From Proxmox VE
Jump to: navigation, search


With the Proxmox VE ZFS replication manager (pve-zsync) you can synchronize your virtual machine (virtual disks and VM configuration) or directory stored on ZFS between two servers. By synchronizing, you have a full copy of your virtual machine on the second host and you can start your virtual machines on the second server (in case of data loss on the first server).

By default, the tool is syncing every 15 minutes, but the synchronization interval is fully configurable via the integrated cron job setup. The configuration of pve-zsync can be done either on the source server or on the target server.

This is useful for advanced backup strategies.

Note: pve-zsync is introduced for Proxmox VE 3.4 as technology preview. The package can be installed on plain Debian Wheezy or Jessie or Strech servers too, as long as ZFS is configured.

Note: Our ZFS is configured to automount all subvols so keep in mind if you use pve-szync and also zvol will be scanned by lvm.

Main features

  • Speed limiter
  • Syncing interval can be set by cron
  • Syncing VM (disks and config) but also ZFS Datasets
  • Can keep multiple backups
  • Can be used in both directions
  • Can send on local host
  • Traffic is encrypted


  • not possible to sync recursive
  • only ssh for transfer
  • email notification is done by cron
  • Name resolution not taken into account, you have to use IP Addresses, even using hosts file hostnames will not work

System requirements

  • Both, target and source server must support ZFS (best practice: use Proxmox VE hosts).
  • SSH must be installed and configured
  • to receive email-notifications, a working mail server is required (e.g. postfix).
  • cstream
  • perl
  • scp
  • JASON.pm (libjson-perl this is the Debian package name, e.g. needed on Wheezy)

PVE Storage Replication and PVE-zsync

PVE Storage Replication and PVE-zsync work completely independent and do not harm each other, as long the following requirements are followed:

  • The destination pool / subset are different.
  • You do not migrate a guest to an other node.

Sum up of the differences:

Characteristic PVE Storage Replication PVE-zsync
Replication Cluster Wide to every Node which meets the requirements
Operation mode push push or pull
Management GUI and Command line Command line only
Keep snapshot no yes
Migration yes no
Main Goal redundancy offsite backup

Migrate form pve-zsync to Storage Replication

When you have a cluster and would like to switch to Storage Replication, you have to destroy the pve-zsync job, clean up the storage and create a new Storage Replication job.

pve-zsync destroy <vmid> [--jobname <test>]

Then you have to cleanup the zfs storage.

Start at the source side and remove all snapshots which start with '@rep_'

zfs destroy <pool>/[<path>/]vm-<VMID>-<type>-<number>@<rep_snapshots> 

Then destroy all guest datasets on the destination side.

zfs destroy -R vm-<vmid>-*-<DiskNO>

Now you can create a new Storage Replication job More information about see Storage Replication

Configuration and use

Install the package with apt, on your Proxmox VE host:

apt-get install pve-zsync

This tool need basically no configuration. On the the first usage, when you create a job with a unknown host, the tool will ask you about the password for the remote server.

Sync a VM or ZFS dataset one time

(N.B. this is also possible if a recurring job for that VM already exists, here you must have in mind that the naming in --name must be the same).

root@zfs1:~# pve-zsync sync --source 100 --dest --verbose --maxsnap 2 --name test1 --limit 512

This command sync VM 100 witch is located on the server where the tool is called and sent it to the server on the zpool tank which has an subset backup. The --maxsnap tells that it should be kept 2 backups, if there are more then 2 backups, the 3 one will be erased (sorted by creation time). Name is only needed if there is already a sync job. The --limit param sets the speed limit what is used for syncing, here it would be 512 KBytes/s.

Create a recurring sync job

root@zfs2:~# pve-zsync create --source --dest tank/backup --verbose --maxsnap 2 --name test1 --limit 512 --skip

The --skip parameter disables the initial sync, that normally would be done immediately but can take a while, depending on the size of the backup. The initial sync will be done at the first sync time.

Delete a recurring sync job

if you delete a job, the former backup data will not be erased only the config will be erased.

root@zfs2:~# pve-zsync destroy --source --name test1

Name is not necessary if it is default.

