Add support for systemd-less systems

Add timeout for systemd connection
Autodetect systemd availability

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

Author: Arc676

cmd/nvidia-mig-manager/main.go

@@ -58,6 +58,7 @@ const (
 	DefaultDriverRootCtrPath         = "/run/nvidia/driver"
 	DefaultNvidiaCDIHookPath         = "/usr/local/nvidia/toolkit/nvidia-cdi-hook"
 	DefaultGeneratedConfigFile       = "/etc/nvidia-mig-manager/generated-config.yaml"
+	DefaultSystemdTimeout            = 1.0
 )
 
 var (
@@ -80,6 +81,7 @@ var (
 	devRoot           string
 	devRootCtrPath    string
 	nvidiaCDIHookPath string
+	systemdTimeout    float64
 )
 
 type GPUClients struct {
@@ -266,6 +268,13 @@ func main() {
 			Destination: &nvidiaCDIHookPath,
 			Sources:     cli.EnvVars("NVIDIA_CDI_HOOK_PATH"),
 		},
+		&cli.Float64Flag{
+			Name:        "systemd-timeout",
+			Value:       DefaultSystemdTimeout,
+			Usage:       "Timeout for systemd connection (in seconds)",
+			Destination: &systemdTimeout,
+			Sources:     cli.EnvVars("SYSTEMD_CONNECTION_TIMEOUT"),
+		},
 	}
 
 	err := c.Run(context.Background(), os.Args)
@@ -522,6 +531,7 @@ func migReconfigure(ctx context.Context, migConfigValue string, clientset *kuber
 		DefaultGPUClientsNamespace: defaultGPUClientsNamespaceFlag,
 		WithReboot:                 withRebootFlag,
 		WithShutdownHostGPUClients: withShutdownHostGPUClientsFlag,
+		SystemdTimeout:             systemdTimeout,
 	}
 
 	if cdiEnabledFlag {

internal/systemd/systemd.go

@@ -35,10 +35,22 @@ type Manager struct {
 }
 
 // NewManager creates a new Manager instance
-func NewManager(ctx context.Context) (*Manager, error) {
-	conn, err := dbus.NewSystemConnectionContext(ctx)
+func NewManager(ctx context.Context, timeout time.Duration) (*Manager, error) {
+	deadline, cancel := context.WithTimeout(ctx, timeout)
+	defer cancel()
+
+	conn, err := dbus.NewSystemConnectionContext(deadline)
+	if err != nil {
+		err = fmt.Errorf("failed to connect to systemd D-Bus: %w", err)
+		if timeout := deadline.Err(); timeout == context.DeadlineExceeded {
+			err = errors.Join(err, timeout)
+		}
+		return nil, err
+	}
+
+	conn, err = dbus.NewSystemConnectionContext(ctx)
 	if err != nil {
-		return nil, fmt.Errorf("failed to connect to systemd D-Bus: %w", err)
+		return nil, fmt.Errorf("failed to recreate connection with no timeout")
 	}
 
 	return &Manager{

pkg/mig/reconfigure/reconfigure.go

@@ -78,6 +78,7 @@ type Options struct {
 	DriverLibraryPath          string
 	NvidiaSMIPath              string
 	NvidiaCDIHookPath          string
+	SystemdTimeout             float64
 }
 
 // Reconfigure handles the MIG reconfiguration process
@@ -107,9 +108,14 @@ func New(ctx context.Context, clientset *kubernetes.Clientset, migPartedBinary [
 		_ = os.Setenv("DBUS_SYSTEM_BUS_ADDRESS", hostSystemBusAddress)
 	}
 
-	systemdManager, err := systemd.NewManager(ctx)
+	systemdManager, err := systemd.NewManager(ctx, time.Duration(opts.SystemdTimeout)*time.Second)
 	if err != nil {
-		return nil, fmt.Errorf("failed to initialize systemd manager: %w", err)
+		if errors.Is(err, context.DeadlineExceeded) {
+			log.Warn("connection to systemd timed out; skipping systemd-related tasks for this reconfiguration")
+			systemdManager = nil
+		} else {
+			return nil, fmt.Errorf("failed to initialize systemd manager: %w", err)
+		}
 	}
 
 	return &Reconfigure{
@@ -331,6 +337,10 @@ 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 +430,11 @@ 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 +720,11 @@ 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 {