configures the Nvidia GPU runtime.
+	Nvidia *NvidiaConfig `json:"nvidia,omitempty"`
+}
+
+type NvidiaConfig struct {
+	// Package is the name of the nvidia driver package that will be installed.
+	// Default is "nvidia-headless-460-server".
+	DriverPackage string `json:"package,omitempty"`
+	// Enabled determines if kOps will install the Nvidia GPU runtime and drivers.
+	// They will only be installed on intances that has an Nvidia GPU.
+	Enabled *bool `json:"enabled,omitempty"`
 }
 
 // ServiceAccountIssuerDiscoveryConfig configures an OIDC Issuer.

pkg/apis/kops/v1alpha2/zz_generated.conversion.go

@@ -883,6 +883,16 @@ func RegisterConversions(s *runtime.Scheme) error {
 	}); err != nil {
 		return err
 	}
+	if err := s.AddGeneratedConversionFunc((*NvidiaConfig)(nil), (*kops.NvidiaConfig)(nil), func(a, b interface{}, scope conversion.Scope) error {
+		return Convert_v1alpha2_NvidiaConfig_To_kops_NvidiaConfig(a.(*NvidiaConfig), b.(*kops.NvidiaConfig), scope)
+	}); err != nil {
+		return err
+	}
+	if err := s.AddGeneratedConversionFunc((*kops.NvidiaConfig)(nil), (*NvidiaConfig)(nil), func(a, b interface{}, scope conversion.Scope) error {
+		return Convert_kops_NvidiaConfig_To_v1alpha2_NvidiaConfig(a.(*kops.NvidiaConfig), b.(*NvidiaConfig), scope)
+	}); err != nil {
+		return err
+	}
 	if err := s.AddGeneratedConversionFunc((*OpenstackBlockStorageConfig)(nil), (*kops.OpenstackBlockStorageConfig)(nil), func(a, b interface{}, scope conversion.Scope) error {
 		return Convert_v1alpha2_OpenstackBlockStorageConfig_To_kops_OpenstackBlockStorageConfig(a.(*OpenstackBlockStorageConfig), b.(*kops.OpenstackBlockStorageConfig), scope)
 	}); err != nil {
@@ -2690,6 +2700,15 @@ func autoConvert_v1alpha2_ClusterSpec_To_kops_ClusterSpec(in *ClusterSpec, out *
 	} else {
 		out.SnapshotController = nil
 	}
+	if in.Nvidia != nil {
+		in, out := &in.Nvidia, &out.Nvidia
+		*out = new(kops.NvidiaConfig)
+		if err := Convert_v1alpha2_NvidiaConfig_To_kops_NvidiaConfig(*in, *out, s); err != nil {
+			return err
+		}
+	} else {
+		out.Nvidia = nil
+	}
 	return nil
 }
 
@@ -3102,6 +3121,15 @@ func autoConvert_kops_ClusterSpec_To_v1alpha2_ClusterSpec(in *kops.ClusterSpec,
 	} else {
 		out.SnapshotController = nil
 	}
+	if in.Nvidia != nil {
+		in, out := &in.Nvidia, &out.Nvidia
+		*out = new(NvidiaConfig)
+		if err := Convert_kops_NvidiaConfig_To_v1alpha2_NvidiaConfig(*in, *out, s); err != nil {
+			return err
+		}
+	} else {
+		out.Nvidia = nil
+	}
 	return nil
 }
 
@@ -6122,6 +6150,28 @@ func Convert_kops_NodeTerminationHandlerConfig_To_v1alpha2_NodeTerminationHandle
 	return autoConvert_kops_NodeTerminationHandlerConfig_To_v1alpha2_NodeTerminationHandlerConfig(in, out, s)
 }
 
+func autoConvert_v1alpha2_NvidiaConfig_To_kops_NvidiaConfig(in *NvidiaConfig, out *kops.NvidiaConfig, s conversion.Scope) error {
+	out.DriverPackage = in.DriverPackage
+	out.Enabled = in.Enabled
+	return nil
+}
+
+// Convert_v1alpha2_NvidiaConfig_To_kops_NvidiaConfig is an autogenerated conversion function.
+func Convert_v1alpha2_NvidiaConfig_To_kops_NvidiaConfig(in *NvidiaConfig, out *kops.NvidiaConfig, s conversion.Scope) error {
+	return autoConvert_v1alpha2_NvidiaConfig_To_kops_NvidiaConfig(in, out, s)
+}
+
+func autoConvert_kops_NvidiaConfig_To_v1alpha2_NvidiaConfig(in *kops.NvidiaConfig, out *NvidiaConfig, s conversion.Scope) error {
+	out.DriverPackage = in.DriverPackage
+	out.Enabled = in.Enabled
+	return nil
+}
+
+// Convert_kops_NvidiaConfig_To_v1alpha2_NvidiaConfig is an autogenerated conversion function.
+func Convert_kops_NvidiaConfig_To_v1alpha2_NvidiaConfig(in *kops.NvidiaConfig, out *NvidiaConfig, s conversion.Scope) error {
+	return autoConvert_kops_NvidiaConfig_To_v1alpha2_NvidiaConfig(in, out, s)
+}
+
 func autoConvert_v1alpha2_OpenstackBlockStorageConfig_To_kops_OpenstackBlockStorageConfig(in *OpenstackBlockStorageConfig, out *kops.OpenstackBlockStorageConfig, s conversion.Scope) error {
 	out.Version = in.Version
 	out.IgnoreAZ = in.IgnoreAZ

