Back up Proxmox VE (Virtual Environment) virtual machines (VMs) and LXC (Linux Containers) through built-in backup jobs or a Proxmox Backup Server. Pick a third-party option like Bacula Enterprise when broader backup requirements justify adopting a dedicated platform. Make sure to protect recovery keys and host configuration separately before testing application recovery from an isolated environment against your data loss and recovery time objectives.
There are three different methods of backing up your Proxmox VE, with the first two overlapping significantly. Built-in backups, launched with a web interface or via the vzdump command, write data to the storage of a user’s choice, be it a local disk or an NFS share. Proxmox Backup Server (PBS), distributed separately from Proxmox, attaches to it as a storage destination for those same jobs and stores backups incrementally with deduplication. Enterprise backup software such as Bacula Enterprise always remains another option to choose from, protecting Proxmox environments together with other storage types in the same environment while allowing its management under the same collection of policies.
Picking a specific option largely comes down to just a few questions:
- How fast do you need to recover?
- How much data can you afford to lose?
- Is everything you work with stored in Proxmox?
- How much time does your team have to operate backup infrastructure?
The sections below walk you through each of these approaches one step at a time, followed by an example of how to test a restore and a detailed overview of what each option would cost to implement on a three-year timeline.
Which approach should you choose?
| Approach | Best when | Main trade-off | Cost basis |
| Built-in backups to file or NFS storage | Small setups with simple recovery needs and no wish to run another server | Each backup is a full archive, with no deduplication | Included with Proxmox VE |
| Built-in backup jobs + PBS | Proxmox-focused environments that need frequent, efficient backups they can verify | An extra server to maintain and secure | Free and open source; optional paid subscription |
| Enterprise backup platform | Mixed environments, or any setup where backup and recovery require central management | More to deploy and learn; commercial license | Commercial license (see the cost section) |
How to back up Proxmox using built-in tools
Proxmox VE has a number of internal features for creating backups of virtual machines (VMs) or containers. Each backup contains the guest’s configuration and all included disk data, making every restore point complete on its own. File storage keeps a separate Full archive for each one, while PBS stores the same data incrementally and deduplicated. You can easily start a backup yourself from the web interface, using the vzdump command, or by relying on a scheduled backup job.
A snapshot is not a backup
A snapshot is a short-term rollback point, not a backup. Proxmox VE can snapshot VMs to return them to their earlier states after a failed update. That snapshot would sit on the same storage as the VM’s disks, meaning that the storage being lost means that both the data and its snapshots are lost. Snapshots are not included in backups, either – a backup writes to a separate storage that could survive the loss of the storage holding the guest itself. As such, use snapshots for rollback and backups for recovery.
That said, the “snapshot” backup mode described below is not the same as a snapshot of a Proxmox environment. It is simply a backup variation describing how vzdump approaches a running guest VM to create a backup.
Backup modes
Snapshot mode is the one set by default; it’s used for creating backups of running VMs with the lowest downtime but has a small risk of inconsistency due to the guest being able to keep writing during the backup process. By using the QEMU guest agent with your VM, vzdump can ask it to freeze and thaw the filesystem around the backup, improving consistency. A filesystem freeze doesn’t make the application itself consistent, so databases might need their own handling of this topic.
Stop mode will simply shut the VM down before doing a backup, offering the most consistent results at the cost of downtime. Suspend mode also remains for compatibility purposes, suspending VMs before running the snapshot mode, although this one isn’t recommended anymore.
Fleecing is something that could only be applied to VMs, caching data that the guest is about to overwrite in a fleecing image instead of sending it directly to the backup target. It can reduce stalls in the guest, so its storage should be fast and local, with discard support and thin provisioning; storage without thin provisioning could simply reserve the full disk size instead.
In containers, Stop mode halts a container for the entire backup, leading to potentially long outage periods. Suspend mode allows data copying from a running container via rsync that suspends it briefly to copy the changes, requiring extra temporary storage. Snapshot mode briefly suspends the container for consistency and then backs up a snapshot of its volumes but requires every volume that’s being backed up to sit in storage that supports snapshots.
Only a container’s root disk is included by default. Extra volume mount points are backed up only when explicitly told to do so, while bind or device mounts are never included in backups.
Running a manual backup
You can back up an individual VM’s or container’s data manually using the Backup tab on the Proxmox web interface. The Backup now button opens a dialog where you’re free to choose the target storage and how the backup itself is made:
- Select the VM or container and open its Backup tab.
- Click Backup now.
- Choose the target storage, then set the mode and compression.
- (optional) Mark the backup as protected so it can’t be removed via Proxmox VE; add notes; adjust notification settings.
- Click Backup and wait for the task to finish without errors.
Compression and backup filenames
vzdump compresses backups using:
- gzip
- lzo
- zstd
The file extension shows which one was used: .gz or .tgz, .lzo or .zst. According to the Proxmox documentation, zstd is currently the fastest of the three and supports multi-threading. The pigz tool could also replace gzip as a multi-threaded drop-in, as both of them can configure thread counts. A file having none of these extensions would mean it wasn’t compressed by vzdump at all. File names also usually cover guest types, guest IDs, dates, and times, like this: vzdump-lxc-105-2009_10_09-11_04_43.tar.
Scheduling backup jobs
Regular backups run as jobs under Datacenter > Backup in the web interface. A backup job names the guests it covers and the target storage, applying to every node in the cluster or to just one of them.
Job schedules use a format that’s close to how systemd calendar events look; here’s an example: mon..fri 02:00 for every weekday at two in the morning. If a host was offline whenever a job was due, the Repeat missed option would run it again afterwards.
Jobs provide the same settings as manual backups, plus options that limit the load of the backup process. A bandwidth limit exists to cap I/O in KiB/s, while the thread counts for zstd or pigz limit CPU use during compression. The Advanced tab here also holds the job’s fleecing option. Notes can use template variables like {{cluster}}, {{node}}, {{vmid}}, and {{guestname}}, so every backup is labeled automatically.
Backup commands
The backup command examples below use a storage named backup-nfs, which you’ll be replacing with your own storage’s name.
