cache volume configuration

Signed-off-by: Lyndon-Li <lyonghui@vmware.com>

changelogs/unreleased/9370-Lyndon-Li

@@ -0,0 +1 @@
+Add cache volume configuration

pkg/builder/storage_class_builder.go

@@ -17,6 +17,7 @@ limitations under the License.
 package builder
 
 import (
+	corev1api "k8s.io/api/core/v1"
 	storagev1api "k8s.io/api/storage/v1"
 	metav1 "k8s.io/apimachinery/pkg/apis/meta/v1"
 )
@@ -87,3 +88,9 @@ func (b *StorageClassBuilder) Provisioner(provisioner string) *StorageClassBuild
 	b.object.Provisioner = provisioner
 	return b
 }
+
+// ReclaimPolicy sets StorageClass's reclaimPolicy.
+func (b *StorageClassBuilder) ReclaimPolicy(policy corev1api.PersistentVolumeReclaimPolicy) *StorageClassBuilder {
+	b.object.ReclaimPolicy = &policy
+	return b
+}

pkg/cmd/cli/nodeagent/server.go

@@ -84,6 +84,7 @@ type nodeAgentServerConfig struct {
 	resourceTimeout         time.Duration
 	dataMoverPrepareTimeout time.Duration
 	nodeAgentConfig         string
+	backupRepoConfig        string
 }
 
 func NewServerCommand(f client.Factory) *cobra.Command {
@@ -121,6 +122,7 @@ func NewServerCommand(f client.Factory) *cobra.Command {
 	command.Flags().DurationVar(&config.dataMoverPrepareTimeout, "data-mover-prepare-timeout", config.dataMoverPrepareTimeout, "How long to wait for preparing a DataUpload/DataDownload. Default is 30 minutes.")
 	command.Flags().StringVar(&config.metricsAddress, "metrics-address", config.metricsAddress, "The address to expose prometheus metrics")
 	command.Flags().StringVar(&config.nodeAgentConfig, "node-agent-configmap", config.nodeAgentConfig, "The name of ConfigMap containing node-agent configurations.")
+	command.Flags().StringVar(&config.backupRepoConfig, "backup-repository-configmap", config.backupRepoConfig, "The name of ConfigMap containing backup repository configurations.")
 
 	return command
 }
@@ -140,6 +142,7 @@ type nodeAgentServer struct {
 	csiSnapshotClient *snapshotv1client.Clientset
 	dataPathMgr       *datapath.Manager
 	dataPathConfigs   *velerotypes.NodeAgentConfigs
+	backupRepoConfigs map[string]string
 	vgdpCounter       *exposer.VgdpCounter
 }
 
@@ -254,6 +257,11 @@ func newNodeAgentServer(logger logrus.FieldLogger, factory client.Factory, confi
 	if err := s.getDataPathConfigs(); err != nil {
 		return nil, err
 	}
+
+	if err := s.getBackupRepoConfigs(); err != nil {
+		return nil, err
+	}
+
 	s.dataPathMgr = datapath.NewManager(s.getDataPathConcurrentNum(defaultDataPathConcurrentNum))
 
 	return s, nil
@@ -329,6 +337,14 @@ func (s *nodeAgentServer) run() {
 		}
 	}
 
+	if s.dataPathConfigs != nil && s.dataPathConfigs.CachePVCConfig != nil {
+		if err := s.validateCachePVCConfig(*s.dataPathConfigs.CachePVCConfig); err != nil {
+			s.logger.WithError(err).Warnf("Ignore cache config %v", s.dataPathConfigs.CachePVCConfig)
+		} else {
+			s.logger.Infof("Using cache volume configs %v", s.dataPathConfigs.CachePVCConfig)
+		}
+	}
+
 	pvbReconciler := controller.NewPodVolumeBackupReconciler(s.mgr.GetClient(), s.mgr, s.kubeClient, s.dataPathMgr, s.vgdpCounter, s.nodeName, s.config.dataMoverPrepareTimeout, s.config.resourceTimeout, podResources, s.metrics, s.logger, dataMovePriorityClass, privilegedFsBackup)
 	if err := pvbReconciler.SetupWithManager(s.mgr); err != nil {
 		s.logger.Fatal(err, "unable to create controller", "controller", constant.ControllerPodVolumeBackup)
@@ -557,14 +573,32 @@ func (s *nodeAgentServer) getDataPathConfigs() error {
 
