Question 350
A service uses an RDS Proxy between Lambda and a relational database. Metrics show many client connections still pinning sessions to database connections because the application relies on session-specific SQL state. Which optimization should be investigated?
- Disable RDS Proxy and assume each connection becomes free to the database merely because it is direct.
- Enable performance metrics for RDS Proxy but leave the application session-level SET variables and temporary tables unchanged.
- Reduce session-pinning behavior where safe and tune connection reuse so the proxy can multiplex more requests onto fewer database connections.
- Open an additional unique persistent database connection for every Lambda invocation and retain all session variables.
Explanation
Correct Answer
C:RDS Proxy pooling benefits can be limited by pinned sessions; changing application session-state usage can improve multiplexing.
Incorrect Answers
A:Removing pooling restores connection churn without addressing the application pinning cause.
B:Observability reveals pinning but does not unpin database sessions or make back-end multiplexing more effective.
D:More pinned dedicated sessions increase the database connection pressure that the proxy is intended to reduce.
Question 351
A backend application serves thousands of analytical event consumers from one Kinesis Data Stream. All consumers share the default retrieval throughput and cause read throttling as consumer count grows, even though producer ingest is within limits. What Kinesis feature should be assessed?
- Disable record retention so only the latest event can be read by every consumer simultaneously.
- Encrypt the stream with a customer-managed KMS key and expect that encryption configuration to reserve read bandwidth per consumer.
- Use enhanced fan-out consumers for eligible independent readers and validate dedicated read throughput and cost.
- Keep all readers sharing the same limited shard retrieval throughput and simply add consumer processes indefinitely.
Explanation
Correct Answer
C:Enhanced fan-out can provide per-consumer read capacity and reduce contention among independent stream consumers.
Incorrect Answers
A:Shortening retention risks data loss for slow consumers and does not add dedicated read capacity.
B:KMS encryption protects data but is not enhanced fan-out and does not dedicate per-consumer shard read throughput.
D:More readers do not increase shared per-shard polling throughput and can worsen throttling.
Question 352
A product analytics stream receives broadly distributed partition keys from hundreds of producers. During launch events, aggregate shard write throughput doubles unpredictably and CloudWatch shows many shards reaching their provisioned aggregate capacity together, rather than one hot key dominating. The team wants to remove manual shard-count planning while monitoring documented scaling limits. What change should be evaluated?
- Force every producer to use the same constant partition key so its write traffic is concentrated in one shard for easier operations.
- Switch the Kinesis Data Stream from provisioned capacity to a supported on-demand capacity mode and monitor scaling and traffic distribution.
- Lower the stream retention period, assuming fewer historical records automatically increases per-shard incoming write throughput in provisioned mode.
- Increase only the number of enhanced fan-out consumers, because registered consumers directly raise the producer write quota for every shard.
Explanation
Correct Answer
B:Kinesis on-demand modes manage shard capacity according to observed aggregate stream throughput, reducing the need for manual provisioned resizing.
Incorrect Answers
A:A constant partition key creates a hot shard and aggravates throttling even if aggregate stream capacity is otherwise available.
C:Retention settings manage historical data availability, not the per-shard incoming write capacity needed during current spikes.
D:Enhanced fan-out isolates read consumers; it does not increase incoming producer bytes per second under provisioned shard limits.
Question 353
An event-driven Java Lambda function is invoked infrequently, and trace data shows most p99 latency is Java runtime initialization rather than handler execution. The deployment uses a supported Java runtime and published versions. What startup optimization should be evaluated?
- Evaluate Lambda SnapStart for supported runtime configurations and verify initialization compatibility, unique state and restored performance.
- Increase the Lambda memory allocation without measuring whether JVM dependency loading and cold-start initialization are the dominant issue.
- Disable all function version publication while requiring a snapshot of a published version.
- Increase the Lambda timeout to fifteen minutes and assume clients experience shorter cold-start response latency.
Explanation
Correct Answer
A:SnapStart captures initialized execution state to reduce eligible Java startup latency, subject to snapshot-state correctness.
Incorrect Answers
B:Memory may affect CPU allocation but does not provide the pre-initialized snapshot capability; its benefit requires separate benchmarking.
