You run a stateless API on Azure Virtual Machine Scale Sets. Traffic is highly variable. You must minimize cost by using preemptible capacity for burst traffic, scale out based on CPU and queue depth, and roll out application updates safely with health probes and surge. Which approach should you use?

The correct approach leverages two VM scale sets to balance cost and reliability. A base set of Standard-priority instances handles steady load, ensuring continuous service. A second Spot-priority set provides cost-effective burst capacity, scaling out when demand increases. A load balancer distributes traffic between them. Autoscale rules manage scaling for each set independently based on CPU and queue depth. Rolling upgrades with health probes and maxSurge on the Standard set ensure safe application updates. Incorrect options: A single uniform VM scale set of Spot VMs with Deallocate eviction is too risky for a stateless API, as all instances could be evicted, causing downtime. A single flexible-orchestration VM scale set with only Spot instances and relying on restarts during evictions is also too risky for a production API, as evictions are disruptive. Moving to Azure App Service with manual instance management doesn't address the requirement for preemptible capacity for burst traffic or automatic scaling based on CPU and queue depth. A single uniform VM scale set with Standard-priority instances but disabled health probes and NewestVM scale-in policy compromises safety during rollouts and doesn't utilize cost-effective Spot instances for bursts.