Automate installing NVIDIA Container Toolkit w/ flag

cmd/minikube/cmd/start.go

@@ -1285,7 +1285,7 @@ func validateFlags(cmd *cobra.Command, drvName string) {
 	}
 
 	if cmd.Flags().Changed(containerRuntime) {
-		err := validateRuntime(viper.GetString(containerRuntime), drvName)
+		err := validateRuntime(viper.GetString(containerRuntime))
 		if err != nil {
 			exit.Message(reason.Usage, "{{.err}}", out.V{"err": err})
 		}
@@ -1298,6 +1298,12 @@ func validateFlags(cmd *cobra.Command, drvName string) {
 		}
 	}
 
+	if cmd.Flags().Changed(enableNvidiaGPUs) {
+		if err := validateEnableNvidiaGPUs(viper.GetBool(enableNvidiaGPUs), drvName, viper.GetString(containerRuntime)); err != nil {
+			exit.Message(reason.Usage, "{{.err}}", out.V{"err": err})
+		}
+	}
+
 	if driver.IsSSH(drvName) {
 		sshIPAddress := viper.GetString(sshIPAddress)
 		if sshIPAddress == "" {
@@ -1402,7 +1408,7 @@ func validateDiskSize(diskSize string) error {
 }
 
 // validateRuntime validates the supplied runtime
-func validateRuntime(rtime, driverName string) error {
+func validateRuntime(rtime string) error {
 	validOptions := cruntime.ValidRuntimes()
 	// `crio` is accepted as an alternative spelling to `cri-o`
 	validOptions = append(validOptions, constants.CRIO)
@@ -1431,12 +1437,18 @@ func validateRuntime(rtime, driverName string) error {
 	if !validRuntime {
 		return errors.Errorf("Invalid Container Runtime: %s. Valid runtimes are: %s", rtime, cruntime.ValidRuntimes())
 	}
+	return nil
+}
 
-	if rtime == constants.NvidiaDocker && driverName != constants.Docker {
-		return errors.Errorf("The nvidia-docker container-runtime can only be run with the docker driver")
+// validateEnableNvidiaGPUs validates that the nvidia GPU(s) can be used with the given configuration
+func validateEnableNvidiaGPUs(gpusEnabled bool, drvName, rtime string) error {
+	if !gpusEnabled {
+		return nil
 	}
-
-	return nil
+	if drvName == constants.Docker && rtime == constants.Docker {
+		return nil
+	}
+	return errors.Errorf("The enable-nvidia-gpus flag can only be run with the docker driver and docker container-runtime")
 }
 
 func getContainerRuntime(old *config.ClusterConfig) string {
@@ -1798,7 +1810,7 @@ func validateContainerRuntime(old *config.ClusterConfig) {
 		return
 	}
 
-	if err := validateRuntime(old.KubernetesConfig.ContainerRuntime, old.Driver); err != nil {
+	if err := validateRuntime(old.KubernetesConfig.ContainerRuntime); err != nil {
 		klog.Errorf("Error parsing old runtime %q: %v", old.KubernetesConfig.ContainerRuntime, err)
 	}
 }

cmd/minikube/cmd/start_flags.go

@@ -142,6 +142,7 @@ const (
 	socketVMnetPath         = "socket-vmnet-path"
 	staticIP                = "static-ip"
 	autoPauseInterval       = "auto-pause-interval"
+	enableNvidiaGPUs        = "enable-nvidia-gpus"
 )
 
 var (
@@ -204,6 +205,7 @@ func initMinikubeFlags() {
 	startCmd.Flags().Bool(disableMetrics, false, "If set, disables metrics reporting (CPU and memory usage), this can improve CPU usage. Defaults to false.")
 	startCmd.Flags().String(staticIP, "", "Set a static IP for the minikube cluster, the IP must be: private, IPv4, and the last octet must be between 2 and 254, for example 192.168.200.200 (Docker and Podman drivers only)")
 	startCmd.Flags().Duration(autoPauseInterval, time.Minute*1, "Duration of inactivity before the minikube VM is paused (default 1m0s).  To disable, set to 0s")
+	startCmd.Flags().Bool(enableNvidiaGPUs, false, "If set, allows pods to use your NVIDIA GPU(s) (Docker driver with Docker container-runtime only)")
 }
 
