Accept kubelet socket from command line

cmd/nvidia-device-plugin/main.go

@@ -21,6 +21,7 @@ import (
 	"errors"
 	"fmt"
 	"os"
+	"path/filepath"
 	"syscall"
 	"time"
 
@@ -105,6 +106,12 @@ func main() {
 			Usage:   "ensure that containers are started with NVIDIA_MOFED=enabled",
 			EnvVars: []string{"MOFED_ENABLED"},
 		},
+		&cli.StringFlag{
+			Name:    "kubelet-socket",
+			Value:   pluginapi.KubeletSocket,
+			Usage:   "specify the socket for communicating with the kubelet; if this is empty, no connection with the kubelet is attempted",
+			EnvVars: []string{"KUBELET_SOCKET"},
+		},
 		&cli.StringFlag{
 			Name:        "config-file",
 			Usage:       "the path to a config file as an alternative to command line options or environment variables",
@@ -196,8 +203,10 @@ func loadConfig(c *cli.Context, flags []cli.Flag) (*spec.Config, error) {
 }
 
 func start(c *cli.Context, flags []cli.Flag) error {
-	klog.Info("Starting FS watcher.")
-	watcher, err := watch.Files(pluginapi.DevicePluginPath)
+	kubeletSocket := c.String("kubelet-socket")
+	kubeletSocketDir := filepath.Dir(kubeletSocket)
+	klog.Infof("Starting FS watcher for %v", kubeletSocketDir)
+	watcher, err := watch.Files(kubeletSocketDir)
 	if err != nil {
 		return fmt.Errorf("failed to create FS watcher for %s: %v", pluginapi.DevicePluginPath, err)
 	}
@@ -242,8 +251,8 @@ restart:
 		// 'pluginapi.KubeletSocket' file. When this occurs, restart this loop,
 		// restarting all of the plugins in the process.
 		case event := <-watcher.Events:
-			if event.Name == pluginapi.KubeletSocket && event.Op&fsnotify.Create == fsnotify.Create {
-				klog.Infof("inotify: %s created, restarting.", pluginapi.KubeletSocket)
+			if kubeletSocket != "" && event.Name == kubeletSocket && event.Op&fsnotify.Create == fsnotify.Create {
+				klog.Infof("inotify: %s created, restarting.", kubeletSocket)
 				goto restart
 			}
 
@@ -315,7 +324,8 @@ func startPlugins(c *cli.Context, flags []cli.Flag) ([]plugin.Interface, bool, e
 
 	// Get the set of plugins.
 	klog.Info("Retrieving plugins.")
-	pluginManager, err := NewPluginManager(infolib, nvmllib, devicelib, config)
+	kubeletSocket := c.String("kubelet-socket")
+	pluginManager, err := NewPluginManager(infolib, nvmllib, devicelib, kubeletSocket, config)
 	if err != nil {
 		return nil, false, fmt.Errorf("error creating plugin manager: %v", err)
 	}

cmd/nvidia-device-plugin/plugin-manager.go

@@ -29,7 +29,7 @@ import (
 )
 
 // NewPluginManager creates an NVML-based plugin manager
-func NewPluginManager(infolib info.Interface, nvmllib nvml.Interface, devicelib device.Interface, config *spec.Config) (manager.Interface, error) {
+func NewPluginManager(infolib info.Interface, nvmllib nvml.Interface, devicelib device.Interface, kubeletSocket string, config *spec.Config) (manager.Interface, error) {
 	var err error
 	switch *config.Flags.MigStrategy {
 	case spec.MigStrategyNone:
@@ -67,6 +67,7 @@ func NewPluginManager(infolib info.Interface, nvmllib nvml.Interface, devicelib
 		manager.WithCDIHandler(cdiHandler),
 		manager.WithConfig(config),
 		manager.WithFailOnInitError(*config.Flags.FailOnInitError),
+		manager.WithKubeletSocket(kubeletSocket),
 		manager.WithMigStrategy(*config.Flags.MigStrategy),
 	)
 	if err != nil {

internal/plugin/manager/factory.go

@@ -38,6 +38,8 @@ type manager struct {
 
 	cdiHandler cdi.Interface
 	config     *spec.Config
+
+	kubeletSocket string
 }
 
 // New creates a new plugin manager with the supplied options.

