Velero limitations
Limits
Section titled “Limits”- No file system backup. A PVC saved with Velero file system backup (kopia or restic through
the node agent) cannot be restored without its pod. bret refuses it with
BackupUnsupported. Use CSI snapshots, best withsnapshotMoveData. - Scratch namespace. The check runs in
bret-scratch-<hash>, not in your namespace. A check withserviceAccountName,hostPath, or a configMap, secret, projected or persistentVolumeClaim volume endsFailedwithProviderUnsupported. - Cluster-wide rights. With Velero on, the operator can read and create Secrets and ConfigMaps and manage pods and PVCs in every namespace.
- Velero picks size and storage class. PVCs come back at their original size and storage
class. A Velero source has no
size,autoExpandorstorageClassName. - One Backup per run. Every source of a
Restorenames the samebackupRef, and every source of aRestoreTestthe samebackupSelector. - Restored PVCs only. Deployments, Services and other resources of the backup are not restored or tested.
- Backups without data movement die with the cluster. A CSI snapshot without
snapshotMoveDatalives in the cluster. bret can test it, but it does not prove off-site recovery. - Completed only.
PartiallyFailedbackups are never picked.
Troubleshooting
Section titled “Troubleshooting”Restore fails with BackupUnsupported
Section titled “Restore fails with BackupUnsupported”- Cause: the
BackuphassnapshotVolumes: false, or a source PVC was saved with file system backup. - Fix: back the PVC up as a CSI snapshot. See
Volume data must be a snapshot.
Reason:
BackupUnsupported.
Velero backup is gone after the source cluster is lost
Section titled “Velero backup is gone after the source cluster is lost”- Symptom: the restore works only while the snapshot still exists in the cluster.
- Cause: the backup was created without data movement.
- Fix: create backups with
--snapshot-move-data(orsnapshotMoveData: trueon theSchedule).
Restore warning ConfigMap:kube-root-ca.crt already exists
Section titled “Restore warning ConfigMap:kube-root-ca.crt already exists”- Cause: every namespace already has this ConfigMap.
- Fix: none needed. The warning is harmless.
RestoreTest stays at BackupNotFound
Section titled “RestoreTest stays at BackupNotFound”- Cause: no
CompletedBackupwith the selected labels covers the namespace. - Fix: check
kubectl -n velero get backup --show-labels, the phase, andspec.includedNamespaces/spec.excludedNamespacesof the newestBackup. Reason:BackupNotFound.
Restore fails with RestoredVolumeMissing
Section titled “Restore fails with RestoredVolumeMissing”- Cause: a
sourcePVCwas not in theBackup,spec.velerofiltered it out, or the VeleroRestorecompleted without restoring any item. - Fix: check that the PVC carried the selected labels when the backup ran, and
velero restore describe <name> --details. Reason:RestoredVolumeMissing.
Restore fails with CheckRefMissing
Section titled “Restore fails with CheckRefMissing”- Cause: the check names a Secret or ConfigMap that is not in the
Restorenamespace, so bret cannot copy it. - Fix: create it, or mark the reference
optional: true. Reason:CheckRefMissing.
Restore fails with RestoreFailed
Section titled “Restore fails with RestoreFailed”- Cause: the Velero
RestoreendedFailed,PartiallyFailedorFailedValidation. - Fix:
velero restore describe <name> --detailsandvelero restore logs <name>. The name is instatus.providerRefsof theRestore. Reason:RestoreFailed.