 // initKubernetesFlags inits the commandline flags for Kubernetes related options
@@ -595,6 +597,7 @@ func generateNewConfigFromFlags(cmd *cobra.Command, k8sVersion string, rtime str
 		},
 		MultiNodeRequested: viper.GetInt(nodes) > 1,
 		AutoPauseInterval:  viper.GetDuration(autoPauseInterval),
+		EnableNvidiaGPUs:   viper.GetBool(enableNvidiaGPUs),
 	}
 	cc.VerifyComponents = interpretWaitFlag(*cmd)
 	if viper.GetBool(createMount) && driver.IsKIC(drvName) {

cmd/minikube/cmd/start_test.go

@@ -434,7 +434,6 @@ func TestValidateDiskSize(t *testing.T) {
 func TestValidateRuntime(t *testing.T) {
 	var tests = []struct {
 		runtime  string
-		driver   string
 		errorMsg string
 	}{
 		{
@@ -449,20 +448,10 @@ func TestValidateRuntime(t *testing.T) {
 			runtime:  "test",
 			errorMsg: fmt.Sprintf("Invalid Container Runtime: test. Valid runtimes are: %v", cruntime.ValidRuntimes()),
 		},
-		{
-			runtime:  "nvidia-docker",
-			driver:   "docker",
-			errorMsg: "",
-		},
-		{
-			runtime:  "nvidia-docker",
-			driver:   "kvm",
-			errorMsg: "The nvidia-docker container-runtime can only be run with the docker driver",
-		},
 	}
 	for _, test := range tests {
 		t.Run(test.runtime, func(t *testing.T) {
-			got := validateRuntime(test.runtime, test.driver)
+			got := validateRuntime(test.runtime)
 			gotError := ""
 			if got != nil {
 				gotError = got.Error()

pkg/drivers/kic/kic.go

@@ -77,22 +77,20 @@ func NewDriver(c Config) *Driver {
 func (d *Driver) Create() error {
 	ctx := context.Background()
 	params := oci.CreateParams{
-		Mounts:        d.NodeConfig.Mounts,
-		Name:          d.NodeConfig.MachineName,
-		Image:         d.NodeConfig.ImageDigest,
-		ClusterLabel:  oci.ProfileLabelKey + "=" + d.MachineName,
-		NodeLabel:     oci.NodeLabelKey + "=" + d.NodeConfig.MachineName,
-		CPUs:          strconv.Itoa(d.NodeConfig.CPU),
-		Memory:        strconv.Itoa(d.NodeConfig.Memory) + "mb",
-		Envs:          d.NodeConfig.Envs,
-		ExtraArgs:     append([]string{"--expose", fmt.Sprintf("%d", d.NodeConfig.APIServerPort)}, d.NodeConfig.ExtraArgs...),
-		OCIBinary:     d.NodeConfig.OCIBinary,
-		APIServerPort: d.NodeConfig.APIServerPort,
-	}
-
-	if d.NodeConfig.ContainerRuntime == constants.NvidiaDocker {
-		params.GPUs = true
+		Mounts:           d.NodeConfig.Mounts,
+		Name:             d.NodeConfig.MachineName,
+		Image:            d.NodeConfig.ImageDigest,
+		ClusterLabel:     oci.ProfileLabelKey + "=" + d.MachineName,
+		NodeLabel:        oci.NodeLabelKey + "=" + d.NodeConfig.MachineName,
+		CPUs:             strconv.Itoa(d.NodeConfig.CPU),
+		Memory:           strconv.Itoa(d.NodeConfig.Memory) + "mb",
+		Envs:             d.NodeConfig.Envs,
+		ExtraArgs:        append([]string{"--expose", fmt.Sprintf("%d", d.NodeConfig.APIServerPort)}, d.NodeConfig.ExtraArgs...),
+		OCIBinary:        d.NodeConfig.OCIBinary,
+		APIServerPort:    d.NodeConfig.APIServerPort,
+		EnableNvidiaGPUs: d.NodeConfig.EnableNvidiaGPUs,
 	}
+
 	networkName := d.NodeConfig.Network
 	if networkName == "" {
 		networkName = d.NodeConfig.ClusterName
@@ -455,7 +453,7 @@ func (d *Driver) Stop() error {
 		}
 	}
 
-	runtime, err := cruntime.New(cruntime.Config{Type: d.NodeConfig.ContainerRuntime, Runner: d.exec})
+	runtime, err := cruntime.New(cruntime.Config{Type: d.NodeConfig.ContainerRuntime, Runner: d.exec, EnableNvidiaGPUs: d.NodeConfig.EnableNvidiaGPUs})
 	if err != nil { // won't return error because:
 		// even though we can't stop the cotainers inside, we still wanna stop the minikube container itself
 		klog.Errorf("unable to get container runtime: %v", err)

pkg/drivers/kic/oci/oci.go

@@ -190,7 +190,7 @@ func CreateContainerNode(p CreateParams) error {
 		runArgs = append(runArgs, "--network", p.Network)
 		runArgs = append(runArgs, "--ip", p.IP)
 	}
-	if p.GPUs {
+	if p.EnableNvidiaGPUs {
 		runArgs = append(runArgs, "--gpus", "all")
 	}
 