# Snapshot-mode backup of guest 777 with zstd compression
vzdump 777 –storage backup-nfs –mode snapshot –compress zstd
# Back up every guest except 101 and 102
vzdump –all –exclude 101,102 –storage backup-nfs
# Back up guest 777 and keep its last 3 backups, 13 daily ones, and 9 yearly ones
vzdump 777 –storage backup-nfs –prune-backups keep-last=3,keep-daily=13,keep-yearly=9
Defaults for every run can be configured by modifying the /etc/vzdump.conf file, using one OPTION: value pair for each new line. That said, options given on the command line override the ones saved here. In this file, the total bandwidth limit is counted in KiB/s, turning a number like 51200 into about 50 MiB/s:
storage: backup-nfs
mode: snapshot
bwlimit: 51200
The –mailto option keeps working still but is considered deprecated already, being replaced by notification targets and matchers.
Retention
Retention rules decide which backups are kept and which aren’t. When there are no retention rules, nothing gets deleted, as the default position is to keep all. Without retention rules, a repository that keeps receiving backups might exhaust its capacity eventually.
The Keep Last option is applied before weekly and yearly options, with each one considering only the backups that are older than those already kept by the previous ones. Weeks in the system run from Monday to Sunday, and only the most recent backup is kept when several backups fall in the same period.
Protected backups are excluded from pruning and don’t count toward the totals. This would block ordinary pruning and deletion from Proxmox VE, but it won’t make the data immutable against someone with administrative access to the storage. Proxmox also recommends setting retention on the storage itself instead of on a per-job basis, with its documentation linking a prune simulator for testing a ruleset before starting to use it.
Restoring a VM or container
In the web interface, open the backup in the storage’s Backups view and click Restore. Depending on the guest type, you’ll be using either qmrestore to recover VMs or pct restore to recover containers:
# Restore a VM backup as VM 601 on local-lvm, with new MAC addresses, and start it
qmrestore /mnt/backup/dump/vzdump-qemu-888-2026_11_04-02_00_01.vma.zst 601 –storage local-lvm –unique –start
# Restore a container (CT) backup as CT 600
pct restore 600 /mnt/backup/dump/vzdump-lxc-777-2026_11_04-02_00_01.tar.zst –storage local-lvm
Restoring to the ID of an existing guest is refused by default. Add –force only when you intend to overwrite that guest. Restores aren’t usually throttled by default and can oversaturate your storage or network during the process. To combat that to some degree, there’s –bwlimit to cap the read speed of a single restore (measured in KiB/s), and there’s also pvesm set STORAGEID –bwlimit restore=KiBs to set a default speed for all writes to this storage.
The full list of options can be found in the official vzdump, qmrestore, and Proxmox VE backup documentation.
How to back up Proxmox with Proxmox Backup Server
Proxmox Backup Server (PBS) is a free and open-source backup solution from the creators of Proxmox VE, released under the AGPLv3 (GNU Affero General Public License version 3). Proxmox recommends running it on a dedicated host. It is added as a storage destination to your existing Proxmox environment so your existing VM and container backups can write to it.
PBS stores backups incrementally and deduplicated: once the first full backup is complete, only the changed data is sent and stored by PBS, while keeping identical blocks only once. Each restore point still restores as a complete guest. The PBS client can even back up physical hosts. A subscription is optional and only adds access to the enterprise repository and vendor-level support.
What PBS changes compared with built-in backups
Compared to built-in backups, PBS has different ways to transfer, store, protect, and clean up backup data. The table below shows what exactly the specific features of the solution mean in day-to-day operations:
| Feature | What it does | What it means in practice |
| Incremental backups | PBS avoids uploading chunks already present; how much source data must be read depends on the workload and change-detection mode. VM backups track changed blocks with dirty bitmaps. Container and file backups compare file metadata when the metadata change detection mode is used. | Shorter backup windows, less network traffic. The first backup takes the most time. In legacy mode, file backups still read every file each time but upload only new or changed chunks. |
| Deduplication | Identical data blocks are only stored once, irrespective of how many backups use them. | Many restore points require far less storage space compared to the same number of full copies. Backups sharing blocks could lead to damage to a single block affecting several backups at once. |
| Client-side encryption | The Proxmox VE host encrypts information before sending it, but only when encryption is enabled for the storage. | The PBS server stores the content in encrypted form if encryption is turned on; lost key = lost backups. |
| Pruning | Applies retention rules and removes the metadata of expired snapshots. | Removes snapshot metadata, eligible unreferenced data chunks are deleted by garbage collection later. |
| Garbage collection | Deletes data blocks that no remaining backup references and that are outside the grace period. | Frees up some storage space. Runs on a schedule and is I/O intensive. |
Prepare the PBS server and its storage
A standard PBS server requires a modern 64-bit CPU with 4+ cores, along with 4+ GiB of RAM (+1 GiB for each TiB of backup storage) and generally fast local storage. Enterprise SSDs are advised by the official documentation. Spinning disks also work with a metadata cache, such as a ZFS special device mirror. It’s important to remember that a PBS virtual machine inside the cluster it protects will go down with that cluster, so using a separate machine is recommended.
Backups are stored in a certain directory called “datastore”, supported by ext4, xfs, or ZFS. The commands presented below prepare a disk and create a datastore on it, with the first one being used to identify the correct disk to use:
proxmox-backup-manager disk list
proxmox-backup-manager disk initialize sdX
proxmox-backup-manager disk fs create store1 –disk sdX –filesystem ext4 –add-datastore true
Initializing and formatting a disk destroys its existing data, so run disk list first to confirm that sdX is the unused disk.