C:SnapStart relies on supported published-version snapshot behavior rather than an undefined mutable runtime state.
D:Timeout changes the maximum execution window; it does not reduce runtime initialization time.
Question 354
A CPU-heavy Lambda image processing function finishes slowly even with sufficient concurrency. Its billed duration dominates latency and cost, and performance may improve if the allocated memory raises available CPU. Which tuning exercise is appropriate?
- Increase Lambda reserved concurrency but leave the same CPU-intensive function memory allocation and observed per-invocation duration.
- Use provisioned concurrency exclusively to eliminate cold starts even though warm invocations are themselves CPU-bound.
- Benchmark several Lambda memory sizes and compare duration, p95 latency and total price per request rather than assuming memory only changes RAM.
- Reduce memory to the smallest allowed value without load testing and assume execution CPU remains unchanged.
Explanation
Correct Answer
C:Lambda memory allocation influences available compute power, and tuning requires measured end-to-end cost/performance.
Incorrect Answers
A:Reserved concurrency bounds simultaneous executions but does not raise CPU allocated to one invocation.
B:Provisioned concurrency targets cold-start latency; it does not replace benchmarking CPU allocation and duration on a warm invocation.
D:Lower memory can also reduce compute allocation and lengthen CPU-bound execution.
Question 355
A high-performance computing cluster needs the lowest practical network latency among tightly coupled EC2 instances, and the job is designed to run within one Availability Zone. Which EC2 placement strategy should be evaluated?
- Use an EC2 cluster placement group for supported instances, considering capacity and single-AZ placement tradeoffs.
- Use a spread placement group across separate racks specifically to maximize the proximity of every tightly coupled node.
- Use a partition placement group as though splitting the MPI peers across partitions guarantees the lowest possible cross-peer latency.
- Use a spread placement group across multiple Availability Zones rather than preserving the requested single-AZ tightly coupled network.
Explanation
Correct Answer
A:Cluster placement groups place compatible instances close for low-latency, high-throughput inter-instance networking.
Incorrect Answers
B:Spread groups prioritize failure separation rather than physically close placement for HPC networking.
C:Partition placement primarily separates failure groups for large distributed workloads, not same-network-rack proximity for tightly coupled HPC.
D:Spread placement groups emphasize separating failure domains, not minimizing cross-node communication latency for this workload.
Question 356
An EFS-backed multi-AZ web application suffers throughput bottlenecks during unpredictable bursts of parallel file access. The file size is stable, but file throughput needs vary dramatically hour to hour. Which EFS performance setting should be considered?
- Evaluate a supported Elastic Throughput mode and benchmark the actual burst workload against its price and throughput limits.
- Continue using an EFS throughput mode tuned to a small fixed baseline without validating whether bursts exceed its effective throughput entitlement.
- Change EFS general purpose performance mode to a different latency profile while leaving the unmet throughput budget unresolved.
- Increase EC2 subnet address capacity without assessing the EFS throughput mode or metered throughput.
Explanation
Correct Answer
A:Elastic Throughput responds to variable file I/O demands without manual provisioned-throughput sizing.
Incorrect Answers
B:A capacity mode constrained by the old baseline may fail the sudden workload surges described.
C:Performance mode affects concurrency and latency behavior and does not substitute for selecting sufficient throughput on a bursty workload.
D:A larger subnet alone does not raise EFS service-side file I/O throughput.
Question 357
An international catalog uses CloudFront with a primary custom HTTP origin in one Region and a read-only secondary HTTP origin in another Region. When the primary begins returning supported 5xx errors, CloudFront should attempt the secondary without changing the viewer DNS name. Which CloudFront configuration supports that?
- Create a CloudFront origin group with primary and secondary origins and appropriate failover status codes for supported requests.
- Enable Origin Shield only and assume an extra cache tier automatically chooses a separate origin on supported 5xx failures.
- Increase the cache TTL to a year so the secondary origin becomes the primary without an origin group.
- Change CloudFront cache behavior TTL and rely on caching expired error responses instead of configuring eligible secondary-origin failover.
Explanation
Correct Answer
A:CloudFront origin groups support configured failover between two origins for eligible request and error conditions, keeping distribution routing stable to viewers.
