
We are pleased to announce the release of KubeStash v2026.11.15
. This release rolls up the changes from v2026.11.10
and brings a resident KubeVirt CBT archiver with point-in-time restore, PVC support for kubectl kubestash download, DocumentDB continuous archiving with PITR, and a fix that stops a repository’s lastBackupTime from moving backwards. You can check out the full changelogs for v2026.11.15
and v2026.11.10
. In this post, we’ll highlight the key changes.
Quick highlights
- Added the
KubeVirtArchiverandKubeVirtArchiverRestoreAPIs: continuous, incremental VM backup using KubeVirt changed block tracking (CBT) with Restic, and restore to the latest or to a point in time. - Added PVC as storage support to
kubectl kubestash download, so snapshots in a local PVC backend can be downloaded. - Added DocumentDB continuous archiving with a
DocumentDBArchiver(full, WAL and manifest backups) and point-in-time recovery. - Fixed
RepositorylastBackupTimemoving backwards when snapshots are resynced.
What’s New
KubeVirt: Continuous CBT Backup and Point-in-Time Restore
KubeStash already backs up KubeVirt VMs on a schedule. The new KubeVirtArchiver makes this continuous. A resident job per VM uses KubeVirt’s native CBT pull-mode export and streams each disk’s map-and-extent artifact straight into Restic. That gives you:
- One full backup, then native incrementals. After the first full, every run carries only the dirty extents (plus map operations such as
zero/hole), not a full raw disk. - No VM-sized staging PVC. The only temporary volume is KubeVirt’s per-export scratch PVC, which the archiver deletes after each iteration.
- An ordered checkpoint lineage per disk, recorded on the
Snapshot(status.components.<disk>.resticCBTStats): each entry has akind(fullorincremental), itscheckpoint, and theparentCheckpointit extends. - Safe failure handling. A partial checkpoint is never published. If a run fails, the next one re-anchors with a full backup instead of extending a broken chain.
Prepare the VM first. Label it so it matches a changedBlockTrackingLabelSelectors entry in your KubeVirt CR (here cbt: enabled), plus a label the archiver selects on, and use masquerade networking so the pull-mode export can reach the VM:
apiVersion: kubevirt.io/v1
kind: VirtualMachine
metadata:
name: restic-cbt-vm
namespace: demo
labels:
cbt: enabled # matches KubeVirt's CBT label selector
kubestash.com/cbt-archive: enabled # selected by the KubeVirtArchiver
spec:
runStrategy: Always
template:
spec:
domain:
devices:
interfaces:
- name: default
masquerade: {} # pull-mode export traffic returns via the pod network
networks:
- name: default
pod: {}
# disks, DataVolumes and volumes omitted for brevity
Then create the archiver:
apiVersion: core.kubestash.com/v1alpha1
kind: KubeVirtArchiver
metadata:
name: restic-cbt-archiver
namespace: demo
spec:
virtualMachines:
namespaces:
from: Same
selector:
matchLabels:
kubestash.com/cbt-archive: enabled
backupStorage:
ref:
name: s3-storage
namespace: demo
subDir: /cbt-archiver
encryptionSecret:
name: encrypt-secret
namespace: demo
fullBackup:
driver: Restic
scheduler:
schedule: "0 1 * * *" # a fresh full every night
jobTemplate: {} # required by the API, even when empty
cbtBackup:
interval: 10m # incremental cadence, and so your RPO
scratchStorageClass: local-path
# skipQuiesce: true # only for guests without a QEMU guest agent
retentionPeriod: 7d
retentionSchedule: "0 0 * * *"
successfulHistoryLimit: 5
failedHistoryLimit: 5
To recover, create a KubeVirtArchiverRestore. It restores the newest complete multi-disk checkpoint, or, with recoveryPoint.targetTime, the newest complete checkpoint at or before that UTC time. Each checkpoint holds every disk plus an immutable copy of the VM manifest, so the restored VM is consistent. The target VM must not already exist: the restore refuses to overwrite one, and your source VM is never touched.
apiVersion: core.kubestash.com/v1alpha1
kind: KubeVirtArchiverRestore
metadata:
name: restic-cbt-vm-restore
namespace: demo
spec:
archiverRef:
name: restic-cbt-archiver
source:
name: restic-cbt-vm
target:
name: restic-cbt-vm-restored
recoveryPoint:
targetTime: "2026-11-10T07:47:39Z" # optional; omit to restore the latest checkpoint
A point-in-time target resolves to the newest complete checkpoint at or before the requested time, so your recovery granularity equals the
cbtBackup.interval. The restore can also be run from a different namespace than the source.
