Incremental restore using pbs ?

[DJP]

Hello,

Do you think it’s possible to use PBS to do incremental restore ?

Like so :
- First full backup to PBS on site1 on vm1
- Restore full backup from PBS on site2 on vm1-bis. Do not start VM.
- incremental backup to PBS on site1 on vm1 using dirty bitmaps.
- incremental restore from PBS on site2 on vm1-bis from the only changes bitmaps.

thus, less bandwidth used, less restore time, and a quick “ready to work” vm1-bis on another site...

Since it’s stored like that on PBS, would this be possible ? If yes, is it planned on roadmap ?

regards,

 

[dcsapak]

mhmm... the problem with this is that we do not save the dirty-bitmap information in the backup (each backup is a complete backup), so we do not really have a way to only overwrite the blocks that changed
but there are some patches on the list to have a 'live restore' that starts a vm and streams the necessary blocks from a pbs

https://lists.proxmox.com/pipermail/pbs-devel/2021-January/001819.html

this would even have lower restore time

 

[DJP]

Hello dcsapak,

Oh ! I thought dirty-bitmaps informations would be save since in metadata or something since PBS save only the "incremental" part. I'm not really into dirty-bitmaps so I don't know how it exactly works but, isn't PBS "recreate" a "full" backup based on the incremental one, and thus, need to know where to apply the dirty-bitmaps for each incremental backup ?

FULL (150Go) < INC1 (10Mo) < INC2 (1Go) < INC3 (20Mo)
Therefore, INC3 need to know where to apply changed stuff on INC2, and INC2 need to know where to apply changed stuff on INC1 and so on... That's what PBS verification backup do : for each incremental backup, the full recovery can be done. Or am i wrong ?

For the live restore part, well that's very interresting indeed and is way better than this "incremental restore" idea ! The background process sync is pretty neat/smart, but need no error at all while restoring, else you lose everything.
Anyway, I think both approach have their pro and cons.

Do you know when this feature could be available ?

Regards,

 

[dcsapak]

Oh ! I thought dirty-bitmaps informations would be save since in metadata or something since PBS save only the "incremental" part. I'm not really into dirty-bitmaps so I don't know how it exactly works but, isn't PBS "recreate" a "full" backup based on the incremental one, and thus, need to know where to apply the dirty-bitmaps for each incremental backup ?

no the backup client basically tells the server during backup which parts it should reuse, the actual chunks on the server are deduplicated anyway so they will not be duplicated across backups
when a backup is finished, there is no way to tell which chunks the server/client has reused

the pbs does not save 'incremental' backups, but the chunks are deduplicated and each snapshot references a full set of chunks

 

[DJP]

Ohhhh, ok, crystal clear !

Well... I think the best option is to wait for live restore then ! ^^

Thank you for your clarifications !

 

[Deepen Dhulla]

Awaiting a Comparable Feature in PBS. Presently employing PBS for 7 copy Version Backup as Disaster Recovery storage, specifically to ensure Data Safety from Ransomware attacks. While also utilizing the ceph snapshot rbd export-diff and import-diff , with protect and clone mechanism with three copies of the snapshot between the Primary Data Center (PDC) and the Disaster Recovery Site (DRS) Cluster, thereby optimizing the restoration process for improved uptime.