Proxmox Backup VE as a virtual machine in a Proxmox VE Cluster

[Heracleos]

Hello everyone,
Sorry if this question has already been asked before.
In your opinion, is it possible to install Proxmox Backup VE in a virtual instance running on the same Proxmox cluster that needs to be backed up, and use an external NAS via NFS or SMB as a repository?
If it is possible, are there any specific drawbacks or things to keep in mind during installation, or any steps to take after installation?
Thanks.

 

[Meister Mopper]

Hello everyone,
Sorry if this question has already been asked before.
In your opinion, is it possible to install Proxmox Backup VE in a virtual instance running on the same Proxmox cluster that needs to be backed up, and use an external NAS via NFS or SMB as a repository?
If it is possible, are there any specific drawbacks or things to keep in mind during installation, or any steps to take after installation?
Thanks.

This is possible, but not recommended.

You must make sure that you back up the PBS directly from the PVE, otherwise you will no longer be able to access the backup in the event of damage.

 

[tcabernoch]

Ok, this is going to be not-so-obvious, but give it a chance.

• The more common method of doing this is to mount the NFS share in the PBS filesystem and then setup a PBS datastore there. If you do this, you'll have to make some file system permission changes, and it will not perform well.

• The less common method, which I actually do employ, is to use the NFS as a Proxmox datastore and host a qcows virtual disk there. Attach that qcows disk to the PBS VM and build your PBS datastore there. This option performs MUCH BETTER than the above method. It still sucks.
  • I've actually had to recover this scenario before when the PVE host had crashed, and I had to rebuild the PBS server. Re-attach the qcow disk to the new PBS, mount it, set it as a datastore, and all the backups came back.

• The very uncommon method, which I employ where the hardware supports it, is to run the PBS server as a guest of the NAS itself. TrueNAS (at a couple sites) or a DIY linux NAS (yes I built one out of scraps) can host a VM. PBS doesn't need a whole lot of resources. Putting the VM physically on top of the storage is far more advantageous than any hit you take for minimal resources on the NAS host.

Oh ... and ...

• Probably the most uncommon method is iSCSI to your NAS. That performs very well, but requires specific hardware, network, and skillset that leaves it as an unattractive option for many admins. I don't iSCSI. (And as soon as I declare that I won't learn something, it's forced on me. I'm building an iSCSI-attached-storage bare-metal PBS right now. I will map it to the ZFS dataset that is the current PBS's drive, and hope to migrate in one swell foop.)

 

[pmarasse]

And probably the very most uncommon method I'm testing at work : 2x identical NAS, with 2x 10GbE network cards. each NAS is viewed as a iSCSI target.

PBS is using those targets with multipathing + ZFS Mirror. It needs some tweaking in order to start in the right way.

 

[Heracleos]

the less common method, which I actually do employ, is to use the NFS as a Proxmox datastore and host a qcows virtual disk there. Attach that qcows disk to the PBS VM and build your PBS datastore there. This option performs MUCH BETTER than the above method. It still sucks.

• I've actually had to recover this scenario before when the PVE host had crashed, and I had to rebuild the PBS server. Re-attach the qcow disk to the new PBS, mount it, set it as a datastore, and all the backups came back.

Well, this scenario seems to me perhaps the most suitable one.
Very similar to what I'm seeing around with Veeam on a vmware instance.
Where the virtual disks in the vm where veeam is running are mapped as vmware nfs datastores.
Clearly, as you said, if the PVE cluster goes down, the PBS vm must be reconfigured and restarted in another node or pve cluster proxmox, to recover the other VMs.
However, the case that a whole PVE cluster goes down, seems quite unlikely to me...or does it ? Is it really that common ?

 

[tcabernoch]

the case that a whole PVE cluster goes down

Um ... nah, we didn't connect there.

First off, I don't have HA and clustered storage.
(Jebus it kills me to admit this in public ... and I might delete it later.)

My "PBS hosted on PVE" is usually 2 disks in different spots.
- a 50gb root disk hosted on local-zfs.
- a multiple TB data disk hosted on a NFS datastore

I guess the root disk could live on the NAS too.
The VM would run slower that way.
I think a lot of the stuff it does is in memory, so I'm not sure if it would matter.
It would make recovery easier, but I think recovery is easy anyway, and I want the speed granted from the root disk running locally.

I'd like to add that I use these virtual PBS systems as capacity storage and sync hubs.
That's the design. They are for doing things that are really expensive with physical hardware.
In front of these virtual hubs and doing the backups are other PBS systems on real physical hardware with as much SSD as I had on hand.
But reality inevitably intrudes. I've filled up my physical primary backup servers and had to rely on these virtual ones.
They work.
They don't even suck that bad if you build it right. Enable fleecing in the backup jobs.

 

[Heracleos]

First off, I don't have HA and clustered storage.....

A ok, now it's clear. If you don't have a cluster, it is clear that you are dependent on the PVE server.
My case is different. Mine is a whole 3-node PVE cluster with ceph as storage for vm.
The PBS would be an instance of the same cluster, but the storage would be on the nas.
I think that in my case, the probability of PBS unreachability is very low, and it can happen only in the case of down of the whole cluster, which seems unlikely to me.
Thanks for your clarifications and advice

 

[tcabernoch]

It's an architecture of necessity. No, it's not recommended. It's gonna be used tho ...

... heh. and thx.

 

[VictorSTS]

A ok, now it's clear. If you don't have a cluster, it is clear that you are dependent on the PVE server.
My case is different. Mine is a whole 3-node PVE cluster with ceph as storage for vm.
The PBS would be an instance of the same cluster, but the storage would be on the nas.
I think that in my case, the probability of PBS unreachability is very low, and it can happen only in the case of down of the whole cluster, which seems unlikely to me.
Thanks for your clarifications and advice

Make vzdump backups of the PBS VM to a Samba or NFS share on that same NAS that hosts you datastore. When the cluster breaks or any other disaster, just install PVE anywhere, restore the backup of the VM and you'll get access to your backups. Still, I would install PBS anywhere else but inside the same cluster it is protecting.

And probably the very most uncommon method I'm testing at work : 2x identical NAS, with 2x 10GbE network cards. each NAS is viewed as a iSCSI target.

PBS is using those targets with multipathing + ZFS Mirror. It needs some tweaking in order to start in the right way.

Done that in labs. Besides being completely unsupported by ZFS (must have full access to the drives), PBS performance will suffer due to network latency. And in case of network issues your datasets will become suspended and you will have to hard reboot and possibly have to repair ZFS. It is not reliable enough for me.