The last command of the three creates an ext4 filesystem with a datastore named store1 located at /mnt/datastore/store1. The same tasks could be done via the web interface too, under Administration > Storage/Disks. Additionally, if you want to use an already existing directory, for example /backup/disk1/, then you should run this command instead:
proxmox-backup-manager datastore create store1 /backup/disk1/store1
Create a backup user and add PBS to Proxmox VE
Make sure to give Proxmox VE its own PBS account specifically with the DatastoreBackup role: it will be able to create and restore its own backups but won’t have any prune capabilities, keeping the topic of retention under the control of the PBS:
# On the PBS server
proxmox-backup-manager user create archiver@pbs
proxmox-backup-manager acl update /datastore/store1 DatastoreBackup –auth-id archiver@pbs
proxmox-backup-manager cert info
# On the Proxmox VE host
pvesm add pbs pbs-main –server 192.0.2.10 –datastore store1 –username archiver@pbs –fingerprint <fingerprint> –password
Set a password for the new user in the PBS web interface before copying the fingerprint from the cert info output. The username must also include the realm, as in archiver@pbs. However, the fingerprint will only be required when the PBS certificate is self-signed or otherwise untrusted. The Proxmox documentation shows –password at the end of this command, so double-check how your version takes the password, or add the storage in the web interface under Datacenter > Storage.
acl update sets an entry in the access control list (ACL) at its first use.
Schedule backups to PBS
Backups move to PBS through the same jobs the built-in method uses. Create a job under Datacenter > Backup, choose the PBS storage as the target, and don’t forget to select the guests. The usual option for most running guests is the snapshot mode.
The initial backup to a datastore uploads everything there is, while future backups only upload new or changed chunks. If you’re working with containers, there’s also a nice additional option for PBS targets: the metadata change detection mode can reuse information from the previous backup to immediately skip files that weren’t changed since then. VM backups aren’t affected by this setting.
vzdump 123 –storage pbs-main –pbs-change-detection-mode metadata
Retention, pruning, and garbage collection
Pruning applies your retention rules to data and removes the metadata of expired snapshots, while the actual data gets reclaimed by the garbage collector (GC) feature: it removes the data blocks that haven’t been used by any of the remaining backups for some time and aren’t referenced by any of them. The baseline cutoff point for garbage collection is 24 hours and 5 minutes, or the oldest running backup writer instance. Space is therefore freed later than the pruning. Remember that GC takes a lot of I/O resources, making the quiet hours the most reasonable time to run garbage collection.
Since the backup account doesn’t have pruning privileges, you’ll have to configure retention rules using the PBS datastore. Luckily, it runs the same keep-last to keep-yearly options as Proxmox VE. Its web interface even has a dedicated prune simulator to test any rule set before applying it to production.
The official Proxmox documentation retention example for daily backups kept for ten years straight uses: keep-last 3, keep-daily 13, keep-weekly 8, keep-monthly 11, and keep-yearly 9. It recommends a higher retention than the minimal requirement of your environment.
There’s a dedicated Prune & GC tab in the datastore that can be used to set up schedules for both pruning and GC, if necessary. Make sure to check the datastore’s usage regularly: a full datastore cannot accept new backups.
Verification and notifications
A successful backup job can only show that the data was written, and practically nothing about whether a restore will work to begin with. PBS can re-read stored backups and compare them with their checksums to do exactly that: detect damaged blocks and bit rot.
Verify jobs run on a schedule and can be configured to skip snapshots that were verified before at some point – or it can re-check older ones after a certain period.
It’s a best practice from the PBS documentation to re-verify all backups regularly, for example once a month, conducting daily or hourly verifications for new backups in the meantime. A whole data store, a group, and even a single snapshot can also be verified manually
PBS sends results of scheduled tasks like verification or garbage collection via notification targets and matchers. Set up a target that points to a mailbox people actually read, along with a matcher that at least forwards errors. Some older setups might also still use the legacy per-datastore setting that has three options to choose from: Always/Errors/Never.
Verification’s supposed to confirm that stored data is still intact, but whether that VM would actually function when restored isn’t something that’s within this tool’s capabilities. The restore test described later in this guide answers that question.
Encryption and key management
Proxmox VE can provide AES-256 backup encryption in GCM mode (Galois/Counter Mode) before sending that data to PBS, making sure the server only stores encrypted data. Encryption itself isn’t mandatory and can be set on a per-storage basis where necessary. There’s a dedicated setting called encryption-key that generates a key automatically and stores it in Proxmox VE’s /etc/pve/priv/storage/<STORAGE-ID>.enc, but only keys without a password are stored there.
You can also create a master key pair. Each backup would then carry a copy of its key encrypted with the master public key, allowing the data to be recovered via the master private key even if the original key is lost.
Encrypted backups aren’t readable without the key. Keeping a paper key or its copy in a password manager is recommended – it has to be stored anywhere outside the production environment, so that you don’t lose the backup when its key gets lost or compromised. Here’s how to produce a paper key:
proxmox-backup-client key paperkey /etc/pve/priv/storage/<STORAGE-ID>.enc –output-format text > qrkey.txt
For safety reasons, we recommend testing a restore with the stored copy of the key instead of the one that’s still in the host.
How do you restore a VM or container from PBS?
Restoring a VM or container from PBS is easy:
- Open the PBS storage in the Proxmox VE web interface
- Select the backup
- Click Restore
The Restore dialog would allow you to either keep the original guest ID or pick a new one, along with choosing the target storage for restoration. All the bandwidth limitations from the built-in method also apply here.
The account you use for the storage with Proxmox VE will determine what you can restore. The account that has a DatastoreBackup role will only restore backups that were created under that same account. A dedicated account with the DatastoreReader role, on the other hand, can inspect the contents of and restore backups within the datastore or namespace scope granted to that account (Proxmox Backup 4.2.9-1 documentation), making it a great pick for a dedicated recovery account.
For disaster recovery tasks that involve moving to a new host, install Proxmox VE, add the same PBS storage with its fingerprint, and supply the encryption key. Then and only then you can restore the guests and check dependencies such as network bridges and storage names on the new host before relying on the result.
What are live restore and single-file restore in PBS?
There are two options that are PBS backup-specific here.
The first is called live restore, it starts a VM right away and keeps copying the data in the background. It resumes the service sooner, but comes at a cost of any reads being slower until the copying process finishes. Additionally, the entire VM would be left in an undefined state if the restore process fails mid-way.
Single-file restore uses the File Restore button in the storage view, opening a file browser for a backup to offer the user the ability to download individual files from a backup without restoring the entire guest. If the backup in question covers a VM, PBS starts a temporary recovery environment to access the backed-up files and stop it as soon as the copying is complete or after a timeout.
