Add support for systemd-less systems

Add flag to skip all systemd operations

Signed-off-by: Alessandro Vinciguerra <alessandro.vinciguerra@postfinance.ch>

Author: Arc676

cmd/nvidia-mig-manager/main.go

@@ -81,6 +81,7 @@ var (
 	nvidiaCDIHookPath string
 
 	readonlyRootFS     bool
+	systemdUnavailable bool
 )
 
 type GPUClients struct {
@@ -273,6 +274,12 @@ func main() {
 			Destination: &readonlyRootFS,
 			EnvVars:     []string{"READONLY_ROOTFS"},
 		},
+		&cli.BoolFlag{
+			Name:        "systemd-unavailable",
+			Usage:       "Indicate that systemd is unavailable on the host",
+			Destination: &systemdUnavailable,
+			EnvVars:     []string{"SYSTEMD_UNAVAILABLE"},
+		},
 	}
 
 	err := c.Run(os.Args)
@@ -529,6 +536,7 @@ func migReconfigure(ctx context.Context, migConfigValue string, clientset *kuber
 		DefaultGPUClientsNamespace: defaultGPUClientsNamespaceFlag,
 		WithReboot:                 withRebootFlag,
 		WithShutdownHostGPUClients: withShutdownHostGPUClientsFlag,
+		SystemdAvailable:           !systemdUnavailable,
 		ReadonlyHost:               readonlyRootFS,
 	}
 

pkg/mig/reconfigure/reconfigure.go

@@ -76,6 +76,7 @@ type Options struct {
 	DriverLibraryPath          string
 	NvidiaSMIPath              string
 	NvidiaCDIHookPath          string
+	SystemdAvailable           bool
 	ReadonlyHost               bool
 }
 
@@ -106,9 +107,15 @@ func New(ctx context.Context, clientset *kubernetes.Clientset, migPartedBinary [
 		_ = os.Setenv("DBUS_SYSTEM_BUS_ADDRESS", hostSystemBusAddress)
 	}
 
-	systemdManager, err := systemd.NewManager(ctx)
-	if err != nil {
-		return nil, fmt.Errorf("failed to initialize systemd manager: %w", err)
+	var systemdManager *systemd.Manager = nil
+
+	if opts.SystemdAvailable {
+		mgr, err := systemd.NewManager(ctx)
+		if err != nil {
+			return nil, fmt.Errorf("failed to initialize systemd manager: %w", err)
+		}
+
+		systemdManager = mgr
 	}
 
 	return &Reconfigure{
@@ -300,11 +307,6 @@ func (r *Reconfigure) isConfigAlreadyApplied() bool {
 
 // persistConfigIfNeeded persists the configuration if needed
 func (r *Reconfigure) persistConfigIfNeeded() error {
-	if r.opts.ReadonlyHost {
-		log.Info("Host root FS is mounted read-only, config cannot be persisted")
-		return nil
-	}
-
 	if r.opts.HostRootMount == "" || r.opts.HostMigManagerStateFile == "" {
 		return nil
 	}
@@ -320,6 +322,11 @@ func (r *Reconfigure) persistConfigIfNeeded() error {
 
 // persistConfig persists the configuration to the state file
 func (r *Reconfigure) persistConfig() error {
+	if r.opts.ReadonlyHost {
+		log.Info("Host root FS is mounted read-only, config cannot be persisted")
+		return nil
+	}
+
 	config := fmt.Sprintf(`[Service]
 Environment="MIG_PARTED_SELECTED_CONFIG=%s"
 `, r.opts.SelectedMigConfig)
@@ -331,6 +338,10 @@ Environment="MIG_PARTED_SELECTED_CONFIG=%s"
 		return fmt.Errorf("failed to write config to state file: %w", err)
 	}
 
+	if r.systemdManager == nil {
+		return nil
+	}
+
 	return r.systemdManager.ReloadDaemon()
 }
 
@@ -420,6 +431,11 @@ func (r *Reconfigure) waitForPodsToBeDeleted() error {
 
 // shutdownHostGPUClients shuts down host GPU clients
 func (r *Reconfigure) shutdownHostGPUClients() error {
+	if r.systemdManager == nil {
+		log.Info("No systemd manager available, not shutting down GPU clients on the host")
+		return nil
+	}
+
 	log.Info("Shutting down all GPU clients on the host by stopping their systemd services")
 
 	services := strings.Split(r.opts.HostGPUClientServices, ",")
@@ -705,6 +721,11 @@ func (r *Reconfigure) createCDISpec() error {
 }
 
 func (r *Reconfigure) hostStartSystemdServices() error {
+	if r.systemdManager == nil {
+		log.Info("No systemd manager available, not starting anything via systemd")
+		return nil
+	}
+
 	services := r.stoppedServices
 	var restartServices []string
 	if len(services) == 0 {