pkg/drivers/kic/oci/types.go

@@ -43,23 +43,23 @@ const (
 
 // CreateParams are parameters needed to create a container
 type CreateParams struct {
-	ClusterName   string            // cluster(profile name) that this container belongs to
-	Name          string            // used for container name and hostname
-	Image         string            // container image to use to create the node.
-	ClusterLabel  string            // label the clusters we create using minikube so we can clean up
-	NodeLabel     string            // label the nodes so we can clean up by node name
-	Role          string            // currently only role supported is control-plane
-	Mounts        []Mount           // volume mounts
-	APIServerPort int               // Kubernetes api server port
-	PortMappings  []PortMapping     // ports to map to container from host
-	CPUs          string            // number of cpu cores assign to container
-	Memory        string            // memory (mbs) to assign to the container
-	Envs          map[string]string // environment variables to pass to the container
-	ExtraArgs     []string          // a list of any extra option to pass to oci binary during creation time, for example --expose 8080...
-	OCIBinary     string            // docker or podman
-	Network       string            // network name that the container will attach to
-	IP            string            // static IP to assign the container in the cluster network
-	GPUs          bool              // add GPU devices to the container
+	ClusterName      string            // cluster(profile name) that this container belongs to
+	Name             string            // used for container name and hostname
+	Image            string            // container image to use to create the node.
+	ClusterLabel     string            // label the clusters we create using minikube so we can clean up
+	NodeLabel        string            // label the nodes so we can clean up by node name
+	Role             string            // currently only role supported is control-plane
+	Mounts           []Mount           // volume mounts
+	APIServerPort    int               // Kubernetes api server port
+	PortMappings     []PortMapping     // ports to map to container from host
+	CPUs             string            // number of cpu cores assign to container
+	Memory           string            // memory (mbs) to assign to the container
+	Envs             map[string]string // environment variables to pass to the container
+	ExtraArgs        []string          // a list of any extra option to pass to oci binary during creation time, for example --expose 8080...
+	OCIBinary        string            // docker or podman
+	Network          string            // network name that the container will attach to
+	IP               string            // static IP to assign the container in the cluster network
+	EnableNvidiaGPUs bool              // add NVIDIA GPU devices to the container
 }
 
 // createOpt is an option for Create

pkg/drivers/kic/types.go

@@ -69,4 +69,5 @@ type Config struct {
 	StaticIP          string            // static IP for the kic cluster
 	ExtraArgs         []string          // a list of any extra option to pass to oci binary during creation time, for example --expose 8080...
 	ListenAddress     string            // IP Address to listen to
+	EnableNvidiaGPUs  bool              // add NVIDIA GPU devices to the container
 }

pkg/minikube/config/types.go

@@ -108,6 +108,7 @@ type ClusterConfig struct {
 	SSHAuthSock             string
 	SSHAgentPID             int
 	AutoPauseInterval       time.Duration // Specifies interval of time to wait before checking if cluster should be paused
+	EnableNvidiaGPUs        bool
 }
 
 // KubernetesConfig contains the parameters used to configure the VM Kubernetes.

pkg/minikube/constants/constants.go

@@ -63,8 +63,6 @@ const (
 	CRIO = "crio"
 	// Docker is the default name and spelling for the docker container runtime
 	Docker = "docker"
-	// NvidiaDocker is the default name and spelling for the nvidia-docker container runtime
-	NvidiaDocker = "nvidia-docker"
 	// DefaultContainerRuntime is our default container runtime
 	DefaultContainerRuntime = ""
 

pkg/minikube/cruntime/cruntime.go

@@ -50,7 +50,7 @@ func (cs ContainerState) String() string {
 
 // ValidRuntimes lists the supported container runtimes
 func ValidRuntimes() []string {
-	return []string{"docker", "nvidia-docker", "cri-o", "containerd"}
+	return []string{"docker", "cri-o", "containerd"}
 }
 
 // CommandRunner is the subset of command.Runner this package consumes
@@ -155,6 +155,8 @@ type Config struct {
 	KubernetesVersion semver.Version
 	// InsecureRegistry list of insecure registries
 	InsecureRegistry []string
+	// EnableNvidiaGPUs add GPU devices to the container
+	EnableNvidiaGPUs bool
 }
 
