Cluster1 -> PBS -> Cluster2

[DmitriiSkvortsov]

Good day team! We have this config:
Cluster1 (5 nodes) = C1
PBS in another datacenter = PBS
Cluster2 (3 powerfull nodes) in another datacenter. = C2

C2 maked like hot-reserv for C1.

All conected with 40Gig line. Every night from C1 ~100VM (~4TB) goes to PBS of cource in incremental "durty" way.

PROBLEM: On C2 I can restore all VMs from PBS with "qmrestore --forse", but it can only delete VM and create it again. It take 2 Hour for all VMs. And not good for Ceph.

QUESTION: How make restore process from PBS to C2 in incremental way (write changes only for last day) without overwriting whole VM.

Thank you=)=)=)

 

[dcsapak]

hmm.. there is no "incremental restore" implemented, also i don't think we could do that in a good way, since we have no way of knowing what exactly changed (there is no guarantee that the target wasn't started and the disk content changed)

 

[DmitriiSkvortsov]

hmm.. there is no "incremental restore" implemented, also i don't think we could do that in a good way, since we have no way of knowing what exactly changed (there is no guarantee that the target wasn't started and the disk content changed)

Okay, thank you for fast answer.

Maybe you have some varian, with which we can sync C1 with C2 without PBS? I can try to make it on operation system level, but maybe it is some ready solution?

Thank you

 

[dcsapak]

if you use zfs as vm storage you can use replication with pve-zsync (https://pve.proxmox.com/wiki/PVE-zsync)
otherwise you can omit restoring the vms at all and use the 'live-restore' (see https://pve.proxmox.com/wiki/Backup_and_Restore#vzdump_restore ) feature when you need it
this will start the vm immediatly and restore in the background, prioritising blocks that the vm currently tries to read/write