 	configs, err := getConfigsFunc(s.ctx, s.namespace, s.kubeClient, s.config.nodeAgentConfig)
 	if err != nil {
-		s.logger.WithError(err).Errorf("Failed to get node agent configs from configMap %s, ignore it", s.config.nodeAgentConfig)
-		return err
+		return errors.Wrapf(err, "error getting node agent configs from configMap %s", s.config.nodeAgentConfig)
 	}
 
 	s.dataPathConfigs = configs
 	return nil
 }
 
+func (s *nodeAgentServer) getBackupRepoConfigs() error {
+	if s.config.backupRepoConfig == "" {
+		s.logger.Info("No backup repo configMap is specified")
+		return nil
+	}
+
+	cm, err := s.kubeClient.CoreV1().ConfigMaps(s.namespace).Get(s.ctx, s.config.backupRepoConfig, metav1.GetOptions{})
+	if err != nil {
+		return errors.Wrapf(err, "error getting backup repo configs from configMap %s", s.config.backupRepoConfig)
+	}
+
+	if cm.Data == nil {
+		return errors.Errorf("no data is in the backup repo configMap %s", s.config.backupRepoConfig)
+	}
+
+	s.backupRepoConfigs = cm.Data
+	return nil
+}
+
 func (s *nodeAgentServer) getDataPathConcurrentNum(defaultNum int) int {
 	configs := s.dataPathConfigs
 
@@ -620,3 +654,20 @@ func (s *nodeAgentServer) getDataPathConcurrentNum(defaultNum int) int {
 
 	return concurrentNum
 }
+
+func (s *nodeAgentServer) validateCachePVCConfig(config velerotypes.CachePVC) error {
+	if config.StorageClass == "" {
+		return errors.New("storage class is absent")
+	}
+
+	sc, err := s.kubeClient.StorageV1().StorageClasses().Get(s.ctx, config.StorageClass, metav1.GetOptions{})
+	if err != nil {
+		return errors.Wrapf(err, "error getting storage class %s", config.StorageClass)
+	}
+
+	if sc.ReclaimPolicy != nil && *sc.ReclaimPolicy != corev1api.PersistentVolumeReclaimDelete {
+		return errors.Errorf("unexpected storage class reclaim policy %v", *sc.ReclaimPolicy)
+	}
+
+	return nil
+}

pkg/cmd/cli/nodeagent/server_test.go

@@ -24,6 +24,7 @@ import (
 	"testing"
 
 	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
 	corev1api "k8s.io/api/core/v1"
 	metav1 "k8s.io/apimachinery/pkg/apis/meta/v1"
 	"k8s.io/apimachinery/pkg/runtime"
@@ -34,6 +35,8 @@ import (
 	"github.com/vmware-tanzu/velero/pkg/nodeagent"
 	testutil "github.com/vmware-tanzu/velero/pkg/test"
 	velerotypes "github.com/vmware-tanzu/velero/pkg/types"
+
+	velerov1api "github.com/vmware-tanzu/velero/pkg/apis/velero/v1"
 )
 
 func Test_validatePodVolumesHostPath(t *testing.T) {
@@ -142,11 +145,10 @@ func Test_getDataPathConfigs(t *testing.T) {
 		getFunc       func(context.Context, string, kubernetes.Interface, string) (*velerotypes.NodeAgentConfigs, error)
 		configMapName string
 		expectConfigs *velerotypes.NodeAgentConfigs
-		expectLog     string
+		expectedErr   string
 	}{
 		{
-			name:      "no config specified",
-			expectLog: "No node-agent configMap is specified",
+			name: "no config specified",
 		},
 		{
 			name:          "failed to get configs",
@@ -154,7 +156,7 @@ func Test_getDataPathConfigs(t *testing.T) {
 			getFunc: func(context.Context, string, kubernetes.Interface, string) (*velerotypes.NodeAgentConfigs, error) {
 				return nil, errors.New("fake-get-error")
 			},
-			expectLog: "Failed to get node agent configs from configMap node-agent-config, ignore it",
+			expectedErr: "error getting node agent configs from configMap node-agent-config: fake-get-error",
 		},
 		{
 			name:          "configs cm not found",
@@ -162,7 +164,7 @@ func Test_getDataPathConfigs(t *testing.T) {
 			getFunc: func(context.Context, string, kubernetes.Interface, string) (*velerotypes.NodeAgentConfigs, error) {
 				return nil, errors.New("fake-not-found-error")
 			},
-			expectLog: "Failed to get node agent configs from configMap node-agent-config, ignore it",
+			expectedErr: "error getting node agent configs from configMap node-agent-config: fake-not-found-error",
 		},
 