 // ListContainersOptions are the options to use for listing containers
@@ -210,7 +212,7 @@ func New(c Config) (Manager, error) {
 	sm := sysinit.New(c.Runner)
 
 	switch c.Type {
-	case "", "docker", "nvidia-docker":
+	case "", "docker":
 		sp := c.Socket
 		cs := ""
 		// There is no more dockershim socket, in Kubernetes version 1.24 and beyond
@@ -219,7 +221,6 @@ func New(c Config) (Manager, error) {
 			cs = "cri-docker.socket"
 		}
 		return &Docker{
-			Type:              c.Type,
 			Socket:            sp,
 			Runner:            c.Runner,
 			NetworkPlugin:     c.NetworkPlugin,
@@ -228,6 +229,7 @@ func New(c Config) (Manager, error) {
 			Init:              sm,
 			UseCRI:            (sp != ""), // !dockershim
 			CRIService:        cs,
+			NvidiaGPUs:        c.EnableNvidiaGPUs,
 		}, nil
 	case "crio", "cri-o":
 		return &CRIO{

pkg/minikube/cruntime/cruntime_test.go

@@ -40,7 +40,6 @@ func TestName(t *testing.T) {
 	}{
 		{"", "Docker"},
 		{"docker", "Docker"},
-		{"nvidia-docker", "Docker"},
 		{"crio", "CRI-O"},
 		{"cri-o", "CRI-O"},
 		{"containerd", "containerd"},
@@ -125,7 +124,6 @@ func TestCGroupDriver(t *testing.T) {
 		want    string
 	}{
 		{"docker", "cgroupfs"},
-		{"nvidia-docker", "cgroupfs"},
 		{"crio", "cgroupfs"},
 		{"containerd", "cgroupfs"},
 	}
@@ -157,12 +155,6 @@ func TestKubeletOptions(t *testing.T) {
 		{"docker", "1.24.0", map[string]string{
 			"container-runtime-endpoint": "unix:///var/run/cri-dockerd.sock",
 		}},
-		{"nvidia-docker", "1.23.0", map[string]string{
-			"container-runtime": "docker",
-		}},
-		{"nvidia-docker", "1.25.0", map[string]string{
-			"container-runtime-endpoint": "unix:///var/run/cri-dockerd.sock",
-		}},
 		{"crio", "1.25.0", map[string]string{
 			"container-runtime-endpoint": "unix:///var/run/crio/crio.sock",
 		}},
@@ -688,13 +680,6 @@ func TestEnable(t *testing.T) {
 				"crio":          SvcExited,
 				"crio-shutdown": SvcExited,
 			}},
-		{"nvidia-docker", defaultServices,
-			map[string]serviceState{
-				"docker":        SvcRestarted,
-				"containerd":    SvcExited,
-				"crio":          SvcExited,
-				"crio-shutdown": SvcExited,
-			}},
 		{"containerd", defaultServices,
 			map[string]serviceState{
 				"docker":        SvcExited,
@@ -736,7 +721,6 @@ func TestContainerFunctions(t *testing.T) {
 		runtime string
 	}{
 		{"docker"},
-		{"nvidia-docker"},
 		{"crio"},
 		{"containerd"},
 	}
@@ -746,7 +730,7 @@ func TestContainerFunctions(t *testing.T) {
 		t.Run(tc.runtime, func(t *testing.T) {
 			runner := NewFakeRunner(t)
 			prefix := ""
-			if tc.runtime == "docker" || tc.runtime == "nvidia-docker" {
+			if tc.runtime == "docker" {
 				prefix = "k8s_"
 			}
 			runner.containers = map[string]string{

pkg/minikube/cruntime/docker.go

@@ -68,7 +68,6 @@ func (e *ErrISOFeature) Error() string {
 
 // Docker contains Docker runtime state
 type Docker struct {
-	Type              string
 	Socket            string
 	Runner            CommandRunner
 	NetworkPlugin     string
@@ -77,6 +76,7 @@ type Docker struct {
 	Init              sysinit.Manager
 	UseCRI            bool
 	CRIService        string
+	NvidiaGPUs        bool
 }
 
 // Name is a human readable name for Docker
@@ -561,7 +561,7 @@ func (r *Docker) configureDocker(driver string) error {
 		},
 		StorageDriver: "overlay2",
 	}
-	if r.Type == constants.NvidiaDocker {
+	if r.NvidiaGPUs {
 		if err := r.installNvidiaContainerToolkit(); err != nil {
 			return fmt.Errorf("failed installing the NVIDIA Container Toolkit: %v", err)
 		}

pkg/minikube/registry/drvs/docker/docker.go

@@ -90,6 +90,7 @@ func configure(cc config.ClusterConfig, n config.Node) (interface{}, error) {
 		Subnet:            cc.Subnet,
 		StaticIP:          cc.StaticIP,
 		ListenAddress:     cc.ListenAddress,
+		EnableNvidiaGPUs:  cc.EnableNvidiaGPUs,
 	}), nil
 }
 

site/content/en/docs/tutorials/nvidia.md

@@ -34,7 +34,7 @@ date: 2018-01-02
   ```
 - Start minikube:
   ```shell
-  minikube start --driver docker --container-runtime nvidia-docker
+  minikube start --driver docker --container-runtime docker --enable-nvidia-gpus
   ```
 {{% /tab %}}
 {{% tab none %}}