CLI: Download Snapshots from a PVC Backend
Local backends that are backed by a PersistentVolumeClaim are common in air-gapped and on-prem clusters, but kubectl kubestash download used to fail on them with unsupported type of local backend provided. It now supports a PVC as the storage backend. For example, with this BackupStorage:
apiVersion: storage.kubestash.com/v1alpha1
kind: BackupStorage
metadata:
name: local-storage-with-pvc
namespace: demo
spec:
default: true
deletionPolicy: WipeOut
runtimeSettings:
pod:
securityContext:
fsGroup: 65534
storage:
provider: local
local:
mountPath: /safe/data
persistentVolumeClaim:
claimName: backup-pvc
usagePolicy:
allowedNamespaces:
from: Same
you can pull any snapshot to your workstation:
$ kubectl kubestash download <snapshot-name> -n demo --destination=/tmp/snapshot-data
DocumentDB: Manifest Backup and Continuous Archiving (PITR)
KubeStash can now archive a KubeDB-managed DocumentDB continuously. The new DocumentDBArchiver combines three things:
- Full backups with Restic, taken on a schedule.
- WAL log backups, shipped continuously so you can recover to any second inside the archived window.
- Manifest backups of the Kubernetes objects around the database (the new manifest backup support in
apimachinery), kept in their own repository.
Create the archiver. A DocumentDB joins it through a label selector:
apiVersion: archiver.kubedb.com/v1alpha1
kind: DocumentDBArchiver
metadata:
name: documentdb-archiver
namespace: demo
spec:
databases:
namespaces:
from: Selector
selector:
matchLabels:
kubernetes.io/metadata.name: demo
selector:
matchLabels:
archiver: enabled
fullBackup:
driver: Restic
scheduler:
schedule: "0 1 * * *" # nightly full backup
successfulJobsHistoryLimit: 1
failedJobsHistoryLimit: 1
manifestBackup:
scheduler:
schedule: "*/30 * * * *" # manifests every 30 minutes
successfulJobsHistoryLimit: 1
failedJobsHistoryLimit: 1
logBackup:
runtimeSettings: {}
backupStorage:
ref:
name: s3-storage
namespace: demo
encryptionSecret:
name: encrypt-secret
namespace: demo
retentionPolicy:
name: demo-retention
namespace: demo
Then label the DocumentDB and point spec.archiver.ref at the archiver. WAL is only shipped when a segment fills (16 MB) or when archive_timeout fires, so set archive_timeout to bound how much data you can lose. Here it is 60 seconds:
apiVersion: kubedb.com/v1alpha2
kind: DocumentDB
metadata:
name: dcdb-pitr
namespace: demo
labels:
archiver: enabled
spec:
version: pg17-0.109.0
replicas: 1
storageType: Durable
storage:
accessModes: ["ReadWriteOnce"]
resources:
requests:
storage: 1Gi
configuration:
inline:
user.conf: |
archive_timeout = 60
archiver:
ref:
name: documentdb-archiver
namespace: demo
deletionPolicy: WipeOut
To recover, create a new DocumentDB with spec.init.archiver. Recovery is never done in place. recoveryTimestamp is an RFC3339 UTC time with second precision. manifestRepository is optional; when set, the restore waits for a manifest snapshot before it starts.
apiVersion: kubedb.com/v1alpha2
kind: DocumentDB
metadata:
name: dcdb-pitr-restored
namespace: demo
spec:
version: pg17-0.109.0
replicas: 1
storageType: Durable
storage:
accessModes: ["ReadWriteOnce"]
resources:
requests:
storage: 1Gi
init:
archiver:
recoveryTimestamp: "2026-11-15T11:10:22Z" # a UTC time inside the archived window
encryptionSecret:
name: encrypt-secret
namespace: demo
fullDBRepository:
name: dcdb-pitr-full
namespace: demo
manifestRepository:
name: dcdb-pitr-manifest
namespace: demo
deletionPolicy: WipeOut
In our test we inserted a batch of rows, noted the time, inserted a second batch, and restored to a timestamp between the two. The restored database contained only the first batch.
Bug Fixes and Improvements
lastBackupTimeno longer goes backwards. When snapshots were resynced, theRepository’slastBackupTimecould be set to an older snapshot’s time. That made a healthy repository look stale and could trigger false RPO alerts. It now only moves forward.
Component Versions
The following components have been released as part of KubeStash v2026.11.15:
| Component | Version |
|---|---|
| kubestash/apimachinery | v0.32.0 |
| kubestash/kubestash | v0.32.0 |
| kubestash/cli | v0.31.0 |
| kubestash/kubedump | v0.31.0 |
| kubestash/pvc | v0.31.0 |
| kubestash/workload | v0.31.0 |
| kubestash/manifest | v0.24.0 |
| kubestash/volume-snapshotter | v0.31.0 |
| kubestash/vault | v0.6.0 |
| kubestash/kubevirt | v0.4.0 |
| kubestash/ui-server | v0.4.0 |
| kubestash/installer | v2026.11.15 |
| kubestash/docs | v2026.11.15 |
What Next?
Please try the latest release and give us your valuable feedback.
- If you want to install KubeStash in a clean cluster, please follow the installation instruction from HERE .
- If you want to upgrade KubeStash from a previous version, please follow the upgrade instruction from HERE .
Support
To speak with us, please leave a message on our website .
