Cloud Architect Advanced Practice Exam: Hard Questions 2025

A global retailer is migrating a multi-region commerce platform to Google Cloud. The platform has microservices, a shared database tier, and a new analytics pipeline. Requirements: (1) 99.99% availability for checkout, (2) RPO <= 1 minute and RTO <= 5 minutes for the database, (3) minimize operational burden, and (4) handle regional outages without manual failover. Which architecture best meets these requirements?

A financial services company must segment workloads by environment (dev/test/prod) and by business unit. They require centralized security controls, a shared networking hub with multiple VPCs, and least-privilege access where teams can self-service project creation but cannot weaken org policies. They also need to onboard acquired companies with minimal refactoring. What is the best Google Cloud resource hierarchy and governance approach?

A data science platform uses Vertex AI and BigQuery. Data is stored in Cloud Storage and must be accessed by notebooks running in a managed environment. The security team mandates: (1) no long-lived service account keys, (2) per-user attribution in audit logs when accessing BigQuery datasets, and (3) fine-grained restriction so only a subset of users can access specific datasets based on group membership. Which approach best satisfies these requirements?

An enterprise runs sensitive workloads on GKE. They need to ensure that only images from an approved registry can run, pods cannot escalate privileges, and secrets are never stored in plaintext in etcd. They also must demonstrate policy enforcement for compliance audits. Which combination best meets these requirements with strong, centralized controls?

A company runs a multi-tenant SaaS on Cloud Run. Tenants require private connectivity from on-premises to the service, and the company must ensure tenant A cannot access tenant B’s data even if they guess URLs. The platform team also needs to rotate encryption keys regularly and immediately revoke a compromised tenant’s access without redeploying services. What is the best design?

A company manages 300 projects with Terraform and uses a CI/CD pipeline. They recently enabled an organization policy that disables service account key creation. After the change, multiple pipelines fail when applying Terraform because the runners previously used downloaded JSON keys. They need a secure, scalable fix that avoids static secrets and works for both Google-hosted and external CI runners. What should they implement?

You operate a fleet of GKE clusters across multiple regions. A sudden rollout caused intermittent 503 errors for a subset of users. Observations: (1) errors correlate with new pods starting, (2) metrics show increased CPU throttling on nodes, (3) HPA is scaling but new pods take long to become Ready, and (4) preStop hooks execute for 30 seconds during termination. You need to reduce user impact during rollouts while maintaining cost efficiency. What is the best remediation plan?

A company’s batch pipeline processes 50 TB/day and writes results to BigQuery. They use Dataflow streaming for near-real-time ingestion and a separate Dataflow batch job for daily recomputation. They are experiencing unpredictable batch completion times and occasional BigQuery load quota errors. They must maintain end-to-end SLA and minimize reprocessing costs. What should you do?

A company wants to implement chargeback and optimize spend for a shared platform running in multiple projects. Requirements: (1) attribute costs to internal teams using labels, (2) detect anomalies in near real time, (3) optimize compute usage by rightsizing and committed use planning, and (4) ensure recommendations do not violate SLOs. What is the best approach?

You are leading an implementation where an on-prem Oracle database is being replaced by Cloud SQL for PostgreSQL. The application has strict cutover windows and must support dual writes for a short period. The team has already built Terraform modules for infrastructure but deployments are inconsistent across environments, and rollback is error-prone. You need to improve implementation reliability and governance while enabling progressive delivery. What should you do?