fix: abort restore when node connection fails

shim/restore.go

@@ -30,6 +30,7 @@ import (
 var (
 	ErrAlreadyRestored      = errors.New("container is already restored")
 	ErrRestoreRequestFailed = errors.New("restore request failed")
+	ErrRestoreDial          = errors.New("failed to connect to node socket")
 )
 
 func (c *Container) Restore(ctx context.Context) (*runc.Container, process.Process, error) {
@@ -199,7 +200,7 @@ func createContainerLoggers(ctx context.Context, logPath string, tty bool) (stdo
 func MigrationRestore(ctx context.Context, r *task.CreateTaskRequest, cfg *Config) (skipStart bool, err error) {
 	conn, err := net.Dial("unix", nodev1.SocketPath)
 	if err != nil {
-		return false, fmt.Errorf("dialing node service: %w", err)
+		return false, fmt.Errorf("%w: dialing node service: %w", ErrRestoreDial, err)
 	}
 	log.G(ctx).Infof("creating restore request for container: %s", cfg.ContainerName)
 

shim/task/service_zeropod.go

@@ -159,13 +159,17 @@ func (w *wrapper) Create(ctx context.Context, r *taskAPI.CreateTaskRequest) (_ *
 	if cfg.AnyMigrationEnabled() {
 		skipStart, err := zshim.MigrationRestore(ctx, r, cfg)
 		if err != nil {
-			if !errors.Is(err, zshim.ErrRestoreRequestFailed) {
+			if errors.Is(err, zshim.ErrRestoreRequestFailed) ||
+				errors.Is(err, zshim.ErrRestoreDial) {
+				// if the restore fails with ErrRestoreRequestFailed it's very
+				// likely it simply did not find a matching migration. Equally,
+				// if the shim can't manage to dial the node service there's no
+				// chance it can be restored. We log it and create the container
+				// from scratch.
+				log.G(ctx).Errorf("restore request failed: %s", err)
+			} else {
 				return nil, err
 			}
-			// if the restore fails with ErrRestoreRequestFailed it's very
-			// likely it simply did not find a matching migration. We log it and
-			// create the container from scratch.
-			log.G(ctx).Errorf("restore request failed: %s", err)
 		}
 		zeropodContainer.SetSkipStart(skipStart)
 	}