Rollback needs all IO ressources (ceph storage)

[mohnewald]

Hello,

my cluster struggles with IO if i rollback a VM.
Is there a way to IO Limit Rollback (ceph storage)?

Cheers,
Mario

 

[mohnewald]

Any news on this since March 21?

 

[t.lamprecht]

Hi,

Some more details would help, first your Proxmox VE and ceph versions, and some basic overview about your ceph OSD disk/network setup.

For now there's no real way to limit rollback impact FWICT. Potentially there could be a slightly more efficient way to handle rollbacks in Ceph though. Ceph actually recommends doing a clone of the old snapshot over a rollback if possible, as that avoids needing to actively rewrite the image data due to better use of the copy-on-write semantics. But, that needs some thoughts about if Proxmox VE can actually use that without risking any period of data-loss and while keeping the other snapshots intact.

 

[VictorSTS]

With the current method of rolling back a VM with Ceph storage, does Promox have to rewrite the whole disk image or just writes the changed blocks from the snapshot back to the disk image? I suppose it's the later as I never had I/O issues while rolling back VMs, so I never really checked how Proxmox does it.

 

[t.lamprecht]

Ceph needs to bring the current volume into the original state, so the difference between now-state and snapshot-state has to be written, IIRC.