Question 30 A developer expects a short-lived serverless function instance to retain reliable application state in local memory across all future invocations. Which response is correct? Persist durable state in an external appropriate service Write records only to temporary execution disk and never persist them Assume each invocation reuses the same process indefinitely Disable all application authentication to make state survive Explanation Correct Answer A:Function execution instances can change; local process memory is not a durable cross-invocation store. Incorrect Answers B:Ephemeral local storage is not reliable for durable application state. C:Event-driven compute can scale and recycle instances. D:Authentication settings do not create durable process memory.
Question 31 A small event handler rarely runs. What billing characteristic is commonly associated with eligible serverless consumption plans? Function execution never has memory or duration constraints Charges relate to executions and consumed compute rather than a dedicated idle VM Every event handler must buy a dedicated physical server Every plan is guaranteed free for unlimited executions Explanation Correct Answer B:Execution-based charging can match sporadic events, subject to plan and related service costs. Incorrect Answers A:Serverless hosting plans impose resource and runtime characteristics. C:Serverless platforms abstract host procurement. D:Free grants and charge models have limits and vary by plan.
Question 32 A web API becomes busy and engineers add more identical application instances behind a traffic distributor. What scaling method is used? Removing worker instances during peak demand Moving data to a colder storage tier Vertical scale-up only Horizontal scale-out Explanation Correct Answer D:Adding instances increases capacity through more workers rather than larger individual machines. Incorrect Answers A:Scale-in would shrink capacity at the wrong time. B:Storage temperature does not add API compute workers. C:Scale-up changes the size of a single instance, not the count.
Question 33 A database workload needs more memory and CPUs on a single compatible VM rather than more VMs. What scaling action is this? Horizontal scale-out by adding machines A region-wide failover without changing VM size Vertical scale-up Removing CPU resources from the VM Explanation Correct Answer C:Increasing the capacity of one machine is vertical scaling. Incorrect Answers A:Adding peers changes instance count, not the size of the existing VM. B:Failover concerns location and continuity rather than VM capacity. D:Scaling down would reduce the requested capacity.
Question 34 An application has many instances after a campaign; traffic has returned to normal and costs should fall. Which action is appropriate? Scale in by reducing excess instances Scale out further despite falling demand Keep permanent peak instance count by policy Increase every instance size while demand falls Explanation Correct Answer A:Removing unnecessary workers lowers capacity and can reduce compute charges. Incorrect Answers B:Adding instances raises capacity when less is needed. C:Fixed peak capacity prevents cost reduction after the surge. D:Upsizing raises capacity and potentially cost rather than releasing it.
Question 35 A team wants capacity to grow when monitored CPU utilization crosses a threshold and shrink after sustained lower use. What should it configure? A one-time manual change to the VM size Assume a fixed worker count can handle every demand spike Autoscale rules tied to appropriate metrics A purchase order for new equipment on each event Explanation Correct Answer C:Metric-based policies adjust capacity as traffic changes without manual intervention. Incorrect Answers A:A static change will not respond to future metric fluctuations. B:A static number of workers cannot add capacity automatically when traffic exceeds the normal load. D:Hardware procurement is not automated elasticity.
Question 36 An architect says increasing server count during demand spikes automatically protects against a full-region outage. What is the distinction? Autoscaling guarantees global failover by default Application availability never depends on deployment topology All regions share one failure boundary Autoscaling addresses capacity; regional resilience needs separate geographic design Explanation Correct Answer D:More instances within one affected region cannot by themselves survive loss of that region. Incorrect Answers A:Scaling policies do not create a second-region deployment. B:Availability is affected by redundancy and fault domains. C:Regional geography can provide independent failure boundaries when designed properly.
Question 37 An API is designed to add and remove many instances dynamically. Why should per-user session state not be stored only in one instance's memory? Scaling out always makes every process share one memory address Store all application session state on one worker's local disk Requests may reach different instances or an instance may be removed Autoscaling permanently binds each user to the same original process Explanation Correct Answer C:External durable session storage avoids losing state when horizontally scaling. Incorrect Answers A:Independent instances do not automatically share volatile memory. B:A single node's private state becomes unavailable when requests are distributed across changing instances. D:Instances can be replaced or scale in.
Question 38 A customer-facing checkout must remain accessible when one application instance fails. What design principle primarily addresses this? Redundant healthy instances with failover or load distribution A single oversized server with no backup instance A private Git branch of application code alone A nightly cost report without operational redundancy Explanation Correct Answer A:A surviving instance can continue serving requests after a single instance outage. Incorrect Answers B:More CPU on one server does not eliminate its single point of failure. C:Source control does not provide runtime failover. D:Cost data does not keep the checkout live during failure.
Question 39 A service must continue processing despite occasional infrastructure component failures. Which approach most directly supports reliability? Removing all telemetry because hardware is managed Fault-tolerant design with retries and redundant components Assuming the cloud provider prevents every component failure Keeping one manually managed machine as the only runtime Explanation Correct Answer B:Recoverable errors and redundancy reduce the effect of individual failures. Incorrect Answers A:No observability would hinder detection and recovery. C:No infrastructure can guarantee zero failures. D:A single node increases failure impact.
Question 40 A business asks whether stored objects can survive hardware loss, independently of whether a web application stays reachable. Which property is being evaluated? Application responsiveness at peak request rate Interactive user sign-in latency Data durability A software subscription's monthly cost Explanation Correct Answer C:Durability concerns preservation of stored information over time despite component failures. Incorrect Answers A:Throughput is different from preservation of stored bytes. B:Authentication performance does not measure data survival. D:Price alone does not establish the probability of data retention.