Back up the Proxmox host configuration
Guest backups only cover VMs and containers. A rebuilt host would need a fresh Proxmox VE installation, with all the network definitions and storage definitions, as well as its membership (in a cluster). In a cluster, the configurations under /etc/pve are shared between nodes, making this reinstallment possible.
A single-node environment would have no second copy to work with, so make sure to back up its configuration ahead of time. The PBS client can store host backups:
proxmox-backup-client backup etc.pxar:/etc –include-dev /etc/pve –backup-type host –repository archiver@pbs@192.0.2.10:store1
Due to /etc/pve being its own separate mount, we had to include the –include-dev /etc/pve option here, as well. Backing up /etc is partial preparation and not a full host backup, while the archive can still contain secrets such as keys and credentials, so apply appropriate security measures. A host backup will only be useful in combination with a written rebuild guide that records:
- Disk layouts;
- Network definitions;
- Storage definitions;
- Subscription keys;
- Encryption key location.
You can test that guide by rebuilding a node in a lab environment.
Six mistakes that can make Proxmox backups unusable
Backup failures often occur well in advance of the point in time where you are forced to restore. Here are six common errors with a failure case and an easy fix for each one. While we’ll be using Proxmox Backup Server (PBS) as our backup target in our examples, the concepts can be transferred to all other types of backup targets.
Mistake #1: The backup system depends on the production it protects
Backups that share a common failure point with production will inevitably fail together with it. Imagine a PBS server running a VM on the cluster it backs up, with its datastore being on the same storage array. Issues like a storage failure or a power loss take out the VMs and their backups in a single move, making restoration unavailable exactly when it’s needed the most.
Create a list of every item the backup system needs to run: hosts, storage, network, power, name resolution. Every item the backup shares with production is a shared point of failure. Try to run PBS on dedicated hardware with its own datastore disks, ideally in a separate rack or room. Address this hardware by IP address or a local hosts entry so that DNS outage won’t be a problem for your restore either. The installation media and the rebuild notes should always be within an arm’s reach to be ready for production going down at any point in time.
Mistake #2: Production and backup administration share credentials or excessive privileges
An intruder that manages to obtain a domain administrator account will be looking for the backup server as one of his top priorities. If PBS uses the same account, the backups can be deleted before production is extracted or encrypted, transforming a single point of compromise into a total loss. To fix that, go through the following controls in order:
- Create dedicated, named administrator accounts in PBS’s own user realm, with no link to Active Directory to prevent a stolen domain account from opening anything on the backup server.
- Enable two-factor authentication (WebAuthn or TOTP) for people and use API tokens for automation purposes – as the PBS documentation recommends.
- Let Proxmox VE connect with an account holding only the DatastoreBackup role, which should be able to create and restore its own backups without any pruning privileges attached.
- Perform pruning on PBS itself as a means of preventing a compromised Proxmox VE host from using its account to delete backup data.
- Place PBS in a segmented network zone and allow access to port 8007, which serves the web interface, the API, and backup clients, only from the Proxmox VE hosts or designated administration workstations.
Mistake #3: There’s no suitably isolated recovery copy
Imagine that you have a business that stores its second backup copy on a NAS in another site, mounted using the same administrator credentials as the primary backup. Any intruder who gets these credentials will be able to delete both copies in the same session. The copy itself was off-site, but it was still reachable and deletable. To avoid this, try to classify every copy by the isolation it actually offers:
| Term | Meaning | Examples with PBS |
| Off-site | Stored in another location; may still be reachable over the network with valid credentials | A second PBS at another site, fed by sync jobs |
| Offline | Stored on a medium disconnected from any information system | Tape, or a removable datastore that’s unplugged after the copy |
| Immutable | Cannot be changed or deleted for a set period by a mechanism in the storage; the strength depends on the mechanism | The PBS documentation describes no immutability option for datastores (as of November 2026) |
ANSSI, the French national cybersecurity agency, recommends in its IT Backup Fundamentals guide (ANSSI-BP-100, version 1.1) three copies on two media with one offline, and deems an offline solution more robust than an online WORM (write once, read many) one.
With PBS, tape and removable datastores are the offline copies. To get an off-site copy on a second PBS server, create the sync job as a pull job on that server, using a remote credential that can only read the source. This way, the production side holds no credentials for the destination. You can also leave remove-vanished off by default to ensure that deletions at the source don’t propagate. If you choose to push instead, make sure to grant the remote credential the backup rights only.
Mistake #4: The encryption keys are unavailable during recovery
If the only copy of the encryption key is on the Proxmox VE cluster that was just destroyed, the backups it encrypts cannot be read. A password manager hosted on the same cluster would fail in the same way.
Store the key in at least two separate locations where a production failure won’t reach, like a printed paper key in a physical safe and a corporate password manager hosted elsewhere. Also create a master key pair: a master public key goes into the storage configuration, so each backup could carry a copy of its encryption key that only the master private key can decode. Meanwhile, the master private key itself stays offline.
Decide in advance who can retrieve the key and keep track of where each copy’s stored. The encryption section explained earlier in this guide shows how such keys could be created.
Mistake #5: Retention, capacity, verification, or alerts are misconfigured
An example: a PBS datastore with pruning scheduled but no garbage collection. The disk fills up over time, new backups start failing, and the failure emails appear in a root mailbox nobody bothers to open. Two weeks later you realize that your most recent restore point is, you guessed it, two weeks old. You can still solve this but you’ll have to go through all of these in no particular order:
- Check that garbage collection’s schedule follows pruning, as pruning itself only removes snapshot metadata.
- Make sure retention also covers the time it takes to notice an issue. Proxmox has its own ransomware guidance advising users to keep backups for longer periods, as dormant malware could’ve already infected the recent data copies.
- Verify jobs also have to have a schedule, with the monthly all-encompassing verification being strongly suggested for everyone.
- Notifications must lead to a mailbox someone actually opens regularly.
- Datastore usage in general should be monitored from time to time to be able to react accordingly to a filling disk instead of being surprised by it.
