Snapshots (LVM, ZFS, Btrfs, VMware, Proxmox, cloud volumes) take seconds because they only track changes from that point on. They are an excellent fallback point before an update.
But they share the fate of the system: if the storage fails, the account is deleted or ransomware encrypts the hypervisor, the snapshots go with it. A backup lives on separate storage, ideally following the 3-2-1 rule with one off-site and one immutable copy.
Two mistakes from practice: snapshots left in place for weeks that slow down the VM or fill the storage, and snapshots of a running database that are only crash-consistent. Clean backups need a database dump or a freeze.
Example
lvcreate -s -n data_pre_update -L 10G /dev/vg0/data