 		{
@@ -177,23 +179,21 @@ func Test_getDataPathConfigs(t *testing.T) {
 
 	for _, test := range tests {
 		t.Run(test.name, func(t *testing.T) {
-			logBuffer := ""
-
 			s := &nodeAgentServer{
 				config: nodeAgentServerConfig{
 					nodeAgentConfig: test.configMapName,
 				},
-				logger: testutil.NewSingleLogger(&logBuffer),
+				logger: testutil.NewLogger(),
 			}
 
 			getConfigsFunc = test.getFunc
 
-			s.getDataPathConfigs()
-			assert.Equal(t, test.expectConfigs, s.dataPathConfigs)
-			if test.expectLog == "" {
-				assert.Empty(t, logBuffer)
+			err := s.getDataPathConfigs()
+			if test.expectedErr == "" {
+				require.NoError(t, err)
+				assert.Equal(t, test.expectConfigs, s.dataPathConfigs)
 			} else {
-				assert.Contains(t, logBuffer, test.expectLog)
+				require.EqualError(t, err, test.expectedErr)
 			}
 		})
 	}
@@ -416,3 +416,117 @@ func Test_getDataPathConcurrentNum(t *testing.T) {
 		})
 	}
 }
+
+func TestGetBackupRepoConfigs(t *testing.T) {
+	cmNoData := builder.ForConfigMap(velerov1api.DefaultNamespace, "backup-repo-config").Result()
+	cmWithData := builder.ForConfigMap(velerov1api.DefaultNamespace, "backup-repo-config").Data("cacheLimit", "100").Result()
+
+	tests := []struct {
+		name          string
+		configMapName string
+		kubeClientObj []runtime.Object
+		expectConfigs map[string]string
+		expectedErr   string
+	}{
+		{
+			name: "no config specified",
+		},
+		{
+			name:          "failed to get configs",
+			configMapName: "backup-repo-config",
+			expectedErr:   "error getting backup repo configs from configMap backup-repo-config: configmaps \"backup-repo-config\" not found",
+		},
+		{
+			name:          "configs data not found",
+			kubeClientObj: []runtime.Object{cmNoData},
+			configMapName: "backup-repo-config",
+			expectedErr:   "no data is in the backup repo configMap backup-repo-config",
+		},
+		{
+			name:          "succeed",
+			configMapName: "backup-repo-config",
+			kubeClientObj: []runtime.Object{cmWithData},
+			expectConfigs: map[string]string{"cacheLimit": "100"},
+		},
+	}
+
+	for _, test := range tests {
+		t.Run(test.name, func(t *testing.T) {
+			fakeKubeClient := fake.NewSimpleClientset(test.kubeClientObj...)
+
+			s := &nodeAgentServer{
+				namespace:  velerov1api.DefaultNamespace,
+				kubeClient: fakeKubeClient,
+				config: nodeAgentServerConfig{
+					backupRepoConfig: test.configMapName,
+				},
+				logger: testutil.NewLogger(),
+			}
+
+			err := s.getBackupRepoConfigs()
+			if test.expectedErr == "" {
+				require.NoError(t, err)
+				require.Equal(t, test.expectConfigs, s.backupRepoConfigs)
+			} else {
+				require.EqualError(t, err, test.expectedErr)
+			}
+		})
+	}
+}
+
+func TestValidateCachePVCConfig(t *testing.T) {
+	scWithRetainPolicy := builder.ForStorageClass("fake-storage-class").ReclaimPolicy(corev1api.PersistentVolumeReclaimRetain).Result()
+	scWithDeletePolicy := builder.ForStorageClass("fake-storage-class").ReclaimPolicy(corev1api.PersistentVolumeReclaimDelete).Result()
+	scWithNoPolicy := builder.ForStorageClass("fake-storage-class").Result()
+
+	tests := []struct {
+		name          string
+		config        velerotypes.CachePVC
+		kubeClientObj []runtime.Object
+		expectedErr   string
+	}{
+		{
+			name:        "no storage class",
+			expectedErr: "storage class is absent",
+		},
+		{
+			name:        "failed to get storage class",
+			config:      velerotypes.CachePVC{StorageClass: "fake-storage-class"},
+			expectedErr: "error getting storage class fake-storage-class: storageclasses.storage.k8s.io \"fake-storage-class\" not found",
+		},
+		{
+			name:          "storage class reclaim policy is not expected",
+			config:        velerotypes.CachePVC{StorageClass: "fake-storage-class"},
+			kubeClientObj: []runtime.Object{scWithRetainPolicy},
+			expectedErr:   "unexpected storage class reclaim policy Retain",
+		},
+		{
+			name:          "storage class reclaim policy is delete",
+			config:        velerotypes.CachePVC{StorageClass: "fake-storage-class"},
+			kubeClientObj: []runtime.Object{scWithDeletePolicy},
+		},
+		{
+			name:          "storage class with no reclaim policy",
+			config:        velerotypes.CachePVC{StorageClass: "fake-storage-class"},
+			kubeClientObj: []runtime.Object{scWithNoPolicy},
+		},
+	}
+
+	for _, test := range tests {
+		t.Run(test.name, func(t *testing.T) {
+			fakeKubeClient := fake.NewSimpleClientset(test.kubeClientObj...)
+
+			s := &nodeAgentServer{
+				kubeClient: fakeKubeClient,
+			}
+
+			err := s.validateCachePVCConfig(test.config)
+
+			if test.expectedErr == "" {
+				require.NoError(t, err)
+			} else {
+				require.EqualError(t, err, test.expectedErr)
+			}
+		})
+	}
+}