Mistake #6: Successful jobs are mistaken for proven recoverability
A successfully finished job only shows that the data was written somewhere. Consider a case where nightly jobs were reported as successful for over a year before turning up an incident where the main database server takes far longer to restore than the business can tolerate – or simply boots with an application that’s incapable of finding its data. Nobody would know the source of the issue because nobody tried to restore said data.
Schedule restore tests and keep track of all their results in detail. Restore into an isolated network or to a new guest without overwriting the existing one. Make sure that the application in question works and record how long the entire process took you to go through.
How do you test a Proxmox restoration?
Recover a representative workload from a backup into an isolated environment. After that, make sure the application still works, and the data is there. You’ll also want to compare the time it took to recover and the age of the recovery point with your business objectives.
A restore test illustrates the fact that a backup will actually bring a workload back within certain outage and data loss limits that the business accepts. These tests are more effective when run on a schedule a company can maintain, like once per quarter + after every major change, such as data migration or encryption key change.
What is the difference between backup verification and a recovery test?
Verification checks whether the stored data is currently intact. In PBS, verification jobs read backup data and compare their checksums with the stored values to ensure that no data got damaged since the backup was written.
A recovery test verifies whether the business can run from the backup data: the restore completes, the VM boots, the application starts, and the data inside it is correct.
Both are needed in a workload, as the restore can still fail even if the data is intact because of reasons that verification alone can’t detect, like missing encryption keys or dependencies like authentication being unavailable. The two can be planned to run automatically, but application acceptance criteria and test results would still need a review.
Restore test checklist
A restore test contains six specific steps in this order:
- Write down the targets: recovery time objectives (RTO) signify the longest outage the business accepts for this workload, and recovery point objectives (RPO) represent the largest acceptable data loss, measured in time.
- Pick a workload that matters, like a database server with dependencies, and the recovery point you would actually use after the simulated incident; try to rotate the workloads and recovery points between tests.
- Restore data to a new guest ID, leaving the production guest untouched, but don’t start the restore guest yet. Rely only on what you’d use in a disaster: the stored encryption key copy, the recovery account, etc. Follow your own written procedures for disaster mitigation.
- Connect the newly restored VM to either a bridge or VLAN (virtual local area network) that has no access to production, then start it. If this step is absent, the restored copy can still reach production systems or duplicate addresses.
- Verify that the application works and the data itself is present by signing in as a user and opening recent records; make sure that all services start and all dependencies (authentication, name resolution) are still reachable.
- Record the results. Calculate the difference between the simulated incident time and the recovery-point timestamp before comparing it with the RPO. Also check whether the backup schedule would be able to meet that objective with a high degree of consistency. A four-hour RPO would fail in an organization with nightly backups, as nightly backups mean up to 24 hours of data loss in case of emergency. Take note of all missing credentials or keys, along with any undocumented dependencies; assign an owner and a date to each corrective action.
Which command restores a backup for a test?
For a backup on file storage, the command presented below restores data to the new ID 9001. The –unique option can generate new MAC (media access control) addresses for the network devices to avoid address conflicts. However, it doesn’t isolate the application itself, so step 4 is still very much needed. If you’re performing a backup that’s stored on PBS, use the web interface instead and make sure the guest isn’t started after the restore.
qmrestore /mnt/backup/dump/vzdump-qemu-888-2026_11_04-02_00_01.vma.zst 9001 –storage local-lvm –unique
What should a restore test results table record?
Record one row per test. The first row of information provided below is an illustrative example and cannot be considered a measured result. It compares the data loss at the simulated incident with the RPO and the measured recovery time with the RTO.
| Test date and
workload |
Incident and recovery point | Data loss vs RPO | Recovery time vs RTO | Findings and corrective actions | ||
| 5 Nov 2026, accounting database VM | Simulated incident 09:00; recovery point 02:00 the same day | 7 h; RPO 4 h not met | Restore 1 h 40 min (09:30 to 11:10); total outage 2 h 10 min; RTO 4 h met | The restore key existed only on the production host. A copy was added to the safe, and the backup schedule was changed to every four hours. | ||
Built-in backups vs PBS vs Bacula Enterprise
There are a few similarities and differences that could be noted among built-in Proxmox backups, PBS, and third-party software like Bacula Enterprise.
The information below compares them across eight factors that contribute to the software choice, using each vendor’s official documentation as of November 2026. Built-in backups and PBS often work together, so the first two columns could be treated as two stages of the same workflow.
Protection and recovery compared
The first comparison table highlights what each option protects and how it approaches data restoration. Container coverage and restore granularity usually separate the options here.
| Criterion | Built-in backups | PBS | Bacula Enterprise |
| Suitable environment and use case | Small Proxmox-only setups with basic recovery needs. | Proxmox-focused environments in need of efficient backups they can verify. | Mixed environments, or any setup in need of centralized backup and recovery management. |
| VM and container protection | Full backups of VMs and containers from Proxmox VE, in different modes. | The same guests, stored incrementally and deduplicated. The PBS client can also back up physical hosts. | The Proxmox plugin protects VMs and LXC containers, while the QEMU plugin only protects running VMs. |
| Recovery options and application consistency | Restore a whole guest. Filesystem freezing can improve consistency; application-consistent recovery depends on the guest application and its backup preparation. | Restore a whole guest, live restore, or single files from the web interface. Filesystem freezing can improve consistency; application-consistent recovery depends on the guest application and its backup preparation. | Guest images restore as a new VM or as archives in a directory. File-level restore needs a File Daemon inside the guest. The documentation describes snapshot-based reads without saying how guest filesystems are quiesced. |
| Coverage beyond Proxmox | Proxmox VE guests only. | Proxmox VE guests and file-level backups of physical hosts | Other hypervisors, databases, Microsoft 365, physical servers – all in the same product. |
Management, storage, and licensing compared
| Criterion | Built-in backups | PBS | Bacula Enterprise |
| Centralized management | Backup jobs for a whole cluster under Datacenter > Backup | One web interface per PBS server. Several Proxmox VE clusters can back up to one server and stay separate through namespaces. | One Director and catalog manage jobs and reports for all workloads, with the BWeb Management Suite web interface. |
| Storage destinations and isolated copies | File-level storage that Proxmox VE supports, or PBS. Isolation depends on the storage you choose. | Local or removable datastores, tape, S3-compatible object storage, and sync to another PBS. The PBS documentation describes no immutability option for datastores. | Disk, tape, and cloud object storage with deduplication. Immutability comes from provider locks like S3 Object Lock, set on the destination storage. Bacula’s retention must be longer than the lock. |
| Operational requirements and support | No extra components. Support depends on your Proxmox VE subscription or community resources. | A separate server to install, secure, and maintain. Subscription support tiers are optional. | A Director, storage daemon, and catalog database, along with the File Daemon and the plugin on the hypervisor host. |
| Licensing basis | Included with Proxmox VE. | Free and open source (AGPLv3), with optional per-server subscription: €560 a year (Community, non-production), and from €1,120 for production (Basic) up to €4,480 (Premium). | A subscription priced per endpoint. Prices are on a per-request basis. |
PBS subscriptions are priced on a per-backup server basis, regardless of how much data they protect or how many guests there are. Bacula scales from the number of endpoints in its subscription tiers. PBS and Bacula can both write to tape. Bacula also documents provider-side immutability for cloud storage, which the PBS documentation does not describe for datastores.
