Question 120 A team wants to know when autoscale should add web instances. What evidence should guide the threshold? A contract that requests never fluctuate Only the VM resource display names Measured traffic, response times, utilization and capacity tests A universal CPU threshold guaranteed optimal for every application Explanation Correct Answer C:Workload behavior provides better scaling triggers than arbitrary guesses. Incorrect Answers A:Demand should be measured, not assumed constant. B:Names provide no capacity measurements. D:Appropriate signals vary by workload.
Question 121 A service processes independent work items and is limited by a single worker's throughput. What scale-out property is useful? All work must run sequentially on one process forever A service-level agreement changes application algorithms automatically Storage account rename increases CPU cores Multiple workers can process independent items concurrently Explanation Correct Answer D:Parallelizable work can be distributed across additional instances. Incorrect Answers A:Serialization prevents use of more workers. B:An SLA is not a parallel processing implementation. C:Renaming does not add compute throughput.
Question 122 A dataset grows beyond efficient processing on one node. Which approach can help distributed analysis? Put every record in one process's memory regardless of size Change only the storage container's display name Assume more storage capacity always means more CPU Partition suitable independent work across multiple compute workers Explanation Correct Answer D:Parallel execution can handle larger workloads when algorithms and data support it. Incorrect Answers A:Single-process limits can cause bottlenecks. B:Naming does not partition computation. C:Storage volume alone is not processing power.
Question 123 An application requests more instances than its regional quota permits. What should operators do? Delete monitoring to conceal quota errors Review applicable service quotas and request or design for additional capacity Increase the count of dashboard widgets Assume the scale request always succeeds without capacity checks Explanation Correct Answer B:Autoscale cannot exceed actual resource limits. Incorrect Answers A:Removing visibility does not resolve constrained capacity. C:UI widgets do not modify compute limits. D:Cloud limits can reject deployment.
Question 124 Autoscale adds new app instances, but they cannot serve traffic until dependencies initialize. What should the team account for? Disable all probes so new instances receive traffic sooner Assume every new VM starts fully warmed instantaneously Startup and readiness checks before routing real requests Treat a new public IP as proof application initialization completed Explanation Correct Answer C:New capacity should become healthy before serving application traffic. Incorrect Answers A:Premature routing can cause errors. B:Initialization can take time. D:Address allocation is not readiness validation.
Question 125 A service adds instances under load but all instances remain in one fault domain. Which limitation persists? A regional failure never matters after scaling The instances cannot improve request throughput at all Autoscaling alone does not ensure failure-domain diversity Resource tags automatically diversify physical hosts Explanation Correct Answer C:The scaling policy must be combined with suitable zone or availability design. Incorrect Answers A:More nodes in one failed region still fail. B:Additional compute may raise throughput. D:Labels do not control all placement behavior.
Question 126 An API experiences occasional short network errors when calling a managed service. Which practice helps withstand transient failures? Removing all request timeouts Assuming cloud networking never fails Bounded retries with appropriate backoff and idempotency Retrying without delay indefinitely under every failure Explanation Correct Answer C:Transient recovery works better when retries avoid overload and duplicate effects. Incorrect Answers A:No timeouts can leave operations hanging. B:Distributed systems can experience transient faults. D:Unbounded retry storms can worsen incidents.
Question 127 A message can be delivered twice after a consumer restarts. Which application property prevents duplicate financial effects? Skipping all message acknowledgments indefinitely Keeping separate anonymous copies of the transaction on every retry Increasing the VM size regardless of business identifiers Idempotent processing keyed to the logical operation Explanation Correct Answer D:A repeated request should not apply the same business effect twice. Incorrect Answers A:Missing acknowledgments can cause further redelivery. B:Uncoordinated copies increase duplicate risks. C:Compute capacity does not prevent duplicate business writes.
Question 128 A dependent service fails repeatedly, and the caller is overwhelming it with new requests. Which resilience idea is appropriate? Circuit breaking or controlled fallback during persistent failures Disable all application error reporting Force every user to refresh their browser continuously Increase infinite retries without a backoff limit Explanation Correct Answer A:Avoiding futile calls protects resources while the dependency recovers. Incorrect Answers B:No telemetry makes the dependency failure harder to manage. C:Repeated manual requests worsen overload. D:Retry storms can amplify failures.
Question 129 A checkout API receives requests faster than a downstream fulfillment worker can process them. Which architecture supports reliable buffering? Write every pending order directly to a single worker's local memory queue Store all pending work only in web-server RAM Place work on a durable queue for asynchronous processing Require every customer to wait for one synchronous worker indefinitely Explanation Correct Answer C:A queue can absorb demand spikes and decouple request handling from slower workers. Incorrect Answers A:Local memory queues disappear on worker loss and do not provide a durable shared message backlog. B:A server failure can erase volatile queued work. D:Tight coupling increases timeout and availability risk.
Question 130 An application consists of unrelated services; one reporting function fails and should not bring down core checkout. What architecture principle is relevant? Make every checkout request block on the reporting service Give all components one required shared process Isolate failures and critical dependency paths Disable independent health checks Explanation Correct Answer C:Separating nonessential components can prevent a reporting outage from propagating. Incorrect Answers A:That couples critical operations to an optional dependency. B:A process failure could take down every function. D:Without component health information, isolation is harder to operate.
