Multiple PBS with one storage

[StephaneG]

Hi,
My backup storage is a distant NFS.
To prevent high network usage (proxmox --> PBS --> NFS on each nodes), I've installed a pbs server on each node, then each pbs server will backup its own node and not the other.

The point is, is there's a way to share the same storage for each pbs ? If I try to add the same storage on a second pbs server, it fails as the .chunks already exist (understandable).

I've created a dedicated directory for each node and it's working fine, but is there's a better way to achieve my goal ?
The best would be to be able to have a pbs cluster sharing the same .chunks/storage of course, but seems impossible actually.

Remotes doesn't seem an option as it would duplicate all data on all pbs storage.

Is my way is a good one ? what would be the better/best way ?

 

[fiona]

Hi,
I'm afraid PBS is not designed to literally share its datastore with other instances. If there never is any metadata conflict (which probably doesn't happen if all instances live in the same cluster) then there's still the question whether locking would work out (when the whole datastore is locked, it should work, but I think not all locks are that coarse).

In fact, it's recommended to use local storage for your PBS instance. If you don't want to replace your NFS with a PBS and if the local traffic PVE -> PBS really is such a big deal, I don't really see a better solution either.

 

[StephaneG]

Thanks for the answer, yes local traffic is an issue (1G local network on a 21 node pmx cluster, against 10G dédicated card for backups on each nodes). It'll change in the futur but for the moment got to do my best with what I got.

Last question : In case of failure of pve/pbs server 1, will I be able to restore the backup on pve/pbs 2 ? (I'm testing differents ways right now)

 

[fiona]

Thanks for the answer, yes local traffic is an issue (1G local network on a 21 node pmx cluster, against 10G dédicated card for backups on each nodes). It'll change in the futur but for the moment got to do my best with what I got.

Last question : In case of failure of pve/pbs server 1, will I be able to restore the backup on pve/pbs 2 ? (I'm testing differents ways right now)

Ah no, that won't be possible without some manual interaction. Either you'll have to make the PBS1 temporarily available to PVE2 and it would end up using the local network for the restore. Or (also not very nice) you stop both PBS instances, exchange their datastores and start them up again.

Also it's not possible to have deduplication between the instances (as that happens on a datastore level), so after migration a VM you'd always have to take a full backup again.

If you can in any way install the PBS close to the NFS (and have a fast connection between those two), that would be a much better solution. Or maybe you can come up with a way of only having a few PBS instances and each one serving several nodes (e.g. if it's likely to restore VM from node A to node B at some point, A and B should share the same PBS instance). IMHO that'd also be better than one instance for each PVE.