Merge branch 'main' into fail-fs-backup-on-windows-nodes

pkg/cmd/cli/install/install.go

@@ -66,6 +66,7 @@ type Options struct {
 	BackupStorageConfig             flag.Map
 	VolumeSnapshotConfig            flag.Map
 	UseNodeAgent                    bool
+	UseNodeAgentWindows             bool
 	PrivilegedNodeAgent             bool
 	Wait                            bool
 	UseVolumeSnapshots              bool
@@ -117,7 +118,8 @@ func (o *Options) BindFlags(flags *pflag.FlagSet) {
 	flags.BoolVar(&o.UseVolumeSnapshots, "use-volume-snapshots", o.UseVolumeSnapshots, "Whether or not to create snapshot location automatically. Set to false if you do not plan to create volume snapshots via a storage provider.")
 	flags.BoolVar(&o.RestoreOnly, "restore-only", o.RestoreOnly, "Run the server in restore-only mode. Optional.")
 	flags.BoolVar(&o.DryRun, "dry-run", o.DryRun, "Generate resources, but don't send them to the cluster. Use with -o. Optional.")
-	flags.BoolVar(&o.UseNodeAgent, "use-node-agent", o.UseNodeAgent, "Create Velero node-agent daemonset. Optional. Velero node-agent hosts Velero modules that need to run in one or more nodes(i.e. Restic, Kopia).")
+	flags.BoolVar(&o.UseNodeAgent, "use-node-agent", o.UseNodeAgent, "Create Velero node-agent daemonset. Optional. Velero node-agent hosts and associates Velero modules that need to run in one or more Linux nodes.")
+	flags.BoolVar(&o.UseNodeAgentWindows, "use-node-agent-windows", o.UseNodeAgentWindows, "Create Velero node-agent-windows daemonset. Optional. Velero node-agent-windows hosts and associates Velero modules that need to run in one or more Windows nodes.")
 	flags.BoolVar(&o.PrivilegedNodeAgent, "privileged-node-agent", o.PrivilegedNodeAgent, "Use privileged mode for the node agent. Optional. Required to backup block devices.")
 	flags.BoolVar(&o.Wait, "wait", o.Wait, "Wait for Velero deployment to be ready. Optional.")
 	flags.DurationVar(&o.DefaultRepoMaintenanceFrequency, "default-repo-maintain-frequency", o.DefaultRepoMaintenanceFrequency, "How often 'maintain' is run for backup repositories by default. Optional.")
@@ -267,6 +269,7 @@ func (o *Options) AsVeleroOptions() (*install.VeleroOptions, error) {
 		SecretData:                      secretData,
 		RestoreOnly:                     o.RestoreOnly,
 		UseNodeAgent:                    o.UseNodeAgent,
+		UseNodeAgentWindows:             o.UseNodeAgentWindows,
 		PrivilegedNodeAgent:             o.PrivilegedNodeAgent,
 		UseVolumeSnapshots:              o.UseVolumeSnapshots,
 		BSLConfig:                       o.BackupStorageConfig.Data(),
@@ -392,7 +395,12 @@ func (o *Options) Run(c *cobra.Command, f client.Factory) error {
 
 		if o.UseNodeAgent {
 			fmt.Println("Waiting for node-agent daemonset to be ready.")
-			if _, err = install.DaemonSetIsReady(dynamicFactory, o.Namespace); err != nil {
+			if _, err = install.NodeAgentIsReady(dynamicFactory, o.Namespace); err != nil {
+				return errors.Wrap(err, errorMsg)
+			}
+
+			fmt.Println("Waiting for node-agent-windows daemonset to be ready.")
+			if _, err = install.NodeAgentWindowsIsReady(dynamicFactory, o.Namespace); err != nil {
 				return errors.Wrap(err, errorMsg)
 			}
 		}

pkg/cmd/cli/nodeagent/server.go

@@ -200,13 +200,31 @@ func newNodeAgentServer(logger logrus.FieldLogger, factory client.Factory, confi
 			},
 		},
 	}
-	mgr, err := ctrl.NewManager(clientConfig, ctrl.Options{
-		Scheme: scheme,
-		Cache:  cacheOption,
-	})
+
+	var mgr manager.Manager
+	retry := 10
+	for {
+		mgr, err = ctrl.NewManager(clientConfig, ctrl.Options{
+			Scheme: scheme,
+			Cache:  cacheOption,
+		})
+		if err == nil {
+			break
+		}
+
+		retry--
+		if retry == 0 {
+			break
+		}
+
+		logger.WithError(err).Warn("Failed to create controller manager, need retry")
+
+		time.Sleep(time.Second)
+	}
+
 	if err != nil {
 		cancelFunc()
-		return nil, err
+		return nil, errors.Wrap(err, "error creating controller manager")
 	}
 
 	s := &nodeAgentServer{

pkg/cmd/server/server.go

@@ -240,17 +240,34 @@ func newServer(f client.Factory, config *config.Config, logger *logrus.Logger) (
 
 	ctrl.SetLogger(logrusr.New(logger))
 
-	mgr, err := ctrl.NewManager(clientConfig, ctrl.Options{
-		Scheme: scheme,
-		Cache: cache.Options{
-			DefaultNamespaces: map[string]cache.Config{
-				f.Namespace(): {},
+	var mgr manager.Manager
+	retry := 10
+	for {
+		mgr, err = ctrl.NewManager(clientConfig, ctrl.Options{
+			Scheme: scheme,
+			Cache: cache.Options{
+				DefaultNamespaces: map[string]cache.Config{
+					f.Namespace(): {},
+				},
 			},
-		},
-	})
+		})
+		if err == nil {
+			break
+		}
+
+		retry--
+		if retry == 0 {
+			break
+		}
+
+		logger.WithError(err).Warn("Failed to create controller manager, need retry")
+
+		time.Sleep(time.Second)
+	}
+
 	if err != nil {
 		cancelFunc()
-		return nil, err
+		return nil, errors.Wrap(err, "error creating controller manager")
 	}
 
 	credentialFileStore, err := credentials.NewNamespacedFileStore(