For the wider range of workloads that Bacula covers, see its enterprise virtual machine backup page.
Which Bacula Enterprise plugins support Proxmox, and with what limits?
A lot of the version, plugin, and configuration dependencies here concern Bacula Enterprise with its two plugins that cover Proxmox support. Built-in backups and PBS depend mostly on the Proxmox VE and PBS releases you use.
Bacula uses two separate plugins with their own goals. The Proxmox plugin secures VMs in any state, along with LXC containers, with the limitation being its restriction to only Full-level backups (as the Proxmox API doesn’t provide the means of supporting other backup types). There’s also the QEMU plugin, the one that adds incremental backups for running VMs but cannot back up containers. This last plugin requires Bacula File Daemon and the plugin installed on the Proxmox host to function properly.
Make sure to double-check version support before committing, as the QEMU plugin page claims that only Proxmox VE 6.4-8.2 are fully tested and supported. In the meantime, the Proxmox plugin page doesn’t mention any versions at all, so ask Bacula about relevant support capabilities.
Bacula’s Proxmox backup and recovery page describes the plugins and their requirements.
How does Bacula Enterprise handle backup immutability?
Immutability in Bacula’s backups comes from the lock features of a cloud provider. Most of these are configured on the destination storage using the provider’s own tools. Bacula’s data immutability documentation mentions several options as supported:
- AWS S3 Object Lock
- Azure Blob Immutable Storage
- Google Cloud Bucket Lock
- Oracle Cloud Retention Rules
It even requires Bacula’s retention for the affected volumes to be longer than the retention that was configured in the cloud.
S3 Object Lock secures specific object versions for their retention period. A new version or a delete marker could change the current object without deleting the protected version in the process.
In governance mode, users who have the bypass permission can override the lock themselves. In compliance mode, there’s no user in the system that can delete the protected data, not even the root user.
Bacula’s page also offers a note stating that S3 doesn’t have a strict WORM policy, so users would still be able to delete or overwrite the current object, and even the protected version can be retrieved manually.
Per-volume locking, the one that uses the Storage Device Volume Immutability and Minimum Volume Protection Time directives, would only work with the AWS cloud driver. Governance is the default setting for its LockMode.
When built-in backups or PBS are enough
Built-in backups or PBS would be enough in cases where everything that needs to be protected runs exclusively on Proxmox VE, with whole-guest and file-level restores fitting within your recovery targets and your team being fine with operating another Linux server alongside all the existing ones. The deciding factors would be your recovery objectives, application requirements, permissions, reporting, and who would operate the backups. The number of hosts decides little here.
Built-in backups to file storage would only fit hosts with very basic recovery needs. After all, every single one of its backups is a full archive, leading to fast storage growth as time goes on. PBS can avoid that by storing backups incrementally while deduplicating them, adding scheduled verification and client-side encryption on the host on top of everything. It can even write offline copies to tape or removable datastores if necessary, and its documented restore options cover whole guests, live restore, and single file restores.
The test for both isn’t that complicated. Run the restore test covered earlier in the article, and see if it meets your recovery time/data loss targets. If it does, PBS may meet your Proxmox-focused environment’s requirements. Look for alternatives when wider operational or recovery requirements remain. The checklist at the end of this guide could help you by listing them.
Other vendors
Multiple third-party vendors provide documented support for Proxmox VE. The examples below are taken from each vendor’s own pages as of November 2026:
- Veeam Backup & Replication supports Proxmox VE 8.2 to 9.2 on x86-64 hosts installed from the official ISO, with Proxmox support introduced in version 12.2. There are no mentions of containers in its user guides.
- NAKIVO Backup & Replication offers native, agentless Proxmox VE backup, with Proxmox itself announcing NAKIVO as an official Solution Provider back in March 2026.
- Acronis lists agentless backup capabilities for Windows and Linux VMs and containers on Proxmox VE 8.x and 9; a lightweight agent runs on the Proxmox host.
- Commvault supports full, incremental, and synthetic full backups of Proxmox VE VMs from Platform Release 2024E (11.36) CU40. Its documentation covers VM support but has no mention of containers and doesn’t include specific Proxmox VE versions.
- Vinchin Backup & Recovery offers protection to QEMU/KVM VMs on Proxmox VE 7.x to 9.x without agents inside the guests, supporting features like incremental backup and instant VM recovery while licensing on a per-host model. Container coverage should be confirmed ahead of time, as per their own FAQs.
Make sure to check any shortlisted product against the criteria we listed in the tables before – especially something like container coverage and the exact versions of Proxmox that are supported.
What does Proxmox backup actually cost?
More often than not, the cost of a backup license is only one component of the total cost of a Proxmox backup. Infrastructure, extra copies, staff time, and tools for other workloads all contribute to the total value, so subscription comparison alone gives an incomplete picture.
