Update CRD and GRPC.

Signed-off-by: Xun Jiang <jxun@vmware.com>

changelogs/unreleased/4818-jxun

@@ -0,0 +1 @@
+Add CSI VolumeSnapshot related metrics.

config/crd/v1/bases/velero.io_backups.yaml

@@ -354,6 +354,19 @@ spec:
                 format: date-time
                 nullable: true
                 type: string
+              csiVolumeSnapshotsAttempted:
+                description: CsiVolumeSnapshotsAttempted is the total number of attempted
+                  CSI VolumeSnapshots for this backup.
+                type: integer
+              csiVolumeSnapshotsCompleted:
+                description: CsiVolumeSnapshotsCompleted is the total number of successfully
+                  completed CSI VolumeSnapshots for this backup.
+                type: integer
+              csiVolumeSnapshotsStorageTotal:
+                description: CsiVolumeSnapshotsStorageTotal is the total storage size
+                  of created snapshots for this backup.
+                format: int64
+                type: integer
               errors:
                 description: Errors is a count of all error messages that were generated
                   during execution of the backup.  The actual errors are in the backup's

pkg/apis/velero/v1/backup.go

@@ -17,7 +17,6 @@ limitations under the License.
 package v1
 
 import (
-	resource "k8s.io/apimachinery/pkg/api/resource"
 	metav1 "k8s.io/apimachinery/pkg/apis/meta/v1"
 )
 
@@ -325,7 +324,7 @@ type BackupStatus struct {
 	// CsiVolumeSnapshotsStorageTotal is the total storage size of created
 	// snapshots for this backup.
 	// +optional
-	CsiVolumeSnapshotsStorageTotal resource.Quantity `json:"csiVolumeSnapshotsStorageTotal,omitempty"`
+	CsiVolumeSnapshotsStorageTotal int64 `json:"csiVolumeSnapshotsStorageTotal,omitempty"`
 }
 
 // BackupProgress stores information about the progress of a Backup's execution.

pkg/backup/request.go

@@ -21,6 +21,7 @@ import (
 	"sort"
 
 	snapshotv1api "github.com/kubernetes-csi/external-snapshotter/client/v4/apis/volumesnapshot/v1"
+
 	"github.com/vmware-tanzu/velero/internal/hook"
 	velerov1api "github.com/vmware-tanzu/velero/pkg/apis/velero/v1"
 	"github.com/vmware-tanzu/velero/pkg/plugin/framework"

pkg/controller/backup_controller.go

@@ -640,7 +640,12 @@ func (c *backupController) runBackup(backup *pkgbackup.Request) error {
 	for _, vs := range backup.CsiSnapshots {
 		if *vs.Status.ReadyToUse {
 			backup.Status.CsiVolumeSnapshotsCompleted++
-			backup.Status.CsiVolumeSnapshotsStorageTotal.Add(*vs.Status.RestoreSize)
+			storageSize, ret := vs.Status.RestoreSize.AsInt64()
+			if !ret {
+				backupLog.WithError(fmt.Errorf("fail to convert CSI snapshot size: %v to int64", backup.Status.CsiVolumeSnapshotsStorageTotal))
+				storageSize = 0
+			}
+			backup.Status.CsiVolumeSnapshotsStorageTotal += storageSize
 		}
 	}
 
@@ -707,12 +712,7 @@ func recordBackupMetrics(log logrus.FieldLogger, backup *velerov1api.Backup, bac
 		serverMetrics.RegisterCsiSnapshotAttempts(backupScheduleName, backup.Name, backup.Status.CsiVolumeSnapshotsAttempted)
 		serverMetrics.RegisterCsiSnapshotSuccesses(backupScheduleName, backup.Name, backup.Status.CsiVolumeSnapshotsCompleted)
 		serverMetrics.RegisterCsiSnapshotFailures(backupScheduleName, backup.Name, backup.Status.CsiVolumeSnapshotsAttempted-backup.Status.CsiVolumeSnapshotsCompleted)
-		storageSize, ret := backup.Status.CsiVolumeSnapshotsStorageTotal.AsInt64()
-		if !ret {
-			log.WithError(fmt.Errorf("fail to convert CSI snapshot size: %v to int64", backup.Status.CsiVolumeSnapshotsStorageTotal))
-			storageSize = 0
-		}
-		serverMetrics.RegisterCsiStorageSizeAdd(backupScheduleName, backup.Name, storageSize)
+		serverMetrics.RegisterCsiStorageSizeAdd(backupScheduleName, backup.Name, backup.Status.CsiVolumeSnapshotsStorageTotal)
 	}
 
 	if backup.Status.Progress != nil {

pkg/metrics/metrics.go

@@ -419,18 +419,6 @@ func (m *ServerMetrics) InitSchedule(scheduleName string) {
 	if c, ok := m.metrics[volumeSnapshotFailureTotal].(*prometheus.CounterVec); ok {
 		c.WithLabelValues(scheduleName).Add(0)
 	}
-	if c, ok := m.metrics[csiSnapshotAttemptTotal].(*prometheus.CounterVec); ok {
-		c.WithLabelValues(scheduleName).Add(0)
-	}
-	if c, ok := m.metrics[csiSnapshotSuccessTotal].(*prometheus.CounterVec); ok {
-		c.WithLabelValues(scheduleName).Add(0)
-	}
-	if c, ok := m.metrics[csiSnapshotFailureTotal].(*prometheus.CounterVec); ok {
-		c.WithLabelValues(scheduleName).Add(0)
-	}
-	if c, ok := m.metrics[csiSnapshotStorageTotal].(*prometheus.GaugeVec); ok {
-		c.WithLabelValues(scheduleName).Add(0)
-	}
 }
 
 // InitSchedule initializes counter metrics for a node.