This document is intended for developers building backup/disaster recovery solutions for KubeVirt.
Such solutions typically perform the following high level operations.
- Build dependency graph for all required Kubernetes resources
- Quiesce (freeze) applications
- Snapshot PersistentVolumeClaim data
- Unquiesce (thaw) applications
- Copy all necessary Kubernetes resource definitions to a shared storage location
- (optional) Export snapshotted PVC data to a shared storage location
Steps 3, 5, and 6 are beyond the scope of this document.
- Populate PersistentVolumeClaims with snapshot data
- Sanitize and apply all relevant Kubernetes resource definitions
Step 1 is beyond the scope of this document.
Velero is a popular tool for backing up/migrating Kubernetes clusters. The KubeVirt team actively maintains a plugin for use with Velero. The plugin implements much of the logic described in this document.
The VirtualMachineSnapshot API provides an easy way for KubeVirt users to backup VirtualMachines within a cluster. On its own, it is not suitable for offsite backup or disaster recovery. But combined with the VirtualMachineExport API, VirtualMachine volumes may be made available for copy to remote locations.
In this section, KubeVirt resources and their relationships will be explored by showing how yaml snippets map to nodes in an object graph.
Nodes in the object graph are represented by the following tuple:
(APIGroup, Kind, namespace, name)
apiVersion: kubevirt.io/v1
kind: VirtualMachine
metadata:
name: vm1
namespace: ns1
...- ("kubevirt.io", "VirtualMachine", "ns1", "vm1")
...
spec:
instancetype:
kind: VirtualMachineInstancetype
name: small
...- ("instancetype.kubevirt.io", "VirtualMachineInstancetype", "ns1", "small")
...
spec:
preference:
kind: VirtualMachinePreference
name: windows
...- ("instancetype.kubevirt.io", "VirtualMachinePreference", "ns1", "windows")
apiVersion: kubevirt.io/v1
kind: VirtualMachineInstance
metadata:
name: vmi1
namespace: ns1
...- ("kubevirt.io", "VirtualMachineInstance", "ns1", "vmi1")
- ("", "Pod", "ns1", "virt-launcher-vmi1-XXXXX") *
* Each VirtualMachineInstance has a corresponding uniquely named Pod. The backup process can look up the name of this pod by using the kubevirt.io/created-by=<UID of VirtualMachineInstance> label selector.
...
spec:
volumes:
- name: v1
persistentVolumeClaim:
claimName: pvc1
...- ("", "PersistentVolumeClaim", "ns1", "pvc1")
- ("cdi.kubevirt.io", "DataVolume", "ns1", "pvc1")
...
spec:
volumes:
- name: v1
dataVolume:
name: dv1
...- ("", "PersistentVolumeClaim", "ns1", "dv1")
- ("cdi.kubevirt.io", "DataVolume", "ns1", "dv1")
...
spec:
volumes:
- name: v1
configMap:
name: cm1
...- ("", "ConfigMap", "ns1", "cm1")
...
spec:
volumes:
- name: v1
secret:
secretName: s1
...- ("", "Secret", "ns1", "s1")
...
spec:
volumes:
- name: v1
serviceAccount:
serviceAccountName: sa1
...- ("", "ServiceAccount", "ns1", "sa1")
...
spec:
volumes:
- name: v1
memoryDump:
claimName: pvc1
...- ("", "PersistentVolumeClaim", "ns1", "pvc1")
...
spec:
accessCredentials:
- sshPublicKey:
source:
secret:
secretName: my-pub-key
...- ("", "Secret", "ns1", "my-pub-key")
...
spec:
accessCredentials:
- userPassword:
source:
secret:
secretName: my-user-password
...- ("", "Secret", "ns1", "my-user-password")
apiVersion: kubevirt.io/v1
kind: VirtualMachineInstanceReplicaSet
metadata:
name: vmirs1
namespace: ns1
...- ("kubevirt.io", "VirtualMachineInstanceReplicaSet", "ns1", "vmirs1")
- ("kubevirt.io", "VirtualMachineInstance", "ns1", "vmirs1XXXX1") *
- ("kubevirt.io", "VirtualMachineInstance", "ns1", "vmirs1XXXX2") *
* there are usually multiple VirtualMachineInstances corresponding to a VirtualMachineInstanceReplicaSet. The backup process can look up the name of this pod by using the kubevirt.io/vmReplicaSet=<name of VirtualMachineInstanceReplicaSet> label selector.
See this guide for how to execute the freeze/thaw hooks for each VirtualMachineInstance encountered in the object graph.
If restoring to a different cluster, and mac address or bios serial number are explicitly set, you should make sure there will be no collisions. These values are set in:
/spec/template/spec/domain/devices/interfaces/<index>/macAddress
/spec/template/spec/domain/firmware/serial
If a VirtualMachineInstance is owned by a VirtualMachine, it should not be restored. The KubeVirt controller will recreate the resource based on the VirtualMachine definition. Otherwise, VirtualMachineInstance definitions may be backed up/restored with the same precautions as VirtualMachine for mac address/bios.
/spec/domain/devices/interfaces/<index>/macAddress
/spec/domain/firmware/serial
A Pod with “virt-launcher-” prefix that is *owned by a VirtualMachineInstance should not be restored.
DataVolumes in Succeeded phase (status.phase) should have the following annotation added at restore time. Otherwise, the associated PersistentVolumeClaim may get corrupted. DataVolumes in any phase other than Succeeded do not need to be annotated.
cdi.kubevirt.io/storage.prePopulated: <datavolume name>PersistentVolumeClaims owned by DataVolumes must have the following annotation added at backup/restore time.
cdi.kubevirt.io/storage.populatedFor: <datavolume name>