BackupZit – open-source (AGPLv3) backup server with Proxmox VE support: instant recovery, replicas, boot tests

S

stefanzeljkic

New Member
Oct 26, 2026
2
3
3
Hello everyone,

Disclosure: I am the developer of BackupZit, a free and open-source (AGPLv3) self-hosted backup server. I'd like to share it here and ask for feedback from Proxmox users.

How it works with Proxmox VE 7 and 8:
  • The BackupZit agent (Debian package) runs on each node and uses the node's own tools (qm, pct, pvesm, LVM, ZFS, Ceph). Nothing is installed in the guests, and no API token or password is stored in the console.
  • For each VM/CT a temporary snapshot is taken (LVM-thin, ZFS, Ceph RBD, qcow2 on directory/NFS/CIFS); with the QEMU guest agent the file systems are frozen. Disks are read in 1 MiB blocks, deduplicated and encrypted on the node; only blocks not yet in the repository are uploaded.
  • Restore as a new guest (next free ID) or with the original ID, on any node with the agent, also in another cluster; a new vmgenid is set on every restore.
  • Instant recovery: the VM starts from the backup (read-only NBD export + qcow2 overlays on a directory storage) and is moved to the final storage with qm disk move while running.
  • Optional replicas: a stopped copy of each guest on another node is updated after every backup.
  • Optional boot test: each VM of the newest backup is started isolated (network disconnected) and must answer through the QEMU guest agent.
  • Single files from VM disks (NTFS, ext2/3/4, XFS, also in LVM) can be browsed and downloaded.

Known limits: Proxmox VE 9 is not on the supported list yet (7 and 8 are tested); container disks on ZFS subvolumes are not supported yet; raw disks on directory/NFS/CIFS storage and thick LVM can only be backed up when the guest is stopped; in a cluster each node needs its own agent and job.

The same console also handles Windows/Linux machines, VMware, Hyper-V, databases and Microsoft 365, which is the main reason it exists — for Proxmox-only setups PBS is of course the natural choice.

Step-by-step guide: https://backupzit.com/guides/proxmox-vm-backup/
Source code and releases: https://github.com/Max-Zit/backupzit

I'm grateful for any bug reports or suggestions.

(Transparency: I used AI tools while writing parts of the code; everything is reviewed and tested by me on a Proxmox lab.)

Stefan
 
  • Like
Reactions: FrankList80 and Gilberto Ferreira
stefanzeljkic said:
Known limits: Proxmox VE 9 is not on the supported list yet (7 and 8 are tested); container disks on ZFS subvolumes are not supported yet; raw disks on directory/NFS/CIFS storage and thick LVM can only be backed up when the guest is stopped; in a cluster each node needs its own agent and job.
Click to expand...

What's the point in using a backup software which only supports EOL releases of the software? PVE8 is EOL since August 2026, PVE7 even longer: https://forum.proxmox.com/threads/proxmox-ve-support-lifecycle.35755/

Even worse: Why should I pay money for this if I only can run it in non-supported versions (https://backupzit.com/support/ )?

Does it have any features compared to the native backup features of ProxmoxVE and ProxmoxBackupServer or etablished enterprise backup software like Veeam?

Right now I wouldn't use (or even pay money for it) this for any backups of produktion workloads if development was done with AI and any testing in your lab (which can mean all from an used mini-PC to a "home datacenter" larger than many smes IT department)
 
  • Like
Reactions: UdoB and Gilberto Ferreira
Bare-metal disk image backup is the missing feature in PBS.
So there is use-case.
Once you've used forever-incremental backups like PBS (or even Restic), it's hard to accept the merge times and disk usage of Veeam and similar apps.
 
Last edited:
(Slighty derailing the conversation) Congratulations for your deliverable. It looks very promising. In General - some anecdotal perception: i see here A LOT of extension of the whole PVE/PBS ecosystem driven by LLMs which imo solve specific problems which is highly appreciated. For Backup: i'am a fanboy of PBS, VEEAM and STORWARE. A lot must happen, before i switch to anything else (No offense to the "Content" creator!) - this is imo a big topic in the AI-age - who to trust with support and maintenance for 3-5 years. P.S: To the other commenters - we would should not fall in the trap of denying credit to new "Content" creators, be it AI/LLM driven or not, no?
 
  • Like
Reactions: stefanzeljkic
What LLM-driven development has done is made it easier for people who would otherwise lack the time or skills to do these kinds of projects. And as per usual the vast majority of them will be abandoned because the person behind it lacks the time or skills to maintain it and/or the paid version didn't bring in enough to justify the effort. That will be even more of an issue now with projects getting blasted by LLM-based bug reports.

Don't get me wrong though. I'm not saying this particular project will die on the vine, just that most of them will based on history. It will just be a faster cycle than in the past because of LLM's.
 
  • Like
Reactions: Johannes S and FrankList80
Thanks for all the replies, honestly some of it stung a bit but you were right.

@Johannes S @Gilberto Ferreira yeah, PVE 8 went EOL in August and shipping something that only did 7/8 was a mistake on my side. 0.43.4 is out now with PVE 9 support: https://github.com/Max-Zit/backupzit/releases/tag/v0.43.4

I upgraded my test box from 8.4 to 9.2 (Debian 13, kernel 7.0, LVM-thin) and ran everything again. Backups of running VMs and CTs, restore as new VM and new CT, instant recovery, and the auto restore test that boots the VM from the backup and checks the data. All passed. 7 and 8 still work if you haven't upgraded yet, but from now on 9 is what I build against. If you buy support for a version Proxmox itself dropped, it's best effort only.

Also added a how-to for backing up the host itself (/etc/pve, network, storage, firewall, the cluster db) and rebuilding a dead node from it.

About PBS vs Veeam: if you only run Proxmox, just use PBS, it's proven and I'm not trying to replace it. BackupZit makes sense when you've got a mix, like some Proxmox, some VMware or Hyper-V, a few physical boxes, SQL/Postgres/MySQL, M365, and you want all of it in one place. @_gabriel bare metal images of physical machines is exactly the thing PBS doesn't do, I wrote up how it works here: https://backupzit.com/guides/windows-bare-metal-restore/ . And it's forever incremental like PBS, dedup chunks, every backup is a full restore point, nothing to merge.

Testing: fair question. My lab is a physical PVE 9 host with Windows Server, Debian VMs and some containers, an ESXi 8 + vCenter, and a Hyper-V box with Win11, Server 2012 R2 and Ubuntu. Targets are SFTP, SMB, MinIO and Azurite. Every release goes through the automated tests plus a manual round of backups and restores on all of that, and releases are signed. That's not the same as years in production, I know. Only way to close that gap is people actually trying it and telling me what breaks.

@FrankList80 @BobhWasatch get it, nobody should trust a backup tool just because someone says so. Can't prove where it'll be in 5 years. The code is AGPLv3 so it doesn't disappear whatever happens, and I'd rather show it with regular releases than talk about it.

Stefan
 
  • Like
Reactions: FrankList80
You must log in or register to reply here.
Top Bottom
Back