pkg/apis/kops/v1alpha2/zz_generated.deepcopy.go

@@ -1185,6 +1185,11 @@ func (in *ClusterSpec) DeepCopyInto(out *ClusterSpec) {
 		*out = new(SnapshotControllerConfig)
 		(*in).DeepCopyInto(*out)
 	}
+	if in.Nvidia != nil {
+		in, out := &in.Nvidia, &out.Nvidia
+		*out = new(NvidiaConfig)
+		(*in).DeepCopyInto(*out)
+	}
 	return
 }
 
@@ -4131,6 +4136,27 @@ func (in *NodeTerminationHandlerConfig) DeepCopy() *NodeTerminationHandlerConfig
 	return out
 }
 
+// DeepCopyInto is an autogenerated deepcopy function, copying the receiver, writing into out. in must be non-nil.
+func (in *NvidiaConfig) DeepCopyInto(out *NvidiaConfig) {
+	*out = *in
+	if in.Enabled != nil {
+		in, out := &in.Enabled, &out.Enabled
+		*out = new(bool)
+		**out = **in
+	}
+	return
+}
+
+// DeepCopy is an autogenerated deepcopy function, copying the receiver, creating a new NvidiaConfig.
+func (in *NvidiaConfig) DeepCopy() *NvidiaConfig {
+	if in == nil {
+		return nil
+	}
+	out := new(NvidiaConfig)
+	in.DeepCopyInto(out)
+	return out
+}
+
 // DeepCopyInto is an autogenerated deepcopy function, copying the receiver, writing into out. in must be non-nil.
 func (in *OpenstackBlockStorageConfig) DeepCopyInto(out *OpenstackBlockStorageConfig) {
 	*out = *in

pkg/apis/kops/zz_generated.deepcopy.go

@@ -1269,6 +1269,11 @@ func (in *ClusterSpec) DeepCopyInto(out *ClusterSpec) {
 		*out = new(SnapshotControllerConfig)
 		(*in).DeepCopyInto(*out)
 	}
+	if in.Nvidia != nil {
+		in, out := &in.Nvidia, &out.Nvidia
+		*out = new(NvidiaConfig)
+		(*in).DeepCopyInto(*out)
+	}
 	return
 }
 
@@ -4313,6 +4318,27 @@ func (in *NodeTerminationHandlerConfig) DeepCopy() *NodeTerminationHandlerConfig
 	return out
 }
 
+// DeepCopyInto is an autogenerated deepcopy function, copying the receiver, writing into out. in must be non-nil.
+func (in *NvidiaConfig) DeepCopyInto(out *NvidiaConfig) {
+	*out = *in
+	if in.Enabled != nil {
+		in, out := &in.Enabled, &out.Enabled
+		*out = new(bool)
+		**out = **in
+	}
+	return
+}
+
+// DeepCopy is an autogenerated deepcopy function, copying the receiver, creating a new NvidiaConfig.
+func (in *NvidiaConfig) DeepCopy() *NvidiaConfig {
+	if in == nil {
+		return nil
+	}
+	out := new(NvidiaConfig)
+	in.DeepCopyInto(out)
+	return out
+}
+
 // DeepCopyInto is an autogenerated deepcopy function, copying the receiver, writing into out. in must be non-nil.
 func (in *OpenstackBlockStorageConfig) DeepCopyInto(out *OpenstackBlockStorageConfig) {
 	*out = *in

pkg/apis/nodeup/config.go

@@ -76,6 +76,8 @@ type Config struct {
 
 	// APIServerConfig is additional configuration for nodes running an APIServer.
 	APIServerConfig *APIServerConfig `json:",omitempty"`
+	// Nvidia contains the configuration for nvidia
+	Nvidia *kops.NvidiaConfig `json:",omitempty"`
 }
 
 // BootConfig is the configuration for the nodeup binary that might be too big to fit in userdata.