Pause a sync job

If you want to pause a job, say maintenance the source server!

root@zfs2:~# pve-zsync disable --source --name test1

Reactivate a sync job

to reactivate a job, because the job was paused or the job failed use.

root@zfs2:~# pve-zsync enable --source --name test1

This will reset the error flag in case of failure.

Changing parameters

You can edit the configuration in /etc/cron.d/pve-zsync or destroy the job and create it new.

Information about the jobs

To get a overview about all jobs use:

root@zfs94:~# pve-zsync list
SOURCE                   NAME           STATE LAST SYNC           TYPE        testing1       ok     2015-05-13_14:44:00 ssh testing1       syncing2015-05-13_14:44:11 ssh  

root@zfs94:~# pve-zsync status
SOURCE                   NAME           STATUS        testing1       ok testing1       syncing 

Recovering an VM

You must recover the VM or dataset manually. (In one of the upcoming releases, the restore for Proxmox VE VM will be integrated).

Fist, stop the sync job for the VM or dataset in question.

NOTE: if not you can be interfere the sync job or your snapshot will removed before you are able to send it.

root@zfs2:~# pve-zsync disable --source --name test
root@zfs2:~# pve-zsync list
SOURCE                   NAME           STATE  LAST SYNC           TYPE        test            stopped2015-06-12_11:03:01 ssh  

Then you can send the VM or Dataset to the selected target. SSH is only needed if you send to a remote sever.

zfs send <pool>/[<path>/]vm-<VMID>-disk-<number>@<last_snapshot> | [ssh root@<destination>] zfs receive <pool>/<path>/vm-<VMID>-disk-<number>

If you have a VM you must also copy the config and you need to correct the virtual disk storage configuration accordingly.

cp /var/lib/pve-zsync/<VMID>.conf.rep_<JOB_NAME><VMID>_<TIMESTAMP> /etc/pve/qemu-server/<VMID>.conf

NOTE: On PVE 4.2+ the path is: /var/lib/pve-zsync/data/<VMID>.conf.rep_<JOB_NAME><VMID>_<TIMESTAMP>

Example restore VM 100 with 2 disk from (pve2) to (pve1) and change VMID from 100 to 200:

root@zfs2:~# zfs send rpool/backup/vm-100-disk-1@rep_test100_2015-06-12_11:03:01 | ssh root@ zfs receive vm/vm-200-disk-1
root@zfs2:~# ssh root@
root@zfs1:~# cp /var/lib/pve-zsync/100.conf.rep_test100_2015-06-11_14:11:01 /etc/pve/qemu-server/200.conf
root@zfs1:~# nano /etc/pve/qemu-server/200.conf

Now you have to change the storage path:

bootdisk: virtio0
cores: 1
memory: 512
name: Debian8min
net0: virtio=12:5E:F6:59:A9:BB,bridge=vmbr0
numa: 0
ostype: l26
smbios1: uuid=11fa2fba-5670-4610-aabb-534ad7edeffe
sockets: 1
virtio0: zfs:vm-100-disk-1,size=10G
virtio1: zfs:vm-100-disk-2,size=10G


bootdisk: virtio0
cores: 1
memory: 512
name: Debian8min
net0: virtio=12:5E:F6:59:A9:BB,bridge=vmbr0
numa: 0
ostype: l26
smbios1: uuid=11fa2fba-5670-4610-aabb-534ad7edeffe
sockets: 1
virtio0: vm:vm-200-disk-1,size=10G
virtio1: vm:vm-200-disk-2,size=10G


Don't forget the command only works with IP Addresses and not hostnames.

Job status is on error and data-set can't be erased on destination system

If you have problems with a sync-job and when you try to erase the destination zvol, you get the error "zfs data-set is busy", then lvm could be the problem.

This can be occur if you sync zvols with a lvm on it.

In this case please insert the following line in /etc/lvm/lvm.conf

filter = [ "r|/dev/zd*|" ]

and reboot the system.


Video Tutorials


  • as of 2016-11 If you migrate a vm that is used by pve-zsync
edit /etc/cron,d/pve-zsync and change the IP address for vm .
run 'pve-zsync enable --source ____ --name ____ ' or else there will be a warning when the job runs.