pkg/install/daemonset.go

@@ -57,6 +57,10 @@ func DaemonSet(namespace string, opts ...podTemplateOption) *appsv1api.DaemonSet
 		daemonSetArgs = append(daemonSetArgs, fmt.Sprintf("--node-agent-configmap=%s", c.nodeAgentConfigMap))
 	}
 
+	if len(c.backupRepoConfigMap) > 0 {
+		daemonSetArgs = append(daemonSetArgs, fmt.Sprintf("--backup-repository-configmap=%s", c.backupRepoConfigMap))
+	}
+
 	userID := int64(0)
 	mountPropagationMode := corev1api.MountPropagationHostToContainer
 

pkg/install/daemonset_test.go

@@ -60,6 +60,10 @@ func TestDaemonSet(t *testing.T) {
 	assert.Len(t, ds.Spec.Template.Spec.Containers[0].Args, 3)
 	assert.Equal(t, "--node-agent-configmap=node-agent-config-map", ds.Spec.Template.Spec.Containers[0].Args[2])
 
+	ds = DaemonSet("velero", WithBackupRepoConfigMap("backup-repo-config-map"))
+	assert.Len(t, ds.Spec.Template.Spec.Containers[0].Args, 3)
+	assert.Equal(t, "--backup-repository-configmap=backup-repo-config-map", ds.Spec.Template.Spec.Containers[0].Args[2])
+
 	ds = DaemonSet("velero", WithServiceAccountName("test-sa"))
 	assert.Equal(t, "test-sa", ds.Spec.Template.Spec.ServiceAccountName)
 

pkg/install/resources.go

@@ -426,6 +426,10 @@ func AllResources(o *VeleroOptions) *unstructured.UnstructuredList {
 			dsOpts = append(dsOpts, WithNodeAgentConfigMap(o.NodeAgentConfigMap))
 		}
 
+		if len(o.BackupRepoConfigMap) > 0 {
+			dsOpts = append(dsOpts, WithBackupRepoConfigMap(o.BackupRepoConfigMap))
+		}
+
 		if len(o.KubeletRootDir) > 0 {
 			dsOpts = append(dsOpts, WithKubeletRootDir(o.KubeletRootDir))
 		}

pkg/types/node_agent.go

@@ -66,6 +66,14 @@ type RestorePVC struct {
 	IgnoreDelayBinding bool `json:"ignoreDelayBinding,omitempty"`
 }
 
+type CachePVC struct {
+	// StorageClass specifies the storage class for cache PVC
+	StorageClass string
+
+	// ResidentThreshold specifies the minimum size of the backup data to create cache PVC
+	ResidentThreshold int64
+}
+
 type NodeAgentConfigs struct {
 	// LoadConcurrency is the config for data path load concurrency per node.
 	LoadConcurrency *LoadConcurrency `json:"loadConcurrency,omitempty"`
@@ -87,4 +95,7 @@ type NodeAgentConfigs struct {
 
 	// PrivilegedFsBackup determines whether to create fs-backup pods as privileged pods
 	PrivilegedFsBackup bool `json:"privilegedFsBackup,omitempty"`
+
+	// CachePVCConfig is the config for cachePVC
+	CachePVCConfig *CachePVC `json:"cachePVC,omitempty"`
 }