Tape backup : filter chuncks for one day in time

[O van D'Ho]

Hi there,

I've started to use proxmox a few months ago and did setup PBS with lots of disks and 1 tape drive.
Disk backup is running smoothly. I can choose any day in the past 4 months and restore a vm.

The thing is the tape backup job seems to backup all "the disk backup store" on tape every day which takes a long time now.
If I understand correctly, disk backup only adds a few chunks every day using deduplication and backuping only what's changed on the vm while tape backup put all chunks on tape so we can recover the complete backup store.

Idealy, I would like to have an end of month tape with all chunks so i can restore all vm at end of month state, then during the month, i would save newer chunks to other tapes.
I would have like 12 end of month tapes per year each with all chunks and catalog to restore vm at the month's state.

Is that possible while keeping all backups on disks?

Thank you in advance for any help.

O

 

[BriBriBri]

I think the "latest only" checkbox in the properties of the tape backup job is close to what you're looking for.



With that checked, if I write to tape on March 1, it's going to grab only the lastest disk-to-disk snapshot for each VM and write it to tape (as a complete backup, generated from prior snapshots and the most current incremental), as of March 1.

Then, if I write to tape on April 1, it'll do the same as of April 1.

And so on.

This appears to be the case based on what I'm seeing in the restore functionality of the PBS web GUI. I'm sure someone will correct me if I'm wrong?

I guess the only place where this differs from your stated needs is that the tapes are independent of each other. Meaning, there are no "incremental tape backups". . . the March 1 tape has everything it needs to restore your VMs to the March 1 state while your April 1 tape will have everything it needs to restore your VMs to the April 1 state (with no need to get anything from the March 1 tapes).

Writing to tape daily isn't practical for us with over 80TB of data. In our environment, we write weekly to tape and organize those tapes appropriately for having offsite/disaster-recovery backups in case of catastrophic loss of our pbs datastore (the only real role for tape given that with dedupe, your disk-to-disk datastore should really be able to handle any restores you need into the foreseeable future unless your data changes constantly).

For those who have been around since we were writing our backups directly to tape, and scheduling "full" vs "incremental" tape backups, etc. . . this migration to a disk-to-disk model with tapes only as an offsite/CYA/utmost-disaster contingency. . . it can be a bit disorienting.

Apologies if any of this is wrong. Hopefully someone with some authority can confirm.

 

[dcsapak]

since the backups we write on tape are fully deduplicated, it only depends on the retention and allocation policies what is written onto the tape
if you e.g. set the allocation policy to once a month, any additional tape backup will continue to write to the same media-set and not rewrite any chunks until the new media-set starts

see the example setups to get an idea how you might achieve what you want: https://pbs.proxmox.com/docs/tape-backup.html#example-setups