This section compares the total cost of ownership over three years for two illustrative scenarios. In it, we used official list prices where they exist and labelled assumptions where they don’t. All amounts are in euros, exclude VAT, and reflect list pricing as of November 2026.
What does a Proxmox backup budget include?
The total cost of a Proxmox backup setup combines five separate cost categories. The table below lists them all and what to count in each one of them.
| Cost category | What to include | Source of figures here |
| Software and support | Subscription or licensing basis, support entitlement, required components | Official price list (PBS), quote required (Bacula) |
| Infrastructure | Backup servers, storage, networking, a second location | Assumptions that you replace with your own quotes |
| Additional copies | Cloud storage, requests, retrieval, transfer, offline media | One published price list, as an example |
| Operations | Setup, monitoring, maintenance, upgrades, recovery testing | Hours, stated as assumptions |
| Wider coverage | Additional tools for workloads outside Proxmox | A vendor quote for the tool you choose |
Which assumptions do both cost scenarios share?
Both scenarios we’ll be going through use a lot of the same protection requirements, with every figure below being an illustrative assumption:
- Scenario A is a Proxmox-only environment that has 3 hosts, 25 VMs, and roughly 5 TB of source data, with an average VM being about 200 GB. Source data implies the data used inside the guests, making the allocated disk capacity bigger.
- Scenario B works off of that baseline and adds 10 endpoints outside it, with 5 physical servers and 5 VMs on a different hypervisor, to a grand total of 35 protected endpoints. A vendor could also count the 3 Proxmox hosts as endpoints, too, which is a question for the quote.
- Both scenarios use nightly backups and keep 7 daily, 4 weekly, and 6 monthly restoration points, bringing the total to 17 per workload at steady state. PBS deduplication means not all of these points need a full copy, while the prune simulator shows the actual calendar.
- One additional copy is kept at a second location in both cases. In scenario A, it’ll be a second PBS server, and the cloud example later in this section illustrates object storage pricing only.
- The horizon for calculations is three years, and all prices are listed without VAT.
How much do PBS subscriptions cost?
The PBS software itself is free, while its optional subscription services are priced on a per-server, per-year basis. Every tier includes all features, with unlimited backup storage and unlimited clients, and access to the enterprise repository. Support performance is where these tiers differ the most:
| Tier | Per server per year | Three years | Support |
| Community | €560 | €1,680 | No ticket support; labeled for non-production use |
| Basic | €1,120 | €3,360 | 5 tickets a year; response time within 1 business day |
| Standard | €2,240 | €6,720 | 15 tickets a year, response time within 4 hours, support for SSH remote |
| Premium | €4,480 | €13,440 | Unlimited tickets, 2 hours of response time for critical requests, 24/7 support from 19 October 2026 |
Response times apply under the coverage and severity conditions on Proxmox’s pricing page. The Community tier claims it’s only available for non-production use, leading to this section using the Basic tier for €1,120 per year. Subscriptions themselves are scaled on a per-server basis, meaning a second PBS server on a different production site would need its own subscription if you want to support it.
What does a three-host, Proxmox-only backup example include?
A Proxmox-only scenario considers two PBS servers with Basic subscription, one in each location, with the second one tasked with receiving and storing synced copies of the first. There are 3 hosts, 25 VMs, and about 5 TB of source data:
| Item | Basis | Three-year cost |
| PBS support subscriptions | 2 servers x €1,120 (Basic) x 3 years | €6,720 (the PBS software itself is free) |
| Backup servers and storage | Capacity from the sizing worksheet in the next section, plus your hardware quote | Your quote |
| Network and second-site hosting | Your own costs | Your costs |
| Staff time | Illustrative assumption of 216 hours: 24 for setup, 4 a month for monitoring and maintenance over 36 months, and four 4-hour test restores a year | 216 hours x your hourly rate |
Three-year cost = €6,720 + servers and storage + site costs + (216 hours x hourly rate).
How do you size PBS storage for a cost estimate?
Backup storage estimations require six inputs, all of which should be measured in your own environment:
- The used data inside the guests, in TB.
- The data that changes each day, in GB.
- The compression and deduplication ratios, which a pilot datastore would show.
- The retention rules and the restore-point calendar they produce, checked with the prune simulator.
- The expected growth over three years.
- The headroom you keep free for garbage collection + verification + growth.
Using a simplified planning model, we can assume that:
PBS capacity = (the first backup after compression and deduplication + the changed data kept across the retention period after deduplication) x growth x headroom
Built-in backups tend to keep a full archive of every restore point, making their capacity roughly equal to the compressed size of one full backup x the number of restore points x growth x headroom.
Compare both models with your own inputs included. This guide also doesn’t cover storage savings, as the result varies significantly depending on how much data changes.
What does an additional cloud copy cost?
The cost of a second copy going to object storage instead of a second PBS server would differ depending on storage volume and the number of requests, retrievals, and transfers. As a published example, Hetzner Object Storage charges a base price of €6.49 a 31-day month without VAT for 1 TB of storage and 1 TB of outgoing traffic (as of November 2026). After that, it charges €0.0087 per TB-hour for additional storage, resulting in about €6.47 per TB per month.
Therefore, an 8 TB copy (used here purely as an example size) would cost you about €51.80 a month, or about €1,865 over 36 months. Requests are not charged separately.
Restores beyond the original 1 TB of traffic cost extra. Hetzner’s page makes the extra-traffic storage cost unclear, so make sure to confirm it before budgeting. The page also uses TB without defining it, even though PBS sizing uses TiB, so remember to convert before comparing.
This example illustrates storage pricing only and isn’t a replacement architecture. It prices capacity only. A PBS datastore on S3 storage would still need a PBS server with a local persistent cache, other providers may charge for requests and egress, and the example doesn’t establish that PBS works with this particular provider or that the data is immutable.
What does a mixed environment with 35 endpoints include?
Scenario B protects the same 25 Proxmox VMs and 10 endpoints elsewhere on top of that: 2 database servers, 2 file servers, 1 domain controller on physical hardware, and 5 application VMs on another hypervisor. The requirements for recovery also differ: file-level restore is needed for the file servers, app-aware recovery is expected for the databases, system recovery – for the domain controller, and image-level restores – for the other VMs.
