Choose Spot eviction state from the workload's restart plan

Source facts

Azure Spot VMs can be evicted when Azure needs capacity and have no availability guarantee. The default Deallocate eviction policy retains a stopped-deallocated VM, but later allocation is not guaranteed and the retained VM still consumes quota. Delete removes the VM and its underlying disks. In-VM Scheduled Events eviction notifications are best effort, arriving at most 30 seconds before eviction. Microsoft Learn.

Applicability

Use Spot only after the workload owner has accepted interruption. Identify which state is disposable, which must survive, and whether processing can resume elsewhere. Review the current VM-size and subscription eligibility separately from the application’s restart design. Do not assume advance notification will always provide a usable cleanup interval.

DSE recommendation

Choose the eviction policy only after documenting where restartable work and durable state will live. For retained instances, assign responsibility for retrying allocation and reconciling the remaining quota footprint. For deletion, require an independently demonstrated rebuild and state-recovery path. Keep the chosen policy visible in the workload’s operating record rather than relying on the default.

Verification

Exercise an approved interruption test without depending on a final notification to save essential progress. Verify the actual surviving resources and the application’s restart behavior against the chosen policy. Record an unsuccessful capacity request as an expected contingency to handle, not proof that retained disks guarantee an immediate restart. Test the alternate processing path before placing deadline-sensitive work on the deployment.

Official references

Microsoft Learn: Azure Spot Virtual Machines. Source reviewed September 9, 2026.