A destination is where backups are stored. VMStack supports five kinds: an SFTP server (a Hetzner Storage Box or any other), an S3-compatible bucket, a Proxmox Backup Server, an NFS export, and a directory on each node's own disk. Nodes connect to the destination; the panel only sends them its details inside each task.
Add a destination#
-
Go to Backups › Destinations › Add a destination.
-
Under General, give it a Name and choose its Kind:
Kind Use it for SFTP (Hetzner Storage Box or any SFTP server) A Storage Box or any server you can reach over SFTP S3-compatible (AWS S3, Hetzner, R2, B2, Wasabi, MinIO) Object storage Proxmox Backup Server A PBS datastore; deduplicated storage NFS export A NAS or file server on your network A directory on each node A second disk in each node (does not survive the loss of the node) -
Fill in the fields of the kind (below).
-
Under Storage, set the Directory for backups (the Path prefix for S3; default
vmstack) and, optionally, an Upload speed limit (MB/s). The limit applies to each upload; restores always run at full speed. PBS has neither field. -
Under Encryption, decide whether to Encrypt backups (see Encryption). This is fixed once the destination holds backups.
-
Save. Then press Check on the destination's row and choose a node to run it from.
On every kind except PBS, backups are laid out the same way: <directory>/<panel id>/<server id>/<UTC time>-<backup id>/ holding disk.qcow2, one disk-<disk id>.qcow2 per extra disk and manifest.json, written last.
The first destination and the default#
The first destination you add becomes the default. To change it, press Make default on another destination's row; a destination can be made default only after it has passed a check. Backups cannot be turned on under Backups › Overview until a default destination exists.
What cannot change later#
While a destination holds backups, the fields that say where it is (host, port and user; bucket, endpoint, prefix and addressing; datastore and namespace; NFS server and export; a local directory; the base directory) are fixed, and encryption cannot be turned on or off. To move backups elsewhere, add a new destination and make it the default. Names, credentials, NFS mount options, TLS settings and speed limits can be changed at any time. A destination that holds backups cannot be deleted.
SFTP and Hetzner Storage Box#
Fields: Host (for a Storage Box u123456.your-storagebox.de), Port (23 by default, as a Storage Box uses), User.
The panel signs in with an Ed25519 key it makes for this destination. The private half is stored encrypted and only sent to a node inside a backup task; it is never written to the node's disk. Password authentication is not used for backups. Under Sign-in, choose How the panel gets in:
Sign in with the password once (the default):
- Press Connect. The panel connects without signing in and lists the server's host keys with their SHA-256 fingerprints. For a Storage Box it links Hetzner's published fingerprints; for another server compare with
ssh-keygen -lfon the server. - Tick These fingerprints are correct.
- Type the account's Password and press Add and install the key. The panel signs in only to a server that shows one of the confirmed keys, adds its public key to
.ssh/authorized_keys(keeping every line already there), signs in again with the key to prove it works, then saves the destination and checks it from a node.
The password is used for that one sign-in and is never stored, logged or put in a task. If the key cannot be installed, nothing is saved.
The key is already installed, or I will install it: the destination is saved and its public key is shown with Copy. Add it to .ssh/authorized_keys on the server yourself. For a Storage Box:
cat key.pub | ssh -p 23 [email protected] install-ssh-key
Then press Check: its first success pins the host key the node saw. No backup is sent to the destination before that.
Host keys and replacing the key#
- If the server later shows a different host key, every backup and check fails until an administrator presses Accept the new host key on the row. Do this only if you know why the key changed (the server was moved or reinstalled).
- Replace key makes a new key pair, installs it with the current key (or with the password, if you type it because the current key no longer works), proves it, and only then removes the old key. It is refused while a backup, restore or delete on the destination runs.
- After a successful check the row shows "Signs in with key SHA256:…"; Show public key reveals it again.
S3-compatible storage#
Fields:
| Field | Notes |
|---|---|
| Provider | AWS S3, Hetzner Object Storage, Cloudflare R2, Backblaze B2, Wasabi, or MinIO or other. Only fills in defaults |
| Endpoint | Scheme, host and port, no path, for example https://fsn1.your-objectstorage.com. Empty for AWS. An http:// endpoint is accepted but sends backups unencrypted: use it only on a private network |
| Region | For example eu-central-1; us-east-1 for MinIO; auto for R2 |
| Bucket | 3 to 63 lower-case letters, digits, dots and hyphens |
| Addressing | Virtual-hosted (bucket.host) or path-style (host/bucket). Default: virtual-hosted without an endpoint, path-style with one. A bucket name with a dot is always path-style |
| Access key ID, Secret access key | The secret is stored encrypted and never shown again |
| Verify the TLS certificate | On by default. For a self-signed MinIO, paste its CA certificate (optional) instead of turning this off |
| Storage class | The bucket's default, STANDARD, STANDARD_IA, ONEZONE_IA, INTELLIGENT_TIERING, GLACIER_IR or another class. GLACIER and DEEP_ARCHIVE are refused: a restore would need a thaw first |
| Server-side encryption | None, SSE-S3 (AES256) or SSE-KMS (with a key ID). Many compatible services refuse these; the check tells you |
Give the nodes a key that is limited to this bucket: every node that runs a backup holds it in memory while it does. The form lists the permissions the key needs: s3:ListBucket, s3:GetBucketLocation, s3:PutObject, s3:GetObject, s3:DeleteObject, s3:AbortMultipartUpload, s3:ListMultipartUploadParts, s3:ListBucketMultipartUploads, and optionally s3:GetBucketVersioning and s3:GetLifecycleConfiguration for the check's warnings.
