Continuous Improvement for Existing Solutions — AWS Certified Solutions Architect SAP-C02 practice questions

Data corruption occurred because the app ignored DynamoDB throttling without retrying. Which improvement is BEST?

1. Enable SDK retries with exponential backoff and jitter, alarm on ThrottledRequests, and match capacity to peak demand
2. Migrate the workload from DynamoDB to RDS PostgreSQL to gain stronger consistency guarantees on write failures
3. Add a circuit breaker that drops writes silently during throttling to prevent retry storms from overloading the backend
4. Disable DynamoDB throttling on the affected tables so the application never encounters throttled write responses at all

Answer: A — Enable SDK retries with exponential backoff and jitter, alarm on ThrottledRequests, and match capacity to peak demand

A) Retry with backoff + observability + capacity fixes the root cause. B) Doesn't fix the retry omission. C) Dropping writes worsens corruption. D) Throttling cannot be disabled.

One downstream microservice occasionally takes 30+ s causing cascading timeouts across the platform. Which change set is BEST?

1. Add per-service circuit breakers, tighter timeouts, jittered exponential backoff, and bulkhead-isolated dependency pools
2. Increase the platform-wide global request timeout to 60 seconds so slow calls succeed instead of being cut off early
3. Disable retries across the entire platform so cascading retry storms cannot compound the underlying slowness
4. Refactor all microservices back into a single monolith so cross-service latency disappears from the request path

Answer: A — Add per-service circuit breakers, tighter timeouts, jittered exponential backoff, and bulkhead-isolated dependency pools

A) Circuit breakers + timeouts + backoff + bulkheading is the canonical resilience bundle. B) Longer timeouts extend pain. C) No retries loses recovery. D) Reverses architecture.

On-demand DynamoDB baseline is 8000 RCU / 2500 WCU with occasional 3x spikes and a high bill. Which optimization is BEST?

1. Migrate the workload from DynamoDB to RDS PostgreSQL to reduce the cost of ownership of the data layer over time
2. Reduce TTL on tables so items expire sooner and less capacity is consumed serving reads over the same working set
3. Switch to provisioned capacity with auto-scaling at 70% utilization sized to steady-state plus Reserved Capacity discount
4. Stay on on-demand pricing indefinitely and accept the current bill as the cost of predictable throughput headroom

Answer: C — Switch to provisioned capacity with auto-scaling at 70% utilization sized to steady-state plus Reserved Capacity discount

C) Provisioned + auto-scaling + Reserved Capacity typically saves 50%+ vs on-demand for predictable workloads. A) Cross-engine migration is invasive. B) TTL affects correctness. D) Ignores the ask.

Nightly on-demand EMR clusters have grown costly. Which optimization is BEST?

1. Buy 3-year Reserved Instances covering the entire EMR fleet for all instance families running any nightly job
2. Use EMR with task nodes on Spot, instance fleets to diversify, and consider EMR Serverless or EMR on EKS for elasticity
3. Migrate the ETL pipeline off EMR and onto Amazon Aurora with stored procedures replacing Spark transformations
4. Replace EMR entirely with AWS Lambda functions triggered by S3 events to handle the nightly batch job workload

Answer: B — Use EMR with task nodes on Spot, instance fleets to diversify, and consider EMR Serverless or EMR on EKS for elasticity

B) Spot task nodes + instance fleets + optionally EMR Serverless is the canonical EMR cost optimization. A) 3-year RIs for nightly-only wastes daytime. C) Aurora is not an ETL engine. D) Lambda 15-min/memory caps unsuitable.

Blue/Green deployment on AWS is BEST characterized as:

1. Rolling in-place updates across each instance one at a time without a parallel environment
2. Deploying new versions to a small percentage of traffic and increasing gradually over hours
3. Running two production-like environments and shifting traffic to the new one after tests
4. Redeploying only during scheduled maintenance windows without traffic-shifting capability

Answer: C — Running two production-like environments and shifting traffic to the new one after tests

C) Blue/Green runs a parallel environment then shifts traffic. A) That is rolling deployment. B) That is canary. D) That describes a maintenance-window approach, not blue/green.

Which statement BEST describes the pillars of the AWS Well-Architected Framework?