With PBS, the Proxmox part would have the same cost as in the previous scenario. The PBS client backs up files on physical hosts, which could cover the file servers (check with the operating systems it supports), but the databases, the domain controller, and the other VMs would all need whatever their requirements demand.
Application-aware or image-level recovery may need a dedicated tool priced by its own vendor. A second system set up to administer, monitor, and test would also add additional hours that were not covered by the previous scenario.
With Bacula Enterprise, subscriptions are priced per-endpoint or per-core without charging for data volume. Every subscription also includes unlimited technical support, the response speed of which will vary depending on the chosen tier. Bacula Systems publishes no price list, so the exact cost would need a quote.
Do not assume that each Proxmox VM equals one licensed endpoint, make sure to ask how exactly VMs, containers, and hosts are counted.
What information does a Bacula Enterprise quote need?
The pricing of Bacula Enterprise is provided on a per-quote basis, with a comparable quote requiring the following information:
- The number of endpoints by type, with a grand total of 35 in this scenario, and whether the Proxmox hosts count as endpoints.
- The number of Proxmox hosts and whether there’s a need for containers or incremental backups (this will decide between the Proxmox and QEMU plugins).
- The required support response time.
- The retention policy, the nightly schedule, and the roughly 5 TB of Proxmox data, which size the infrastructure, even though the license itself doesn’t depend on data volume.
- The storage targets, such as disk, tape, or cloud object storage.
- Whether an immutable copy will be needed.
- The applications on the 10 non-Proxmox endpoints, be it databases or other types that might need dedicated plugins.
How do you compare the total costs?
Compare the options on the same total, like subscriptions, licenses, infrastructure, additional copies, staff hours, and tools for other workloads – all for the duration of three years. Keep the protection requirements identical, meaning the workloads, retention, recovery objectives, and copies.
In a Proxmox-only environment, PBS subscription costs are easy to predict, as they scale with the number of servers and not the number of VMs or the data volume. As such, most of the remaining costs belong to infrastructure and time.
If you’re focusing on a mixed environment, the comparison would be based on the two numbers that we couldn’t provide: the price of the extra tools needed for PBS, and Bacula’s quote. Estimate infrastructure and staff time separately for each architecture, as they differ in the number of platforms, consoles, and restore tests you’ll have to run. Add them to each option’s licensing figures and compare in three-year totals.
When should you evaluate Bacula Enterprise?
The moment your backup requirements go beyond what PBS and built-in backups cover for Proxmox, you should start evaluating solutions like Bacula Enterprise. Built-in options can cover Proxmox well enough for both VMs and containers, and it’s enough for many environments. The checklist below highlights the requirements that could justify considering an enterprise backup platform.
When is Bacula Enterprise worth evaluating for Proxmox?
Consider evaluating Bacula Enterprise if several of the statements below can apply to your business environment:
- Proxmox is only one platform among others, like hypervisors, physical servers, or databases, and you want one policy engine and one catalog for all of them.
- You need centralized backup management across several sites or teams.
- Your recovery plan includes several kinds of storage targets under the same retention policy, like disk, tape, or cloud.
- You seek to create an additional copy in cloud object storage with provider-side object locks that’s written by the backup itself.
- You prefer per-endpoint licensing with no data volume charges and one support contract for all workloads.
Built-in options or PBS are probably enough for you if none of these apply to your business. In that case, going through the cost comparison above and conducting appropriate restore tests would be more beneficial for you than picking a completely new product for adoption.
Bacula’s Proxmox plugins have documented limits covered earlier in this guide.
Where can you learn more about Bacula Enterprise for Proxmox?
To compare Bacula Enterprise with your current setup, discuss your Proxmox backup architecture with Bacula Systems. Useful details worth bringing up include the number of Proxmox hosts and other workloads, your retention, your storage destinations, and your RTO/RPO. For background, see Bacula Enterprise and the page on Proxmox backup and recovery.
Proxmox backup FAQ
Is a Proxmox snapshot a backup?
No. A snapshot lives on the same storage as the VM it belongs to, so if that storage fails, both are gone. Consider it as a quick way to roll back a change, and rely on a backup stored somewhere else for actual recovery.
Is Proxmox Backup Server free to use?
You can download and use PBS without paying, because the software is open-source under the AGPLv3 license. A subscription is optional and adds support and access to the enterprise repository.
How can I make Proxmox backups harder for ransomware to destroy?
Provide the backup server with its own accounts, and make sure the backup creator account doesn’t have the permissions to delete them. Keep at least one more data copy that an attacker with production access won’t reach, so it’ll be either offline or protected by an object lock. A restore test would confirm the surviving copy works.
Can Bacula Enterprise protect LXC containers on Proxmox?
Yes, via its Proxmox plugin that handles both VMs and containers but can only run Full backups. The QEMU plugin, on the other hand, supports Incremental backups but only for those VMs that are currently running. The support for specific versions of Proxmox VE might vary by plugin, so it’s always a good idea to double-check that with Bacula Systems when you can.
Can I restore a Proxmox backup onto a different host?
Usually yes, provided the new node has what the guest depends on. Nothing prevents a backup file or a PBS datastore from being restored to any Proxmox VE node that has access to it. You’ll have to add the same datastore to the new node first, if you’re using PBS, though.
Make sure to check four things before beginning:
- that the target VMID is free or you’ll be creating a new one;
- that the target storage exists on the new node;
- that the bridges the guest used are defined there;
- that the new hardware supports the guest’s configuration (like the CPU type).
If a dependency is missing, the guests would simply start without a functioning network access or even fail to start at all.
Hi,
Does Bacula Enterprise Proxmox Backup support change block tracking backups / incremental or differentials? Or are all backups for Proxmox still full backup copies?
Kind Regards,
Arno
Hello Arno,
Our `QEMU` plugin supports Full and Incremental Proxmox VM backups (Differential backups of VMs are not supported), but it
cannot backup LXC containers on Proxmox. Our `Proxmox` plugin can only do Full backups of Proxmox VMs, but it also supports backups of LXC containers on Proxmox.