Suggestion: improve CT restore

[windowsdesxtop]

Greetings

I'm ~1hour into CT restore (with 1gbps guaranteed) (only ~50GB restored so far)

Would it be possible to make proxmox CT backups faster by:
a) using more than one stream/connection? (needed when latency is like 100ms+)
Now restore speed is max 300mbps (while bandwidth is 1gbps)
I could be wrong but maybe VM restore already do it better somehow?
b) live restore - would be nice addition

And all the best greetings for proxmox/pbs team. Your product group is amazing!

 

[UdoB]

You did not tell us how the actual storage of the PBS is implemented. (Specifically the ".chunks"-store.)

Especially restoring (and verifying) does need IOPS as each and every chunk has to be read from media...

 

[windowsdesxtop]

@UdoB ty for reply

I'm on zfs raidz2 no dedup/compress ~15GB arcz
iostat -mx showed ~10% utilization on each disk
pbs datastore showed only ~5reads/s
(I can get much more utilization/iops when I'm doing non restore jobs)
If needed we can pause this thread till I try much lower latency restore which would confirm/deny slow chunk store.

I could be wrong but I believe adding more TCP connections between proxmox and pbs would help (especially as ping between them increases)
OR maybe more parallel reading threads during restore

 

[SteveITS]

Did you see https://forum.proxmox.com/threads/qmrestore-cli-howto-configure-parallelism.183545/post-852737

 

[windowsdesxtop]

@SteveITS
that sounds really nice and maybe it confirms my initial observation
> I could be wrong but maybe VM restore already do it better somehow?

but as I understand
(from https://lore.proxmox.com/pbs-devel/20250714083439.1240349-3-d.csapak@proxmox.com/)
it's for qmrestore only not vzrestore
is there plans for vzrestore?