Cloud Engineer Case Studies & Practical Application 2 — Questions and Answers
Question 1: A fintech startup experiences intermittent latency spikes in their payment processing API hosted on AWS. CloudWatch shows CPU and memory are normal, but RDS connection errors appear in logs. What is the most likely cause?
- Insufficient EC2 instance size
- RDS connection pool exhaustion under peak load (Correct answer)
- S3 bucket policy blocking API calls
- VPC peering misconfiguration
Correct answer: RDS connection pool exhaustion under peak load
Connection pool exhaustion occurs when concurrent requests exceed the max_connections limit on RDS, causing new connections to fail even though compute resources appear healthy.
Question 2: A retail company migrates an on-premises Oracle database to Google Cloud. After migration, batch ETL jobs run 3x slower than before. The Cloud SQL instance matches the on-premises CPU/RAM specs. What should the engineer investigate first?
- Increase Cloud SQL storage IOPS and check disk throughput limits (Correct answer)
- Upgrade to a larger machine type with more vCPUs
- Enable Cloud SQL automatic backups
- Switch from Cloud SQL to BigQuery
Correct answer: Increase Cloud SQL storage IOPS and check disk throughput limits
Cloud SQL storage IOPS are provisioned based on disk size; if the migrated database uses fast SAN storage on-premises, the default Cloud SQL disk may be the bottleneck.
Question 3: A media company uses Azure Blob Storage for video assets. Egress costs are unexpectedly high despite 80% of traffic coming from users in the same Azure region. What configuration is likely missing?
- Azure CDN is not configured to serve assets from a nearby edge node (Correct answer)
- Blob Storage replication is set to GRS instead of LRS
- The storage account tier is set to Cool instead of Hot
- Azure Traffic Manager is routing users to the wrong region
Correct answer: Azure CDN is not configured to serve assets from a nearby edge node
Without Azure CDN, all requests hit origin Blob Storage and incur egress charges; CDN caches assets at edge nodes, eliminating repeated origin egress for popular content.
Question 4: An e-commerce platform's checkout service on Kubernetes crashes every night at 2 AM. Pod logs show OOMKilled. The team set a 512Mi memory limit. Nightly batch jobs run at 1:45 AM. What is the best fix?
- Increase the pod memory limit and add a Vertical Pod Autoscaler (Correct answer)
- Schedule batch jobs to run at a different time only
- Restart the checkout deployment via a CronJob at 2 AM
- Add a liveness probe that restarts the pod on memory pressure
Correct answer: Increase the pod memory limit and add a Vertical Pod Autoscaler
The batch jobs likely cause memory spikes beyond 512Mi; raising the limit and using VPA ensures the pod has headroom while right-sizing automatically over time.
Question 5: A SaaS company stores tenant data in separate GCP projects for isolation. Cross-project BigQuery queries take 40 seconds. The same data in a single project queries in 2 seconds. What should the engineer recommend?
- Use BigQuery Authorized Views to expose tenant datasets within a shared project (Correct answer)
- Move all tenants back to a single project
- Enable BigQuery BI Engine on each tenant project
- Increase BigQuery slot reservations in each project
Correct answer: Use BigQuery Authorized Views to expose tenant datasets within a shared project
Authorized Views allow cross-project data access without data movement while maintaining IAM-level isolation, avoiding the network overhead of cross-project queries.
Question 6: A logistics company runs a Node.js microservice on AWS Lambda. Cold starts average 4 seconds, causing SLA breaches. The function uses a 256MB memory configuration. What change has the most impact on cold start time?
- Enable Lambda Provisioned Concurrency for the function (Correct answer)
- Increase function timeout from 30s to 60s
- Move the function to a VPC subnet
- Switch the runtime from Node.js 18 to Node.js 16
Correct answer: Enable Lambda Provisioned Concurrency for the function
Provisioned Concurrency pre-initializes Lambda execution environments, eliminating cold starts entirely for the pre-warmed instances.
Question 7: A healthcare company must ensure PHI data in Azure never leaves a specific geographic region due to regulatory requirements. They use multiple Azure services. Which control enforces this most comprehensively?
- Azure Policy with a 'Allowed locations' deny effect applied at the subscription level (Correct answer)
- Configuring each service's geo-replication settings individually
- Using Azure Private Endpoints for all services
- Enabling Azure Defender for all resource types
Correct answer: Azure Policy with a 'Allowed locations' deny effect applied at the subscription level
An Azure Policy deny effect at the subscription level prevents any resource from being deployed outside allowed regions, providing a guardrail that cannot be bypassed per-service.
Cloud Engineer Case Studies & Practical Application 2
A fintech startup experiences intermittent latency spikes in their payment processing API hosted on AWS.
CloudWatch shows CPU and memory are normal, but RDS connection errors appear in logs.
What is the most likely cause?