Check on an S3 destination reaches the bucket, writes, reads, lists and deletes a small test object under <prefix>/.vmstack-check/, and makes a two-part multipart upload. It warns (without failing) when versioning is on (deleted backups keep using space), when no lifecycle rule aborts incomplete multipart uploads, and when the bucket has Object Lock. Add a lifecycle rule that aborts incomplete multipart uploads after one day.
Retention is done by the nodes deleting objects; do not add lifecycle rules that expire backup objects, or the panel will list backups that are gone. To make backups undeletable, see Immutable backups.
Proxmox Backup Server#
The node uses Proxmox's own proxmox-backup-client. Install it on every node that should back up to or restore from a PBS (amd64 nodes on Debian 12 or 13 only) by running the node installer's upgrade with --pbs-client:
curl -fsSL https://panel.example.com/downloads/install.sh | bash -s -- --upgrade --agent-url https://panel.example.com/downloads/vmstack-agent --pbs-client
Use the upgrade command shown on the node's page (it has your panel's addresses and checksums) and add --pbs-client to it; see Node installer options. The option adds Proxmox's pbs-client repository and archive key to apt; a failure is only a warning, and the node then reports PBS as unavailable. The panel offers PBS checks, backups and restores only on nodes that report the client.
Fields: Server, Port (8007), Datastore, Namespace (default vmstack; empty is the datastore's root; the panel adds its own id below it), API token ID in the form user@realm!name, Token secret, and Certificate fingerprint (optional).
On the PBS, give the token the DatastoreBackup role on /datastore/<store>/<namespace>. For the check to create the namespace itself, also give it DatastorePowerUser (or DatastoreAdmin) on the datastore; otherwise create the namespace on the PBS yourself. Schedule garbage collection on the datastore: deleted backups free space only when it runs.
The certificate is pinned. If you leave the fingerprint empty, the first Check only reads the certificate and sends no token. The row then shows the fingerprint: compare it with Dashboard › Show Fingerprint on the PBS (or proxmox-backup-manager cert info) and press Trust this certificate. Pinning is strict: after a certificate renewal (including Let's Encrypt), backups fail until you trust the new fingerprint.
How a PBS destination differs from the others:
- Each backup is one PBS snapshot
vm/<server id>/<time>. PBS stores only chunks it does not already have, so after the first backup of a server only what changed is uploaded. VMStack's own incremental setting does not apply to PBS. - Snapshots on the row lists what the namespace holds, with PBS's own verify state for each.
- Encrypt backups uses PBS's client-side encryption (AES-256-GCM). The key is symmetric, so every node that backs up holds it in memory while it does. The downloaded key file works with
proxmox-backup-client --keyfile. - There is no upload speed limit for PBS.
NFS export#
Fields: NFS server (host name or address), Export (an absolute path such as /srv/backups) and Mount options. Empty options mean vers=4.2,soft,timeo=150,retrans=3: a soft mount, so a server that disappears fails the backup after about 45 seconds instead of hanging. Allowed options are vers, hard, soft, timeo, retrans, proto, port, nconnect, rsize, wsize and sec=sys.
The node mounts the export only for the duration of a backup, restore, delete or check, and unmounts it afterwards. Nodes need nfs-common, which the node installer adds. The export must allow the nodes' addresses read-write access.
A directory on each node#
Each node writes to its own directory, typically a second disk mounted at /backup. A backup there does not survive the loss of the node, and it can be restored only on that node.
Field: Directory on each node, an absolute path. A node accepts it only after you allow the directory on that node, with the installer's --backup-local-root option (repeatable; with --upgrade it adds to the directories already allowed):
curl -fsSL https://panel.example.com/downloads/install.sh | bash -s -- --upgrade --agent-url https://panel.example.com/downloads/vmstack-agent --backup-local-root /backup
Without it the node refuses local destinations ("no local backup directories are allowed on this node"). The directory must already exist; the agent does not create it. The destination's row lists which nodes allow it.
Encryption#
Encrypt backups encrypts every file on the node before it is uploaded. On every kind but PBS it uses age (X25519): a node that backs up only holds the public key, and the secret key is sent to a node only for a restore or verify. The panel keeps the secret key, encrypted with its application key, so restores do not ask for it.
After adding an encrypted destination, press Download encryption key on its row (you type your password again; the download is recorded in the audit log) and store the file outside the panel, for example in a password manager. Without that key, and without the panel's database and application key, encrypted backups cannot be read. When you rebuild a panel against an existing bucket or directory, tick Use an existing key while adding the destination and paste the key.
Encryption is chosen when the destination is added and is fixed while it holds backups. It needs a current agent on every node that backs up there.
Where each server's backups go#
New backups of a server go to the first of these that is set:
- Its plan's destination: Customers › Hosting plans, the plan, section Backups and snapshots, Destination.
- Its node's destination: the node's Maintenance tab.
- The panel's default destination.
Backups already taken stay where they are. The row of each destination shows Used (space and number of backups), Schedules using it and which plans and nodes chose it.
Read-only and paused destinations#
- Make read-only on a row stops new backups to the destination, manual and scheduled. Restores, verifies and deletes by hand still work, and retention deletes nothing from it. Servers whose backups go there get no new backups until it is writable again or they are sent elsewhere.
- Pause schedules stops scheduled backups to the destination, with a reason, until you press Resume schedules. Back up now, restores and deletes go on. See Pausing scheduled backups.
If something fails, see Troubleshooting backups and imports.