Dual server share cephfs

[andmattia]

Hi

i try to understend if is a good idea to have 2 PBS server (in my case in different datacenter) with shared cephfs. In case one of fault of one PBS (es pbs hosted on same node of VMs) I can use second instance to restore VM.

Is it correct? Is there other config/option to do the same failover?

 

[UdoB]

i try to understend if is a good idea to have 2 PBS server (in my case in different datacenter) with shared cephfs.

From my viewpoint PBS and Cephfs are two orthogonal topics. Both can be discussed completely independent from each other.

Yes, having multiple backups and at least one of them being off-site is a good idea. The usual approach in PVE/PBS context is to setup a second (or third) PBS in "the other" datacenter and to "sync" some (or all) backups. See https://pbs.proxmox.com/docs/managing-remotes.html#managing-remotes-sync

In case one of fault of one PBS (es pbs hosted on same node of VMs) I can use second instance to restore VM.

Yes. Just establish "Storage --> Add --> Proxmox Backup Server" to reach out for the remote one. For additional security I recommend to add a "restore-only" user-account on the PBS first and use it for this connection. This way the backups on the remote PBS can not get deleted from the PVE cluster, because nobody has the right to do so. Under some specific circumstances this artificial limitation might be very welcome...

 

[guruevi]

With the new PBS you can also use S3 as a backend. We are interested in trying the Ceph S3 implementation for that in the near future (currently working on getting quotes for 14PBe worth of Ceph) and then tiering off that with rules to Deep Glacier.

There are many ways of getting effective redundant backup storage. I would still treat each “site” as an independent entity and use the built in features of PBS to either make backups to multiple targets (interleaved) or send to a backup site then backup the backup site. Proper backup is 321 (at least) 3 copies in 2 media with 1 offsite.

 

[andmattia]

With the new PBS you can also use S3 as a backend. We are interested in trying the Ceph S3 implementation for that in the near future (currently working on getting quotes for 14PBe worth of Ceph) and then tiering off that with rules to Deep Glacier.

There are many ways of getting effective redundant backup storage. I would still treat each “site” as an independent entity and use the built in features of PBS to either make backups to multiple targets (interleaved) or send to a backup site then backup the backup site. Proper backup is 321 (at least) 3 copies in 2 media with 1 offsite.

I upgrade my PBS node last week and make a simple test on S3 (based on OVH Object Storage) and it works fine.

 

[andmattia]

From my viewpoint PBS and Cephfs are two orthogonal topics. Both can be discussed completely independent from each other.

Yes, having multiple backups and at least one of them being off-site is a good idea. The usual approach in PVE/PBS context is to setup a second (or third) PBS in "the other" datacenter and to "sync" some (or all) backups. See https://pbs.proxmox.com/docs/managing-remotes.html#managing-remotes-sync


Yes. Just establish "Storage --> Add --> Proxmox Backup Server" to reach out for the remote one. For additional security I recommend to add a "restore-only" user-account on the PBS first and use it for this connection. This way the backups on the remote PBS can not get deleted from the PVE cluster, because nobody has the right to do so. Under some specific circumstances this artificial limitation might be very welcome...

I have another doubt about Ceph. What is the benefit of doing a backup on Ceph via PBS rather than on the Proxmox VE node?

 

[UdoB]

I have another doubt about Ceph. What is the benefit of doing a backup on Ceph via PBS rather than on the Proxmox VE node?

In my Homelab(!) I would NOT do backup to Ceph. (In my $dayjob this is a different story: I am offered storage without knowing the backend technology.) I would backup data from Ceph to somewhere else. In my small world Ceph is by far too precious to be used for a stupid backup dump.

Note: I do NOT use Ceph currently at all - but I did try: my conclusion

What I want to have is a chance to restore data when the PVE node dies. For a single node (without Cluster) this means the backup software ("PBS" in itself) and the data (the .chunks-store) needs to be on separate hardware. For other constellations this is recommended too. I my Homelab I have multiple completely independent PBS' with local storage - built from obsolete server hardware. This way I have as few dependencies on other sub-systems as possible. KISS is still a valuable concept - and Ceph is not compatible with that one.