Incorrect Answers
B:Origin Shield consolidates origin cache misses but does not automatically implement dual-origin failover routing and selection.
C:TTL controls cache freshness and is not a failover rule for errors from one of two custom origins.
D:Cache duration alone does not configure CloudFront to retry eligible requests against another healthy custom origin.
Question 358
A data lake stores petabytes of S3 objects that vary unpredictably between frequent and infrequent access. The analytics team requires millisecond access on all routinely queried objects, and cannot forecast when each object's access will change. Which storage optimization is best aligned with the requirements?
- Use S3 Intelligent-Tiering but enable archive access tiers for the whole bucket with no consideration of restore delays on frequently queried objects.
- Use S3 Intelligent-Tiering for eligible objects, and evaluate its tiering and monitoring charges against the observed dataset.
- Use S3 Standard-IA for every object at ingestion regardless of access frequency or minimum storage duration.
- Use S3 Glacier Instant Retrieval for all objects even though a substantial unpredictable fraction is accessed frequently.
Explanation
Correct Answer
B:S3 Intelligent-Tiering can automatically optimize storage costs as access patterns change while supported frequent and infrequent access tiers retain low-latency access; consider monitoring fees and object-size rules.
Incorrect Answers
A:Optional archive tiers can introduce retrieval delays inconsistent with the required millisecond access. Evaluate eligible frequent/infrequent tiers and charges.
C:Frequent or unpredictable retrieval can incur request/retrieval and minimum-duration charges; Intelligent-Tiering better adapts to changing patterns.
D:Instant Retrieval has millisecond access but charges for retrieval and minimum retention; it is not automatically optimal for frequently changing access patterns.
Question 359
Logs in S3 are read frequently for 30 days, rarely for the next 90 days and retained for regulatory purposes for seven years. A restoration delay of hours is acceptable for the oldest records. What storage policy reduces cost while retaining required history?
- Transition all logs directly to Glacier Deep Archive at day one while operational queries need prompt access for their first 30 days.
- Transition logs only to Standard-IA at day 30 and keep them there for the full seven years without evaluating deeper archive tiers.
- Delete S3 object versions after 90 days while claiming a seven-year compliant retention policy.
- Configure a lifecycle policy moving aged logs through appropriate infrequent-access and archive tiers while preserving the required retention period.
Explanation
Correct Answer
D:Lifecycle transitions match declining access frequency with cheaper classes, while retention requirements and retrieval delay constrain archival choices.
Incorrect Answers
A:Deep archive restoration delay conflicts with near-term operational reads; lifecycle transitions should follow the measured hot and cold stages.
B:Standard-IA can be appropriate for near-cold use, but seven-year retention with hours-tolerant access often supports cheaper archive tiers.
C:Deleting required versions violates the seven-year retention; storage-class optimization must preserve all required evidence.
Question 360
A batch-rendering farm can retry individual jobs and checkpoint progress to S3. Jobs have a flexible completion window, and on-demand EC2 charges dominate costs. Which purchase model is most suitable for incremental capacity?
- Use EC2 Spot Instances for interruption-tolerant workers with checkpointing, retry and capacity diversification.
- Use a three-year Compute Savings Plan sized to the one-week peak, assuming it carries no commitment cost during idle weeks.
- Use EC2 On-Demand capacity for all interruption-tolerant jobs solely to eliminate any need for checkpointing, despite checkpoints already existing.
- Use Spot only in one instance type and Availability Zone, treating the cheapest pool as a guarantee of availability.
Explanation
Correct Answer
A:Spot discounts match flexible stateless or checkpointed batch work, provided interruption handling is implemented.
Incorrect Answers
B:A Savings Plan is a continuing hourly dollar commitment; it can be wasteful for highly intermittent, short-lived peaks.
C:On-Demand avoids Spot interruptions but is generally more expensive for these retryable flexible jobs; the workload already supports safe Spot use.
D:Spot capacity varies and can be interrupted; diversification and retry/checkpoint handling reduce risks under this flexible batch requirement.
AWS Certified Solutions Architect - Professional SAP-C02 Practice Tests