Snapshot best practices

[CPO_Siemens]

Are there any best practices for snapshot use?
For VMware there are docs for this but I could not find anything official about performance impacts, limits on VM operations and such for Proxmox.
I strongly suppose, the general "more snapshots -> less performance", "don't keep them too long" and "snapshots are no backups" are no less true for Proxmox but a real source would be nice nonetheless.

 

[longer]

Interesting topic, I am not aware of performance impact with keeping snapshot for too long or more snapshots means less performance.
Indeed "snapshots are no backups" is to be taken seriously therefore ensure to always have your scheduled backup in place.

 

[d.oishi]

As far as I have researched, I understand that there is no well-established best-practices document for this.

This is probably more about the specifications than best practices, but the following patch may eventually be reflected in the documentation:

https://lore.proxmox.com/pve-devel/20260714142629.277917-5-f.ebner@proxmox.com/

I strongly suppose, the general "more snapshots -> less performance", "don't keep them too long" and "snapshots are no backups" are no less true for Proxmox but a real source would be nice nonetheless.

I also think that these general principles apply to Proxmox as well.

 

[Impact]

I'm not aware of an official PVE specific source but if you care, here's some of my opinions about it. This heavilly depends on the storage type.
My experience is with ZFS and LVM-Thin, which are also what I recommend, in that order

more snapshots -> less performance

I haven't noticed this with ZFS or LVM-Thin.

don't keep them too long

Snapshots have to retain data that you delete/change after they were taken and thus old retained data accumulates over time. I guess it can be compared to differential backups.

snapshots are no backups

Indeed. Snapshots stay on the storage. Backups can/should be transferred to a different place. Backups are more important than snapshots. Best to use both.

Virtual machine disks cannot be increased in size while snapshots exist, regardless of whether the virtual machine is powered on or off. Increasing the size of VMDK disks running on a snapshot should never be attempted, even using the CLI.

Works fine for me with PVE.

Ensure that there are no snapshots before increasing the size of a virtual machine disk or virtual RDM, as doing so can cause snapshot corruption and data loss.

No problems with this either.


I handle my snapshots via cv4pve-autosnap. I currently keep 4 and snapshot every 6 hours. This allows very fast restores without needing to involve PBS after testing (or messing up) something.

 

[CPO_Siemens]

This heavilly depends on the storage type.
My experience is with ZFS and LVM-Thin, which are also what I recommend, in that order

We are using Ceph so our mileage may vary...

Thanks a lot nonetheless.

 

[alexskysilk]

ceph and zfs are CoW native; snapshots are not "real" in the sense that the data isnt separated- think of snapshots as bookmarks.
VMDK isnt, and so snapshots are addon files. qcow2 can exhibit similar performance degradation.

In summary- snapshot performance impact is a function of the underlying disk/filesystem. not an inherent limitation.