Local and ephemeral disks
Distinguish node-local persistence from disposable disk lifetime and recovery coverage.
Local storage describes where disk blocks live. Ephemeral storage describes when a disk is discarded. A local persistent volume and an ephemeral scratch disk therefore have different retention behavior.
Persistent local volumes
A local persistent volume belongs to a project and retains its contents through a guest stop. It remains dependent on its storage node. Another compute node cannot treat that disk as replicated storage because its resource ID still exists.
A stopped move requires a supported data copy and enough target capacity. Source failure can make the only live copy unavailable. Independent backup coverage must be established before relying on recovery elsewhere.
Ephemeral scratch disks
Add scratch capacity under Droplet → Hardware → Scratch disks. It is initialized for the placement and is excluded from ordinary snapshots and backup policies. The console marks it separately from persistent volume attachments.
A guest reboot preserves scratch state while the same placement and disk remain active. A stop, replacement placement, spot preemption, or droplet deletion discards it. A failed start must not be used as evidence that previous scratch contents remain recoverable.
Do not put application journals, the only checkpoint, or guest credentials on scratch storage unless their loss is explicitly acceptable. An application must be able to recreate that disk's contents when it starts.
Placement and migration
A droplet with scratch storage is ineligible for ordinary live migration. A cold move creates new scratch disks at the destination; it does not copy their contents. The move preview separates those disks from persistent local volumes that require a data transfer.
Selecting Drain reports the guest as a live-move blocker. The drain cannot silently discard scratch state by stopping it. Use an approved stop-and-restart procedure for restartable workloads.
Quota and observed space
Scratch reservations count against the project's configured ephemeral capacity and the node's local-device admission limit. Guest free space is separate from that reservation. Deleting guest files does not release the configured allocation.
When a guest is removed, capacity is released after the placement's cleanup. An unreachable host can keep its scratch reservation unresolved until the original execution is fenced and reconciled.
Recovery manifests
A backup manifest lists persistent disks included in its capture and separately identifies excluded scratch devices from the source configuration. Restoring the backup recreates eligible scratch capacity empty.
Check device discovery and mount configuration after restoration. A guest that requires a scratch mount must tolerate a new filesystem identifier. See Volume attachments, Spot droplets, and Hardware policy API.