Recheck Spot recovery ownership when enabling scale-set autoscale

Source facts

Spot Try & restore attempts to recover capacity-evicted instances in a scale set to maintain its target instance count. The feature is disabled when the scale set uses Autoscale; autoscale rules then drive the number of VMs. Microsoft also cautions that autoscaling can attempt to replace evicted Spot capacity but does not guarantee successful allocation. Microsoft Learn.

Applicability

Use this check when adding autoscale to an existing Spot workload or explaining why the previously expected restore mechanism is not acting. Confirm the current controller and eviction policy before attributing a reduced instance count to a failed repair.

DSE recommendation

Assign one explicit recovery expectation to the scale set’s selected capacity-control configuration. Have the workload owner state what should happen after capacity loss and compare that expectation with the actual autoscale rules. Review whether the workload can tolerate a period below its desired population. Keep allocation uncertainty in the operating plan rather than promising that either mechanism restores capacity immediately.

Verification

Use an authorized eviction simulation on a disposable instance and observe the selected mechanism’s response. Track the target count, actual instances and any unsuccessful replacement attempt. Verify application work completion separately from the population count. After changing autoscale configuration, revisit the recovery record so responders do not rely on a Try & restore setting that is inactive under the new configuration.

Official references

Microsoft Learn: Azure Spot Virtual Machines for Virtual Machine Scale Sets. Source reviewed September 9, 2026.