1. Speed, Profit, Delivery, and Free-Tier Utilization as the four pillars of every AWS workload
2. Security, Reliability, and Performance are the only three pillars that require a formal review
3. Cost, Compute, Storage, and Network as the four foundational pillars of AWS architecture
4. Operational Excellence, Security, Reliability, Performance Efficiency, Cost Optimization, and Sustainability

Answer: D — Operational Excellence, Security, Reliability, Performance Efficiency, Cost Optimization, and Sustainability

D) The six pillars are the canonical framework. A) These are not framework pillars. B) The framework has six pillars, not three. C) These are resource categories, not the framework pillars.

An ECS cluster has over-provisioned and CPU-throttled services in the same fleet. Which improvement is BEST and most automated?

1. Manually right-size each of the 30 services once per quarter based on Container Insights CPU and memory usage graphs
2. Enable Compute Optimizer for ECS, use service auto-scaling on CPU/memory, and consider Fargate for variable services
3. Migrate every ECS service to AWS Lambda functions to avoid all container capacity planning across the platform
4. Triple the ECS cluster's EC2 capacity to eliminate throttling and accept the higher fixed cost across the fleet

Answer: B — Enable Compute Optimizer for ECS, use service auto-scaling on CPU/memory, and consider Fargate for variable services

B) Compute Optimizer + service auto-scaling + Fargate is AWS-native continuous rightsizing. A) Quarterly manual is slow. C) Lambda does not fit all workloads. D) Makes the utilization problem worse.

A microservices platform needs distributed tracing, correlated logs, and per-service metrics with minimal code changes. Which combination is BEST?

1. Application-level printf logs written to local instance disks and inspected via SSH
2. A single CloudWatch alarm per service that emails engineers whenever it breaches
3. Amazon S3 access logs alone as the tracing and metrics source across every service
4. AWS X-Ray for tracing with CloudWatch Logs Insights and metrics via Container Insights

Answer: D — AWS X-Ray for tracing with CloudWatch Logs Insights and metrics via Container Insights

D) X-Ray + Logs Insights + Container Insights is the AWS observability stack. A) Local logs do not correlate across services. B) One alarm per service is not tracing or metrics. C) S3 access logs cover S3 only.

200 Lambdas log at INFO with 100 KB JSON per invocation and never-expire retention, driving up CloudWatch bill. Which is BEST?

1. Disable CloudWatch entirely across all Lambdas to eliminate the log ingest and storage cost from the bill
2. Set per-group retention, reduce verbosity, archive older logs to S3 via subscription filter + Firehose, and use Logs Insights
3. Stop using AWS Lambda for the 200 workloads and move them all onto EC2 instances with local file logging
4. Print all logs to standard output only so nothing is captured by CloudWatch Logs across the account

Answer: B — Set per-group retention, reduce verbosity, archive older logs to S3 via subscription filter + Firehose, and use Logs Insights

B) Retention + verbosity + archive to S3 + Logs Insights is the AWS-prescriptive CW Logs optimization. A) Loses observability. C) Overkill. D) Lambda stdout still goes to CW Logs.

Aurora MySQL cluster at 80 TB has 4-hour analytical queries blocking OLTP. Which improvement is BEST?

1. Migrate all Aurora workloads including transactional traffic to Amazon Redshift for both OLAP and OLTP simultaneously
2. Run the 4-hour analytics jobs directly on the Aurora writer during off-peak hours to reduce cost of separate infrastructure
3. Migrate the underlying database engine from Aurora MySQL to Amazon RDS PostgreSQL to gain analytical query performance
4. Add a parallel-query reader endpoint for analytics, export to S3 for Athena/Redshift, and consider Aurora Zero-ETL to Redshift

Answer: D — Add a parallel-query reader endpoint for analytics, export to S3 for Athena/Redshift, and consider Aurora Zero-ETL to Redshift

D) Aurora parallel query + Zero-ETL to Redshift is the modern OLTP/OLAP separation pattern. A) Over-migrates. B) Blocking writer is the current problem. C) Engine swap doesn't help.

Cost Explorer Rightsizing recommendations?

Analyzes EC2 usage, suggests smaller types or termination.

Managed Grafana / Managed Prometheus?

AWS-managed open-source observability. Replaces self-hosted setups.

Backpressure pattern?

Slow upstream when downstream overloaded. SQS naturally provides this.

S3 lifecycle to Glacier?

Auto-transition old data to Glacier (cheaper) for archive.