internal/plugin/manager/nvml.go

@@ -34,7 +34,7 @@ func (m *nvmlmanager) GetPlugins() ([]plugin.Interface, error) {
 
 	var plugins []plugin.Interface
 	for _, r := range rms {
-		plugin, err := plugin.NewNvidiaDevicePlugin(m.config, r, m.cdiHandler)
+		plugin, err := plugin.NewNvidiaDevicePlugin(m.config, m.kubeletSocket, r, m.cdiHandler)
 		if err != nil {
 			return nil, fmt.Errorf("failed to create plugin: %w", err)
 		}

internal/plugin/manager/options.go

@@ -68,3 +68,9 @@ func WithConfig(config *spec.Config) Option {
 		m.config = config
 	}
 }
+
+func WithKubeletSocket(kubeletSocket string) Option {
+	return func(m *manager) {
+		m.kubeletSocket = kubeletSocket
+	}
+}

internal/plugin/manager/tegra.go

@@ -34,7 +34,7 @@ func (m *tegramanager) GetPlugins() ([]plugin.Interface, error) {
 
 	var plugins []plugin.Interface
 	for _, r := range rms {
-		plugin, err := plugin.NewNvidiaDevicePlugin(m.config, r, m.cdiHandler)
+		plugin, err := plugin.NewNvidiaDevicePlugin(m.config, m.kubeletSocket, r, m.cdiHandler)
 		if err != nil {
 			return nil, fmt.Errorf("failed to create plugin: %w", err)
 		}

internal/plugin/server.go

@@ -55,6 +55,7 @@ type NvidiaDevicePlugin struct {
 	deviceListEnvvar     string
 	deviceListStrategies spec.DeviceListStrategies
 	socket               string
+	kubeletSocket        string
 
 	cdiHandler          cdi.Interface
 	cdiEnabled          bool
@@ -69,7 +70,7 @@ type NvidiaDevicePlugin struct {
 }
 
 // NewNvidiaDevicePlugin returns an initialized NvidiaDevicePlugin
-func NewNvidiaDevicePlugin(config *spec.Config, resourceManager rm.ResourceManager, cdiHandler cdi.Interface) (*NvidiaDevicePlugin, error) {
+func NewNvidiaDevicePlugin(config *spec.Config, kubeletSocket string, resourceManager rm.ResourceManager, cdiHandler cdi.Interface) (*NvidiaDevicePlugin, error) {
 	_, name := resourceManager.Resource().Split()
 
 	deviceListStrategies, _ := spec.NewDeviceListStrategies(*config.Flags.Plugin.DeviceListStrategy)
@@ -102,6 +103,7 @@ func NewNvidiaDevicePlugin(config *spec.Config, resourceManager rm.ResourceManag
 		mpsDaemon:   mpsDaemon,
 		mpsHostRoot: mpsHostRoot,
 
+		kubeletSocket: kubeletSocket,
 		// These will be reinitialized every
 		// time the plugin server is restarted.
 		server: nil,
@@ -245,7 +247,12 @@ func (plugin *NvidiaDevicePlugin) Serve() error {
 
 // Register registers the device plugin for the given resourceName with Kubelet.
 func (plugin *NvidiaDevicePlugin) Register() error {
-	conn, err := plugin.dial(pluginapi.KubeletSocket, 5*time.Second)
+	if plugin.kubeletSocket == "" {
+		klog.Info("Skipping registration with Kubelet")
+		return nil
+	}
+
+	conn, err := plugin.dial(plugin.kubeletSocket, 5*time.Second)
 	if err != nil {
 		return err
 	}