Spot droplets
Use preemptible compute capacity for restartable workloads.
Spot droplets use capacity that can be reclaimed by its provider. They suit workers, batch jobs, and other workloads that can checkpoint and restart.
Availability and placement
A region and node pool must publish spot capacity before the option appears in the creation form. Spot placement still checks project quota, CPU compatibility, network access, and storage requirements. Availability can change between preview and execution.
Set Capacity type → Spot when creating a droplet. A rejected spot request does not fall back to standard capacity automatically. Select a standard droplet explicitly if its cost and availability are acceptable.
Reclamation lifecycle
When capacity is reclaimed, Infra records a preemption event and sends the configured event notification when a warning is available. A warning period is not guaranteed. The guest can stop before a handler finishes checkpointing.
The droplet moves through Preempting to Preempted. Execution stops and compute reservations are released. Persistent data volumes remain according to their retention settings; ephemeral disks are removed. A retained root volume continues to consume storage and can incur charges.
Design a restartable workload
Store authoritative state on persistent replicated volumes or an external data service. Write checkpoints often enough to meet the application's loss budget. Workers should obtain work through leases so another worker can retry after an interruption.
Do not place the only copy of a database or a control-plane dependency on spot capacity. A backup policy protects retained data but cannot recover writes that existed only in lost memory or ephemeral storage.
Restart and replacement
Open a preempted droplet and select Start to request spot capacity again. The action can fail while the pool has no suitable capacity. An automation rule can request retries with a bounded delay; repeated API calls without backoff do not improve placement eligibility.
To use standard capacity, stop the droplet if required and select Change
capacity type. Infra validates quota and destination capacity before applying
the change. The operation can require a new placement and does not guarantee
that the previous node remains available. Automation uses the
change_capacity_type action, with spot=false
for standard capacity. Wait for its operation before starting the guest.
Notifications and accounting
Use webhooks for preemption events and metrics to observe worker availability. Webhook delivery is asynchronous and can be repeated. A receiver should reconcile current droplet state before submitting a replacement request.
Spot compute rates and retention charges appear separately in usage and billing. A project budget alert does not stop spot resources; use quotas and automation for admission control.
Published spot rates
The standard compute price card defines separate discounts for eligible CPU and GPU spot allocations. The creation review shows the selected rate. Persistent volumes, retained addresses, backups, and unrelated traffic keep their ordinary meters.