You have an application that uses Vertex AI Feature Store to manage and serve product features for real-time recommendations. You want to monitor the performance and health of the application. You need to understand the overall duration of a request. What should you do?
Answer(s): C
C: Measure the Latency of your requests. Latency is the direct metric that tracks the overall duration of a request (the total time elapsed between a client sending a request and receiving the response from Vertex AI Feature Store). Monitoring read/write latency is critical for real-time recommendation engines to ensure that feature retrieval remains fast enough to meet strict user-experience timeouts.
A large healthcare provider's primary electronic health record (EHR) application runs on Compute Engine instances with a Cloud SQL for PostgreSQL database, all located in the us-west1 region. A new regulatory mandate requires you to implement and document a business continuity plan (BCP). This plan must ensure that the EHR application can be fully recovered and operational in a different geographical region with a recovery time objective (RTO) of two hours and a recovery point objective (RPO) of 15 minutes. You need to design a disaster recovery strategy that meets these strict BCP requirements. What should you do?
Answer(s): A
Why option A is the correct choiceActive-active multi-region deployment – Managed Instance Groups (MIGs) are running in both us-west1 and us-east1, so the application is already warm-standby in the DR region. Global external HTTP(S) Load Balancer – Provides a single DNS name and performs health-checks against the MIGs in both regions; when a health-check fails, traffic is automatically redirected to the healthy region, delivering the required RTO ≤ 2 h with near-zero manual intervention. Cross-region Cloud SQL read replica – A synchronous or asynchronous replica in us-east1 keeps the PostgreSQL data within the 15-minute RPO window; in a fail-over the replica can be promoted automatically (or with a single CLI call) to become the new primary. Fully automated fail-over – Health-check-driven load-balancer routing plus replica promotion eliminates manual steps, satisfying the strict BCP automation expectations of the exam.Why the other options are less suitableOption B – Although it uses a cross-region replica, the fail-over process is described as “build the environment in us-east1 and promote the replica,” which implies a manual or semi-manual step and does not address application-level load-balancing or health-check automation. It also introduces Terraform unnecessarily for a DR design that must be operationally resilient. Option C – Relying on daily snapshots and manual restoration of VMs and the database cannot meet a 15-minute RPO or a 2-hour RTO in practice; restoration from snapshots typically takes longer and introduces significant recovery effort. Option D – A regional MIG in a single zone/region cannot provide geographic redundancy; it depends on the Google Cloud SLA to keep the region online, which does not satisfy the requirement to operate in a different geographical region.Result – Option A delivers a true multi-region, automatically fail-over-capable architecture that meets both the RTO and RPO constraints while leveraging Google-managed services (MIGs, Global HTTP(S) Load Balancer, Cloud SQL read replica).
Managed Instance Groups – Autoscaling and Load Balancing : https://cloud.google.com/compute/docs/autoscaling/managed-instance-groups Cross-region Cloud SQL read replicas : https://cloud.google.com/sql/docs/postgres/read-replicas#cross-region-replicas Global HTTP(S) Load Balancing : https://cloud.google.com/load-balancing/docs/https/setup-httpsThese documents provide the technical details needed to design and operate the described disaster-recovery architecture on Google Cloud Platform.
A financial services company is decommissioning one of its on-premises data centers. As part of this initiative, the company needs to perform a one-time migration of 500 ТВ of historical transaction archives to a Cloud Storage bucket for long-term retention. The data center’s internet egress is 1 Gbps, which is shared with critical business operations. You must complete the secure data transfer within a 60-day window to meet the decommissioning deadline. What should you do?
Answer(s): D
Why option D is the optimal choiceData volume vs. bandwidth: 500 TB at a sustained 1 Gbps (≈125 MB/s) would require roughly 47 days of uninterrupted, non-shared throughput. Because the 1 Gbps link is also used by critical business operations, the effective available bandwidth for migration will be far lower, making it impossible to meet the 60-day deadline reliably. One-time, large-scale transfer: A Transfer Appliance is purpose-built for “bulk, one-off” migrations where network bandwidth is constrained. It allows you to copy the data to the appliance over your high-speed local LAN, then ship the sealed unit to Google where Google uploads the data directly into Cloud Storage, bypassing the limited WAN link entirely. Security & integrity: The appliance encrypts data at rest and in transit, and provides checksum verification on upload, ensuring data integrity without additional scripting or custom security measures. Operational simplicity: No need to develop or maintain custom scripts, schedule parallel uploads, or manage concurrent transfers that could interfere with production workloads. The process is a single, well-defined logistics operation.Why the other options are less suitableOption A – Partner Interconnect 10 Gbps: Procuring a 10 Gbps Interconnect would require additional provisioning time, contractual overhead, and would still share the link with business traffic unless fully isolated, which is not indicated. It also does not address the immediate need for a one-time migration within the tight window. Option B – Parallel gsutil / gcloud storage cp over public internet: Uploading 500 TB via the public internet would saturate the already-shared 1 Gbps link, potentially degrading critical services. The approach also lacks built-in resumability, encryption, and verification, requiring extensive custom scripting to handle retries and throttling. Option C – Agent-based Storage Transfer Service: This service is designed for recurring, incremental transfers from VMs or on-premises endpoints that have reliable network connectivity. With a 1 Gbps shared link, the transfer would be too slow and could impact production workloads, making it impractical for a single massive migration.Conclusion Given the constraints of limited bandwidth, a strict 60-day deadline, and the need for a secure, single-shot migration of a large dataset, the Transfer Appliance (Option D) provides the only solution that guarantees completion within the window while preserving network performance for ongoing business operations.
Transfer Appliance Overview: https://cloud.google.com/transfer-appliance/docs/overview Migrating Data to Cloud Storage with Transfer Appliance: https://cloud.google.com/transfer-appliance/docs/migrate-data-cloud-storageThese documents detail the appliance workflow, capacity planning, and security considerations relevant to the scenario described.
Your company wants to optimize Google Cloud costs for their development and staging environments. These environments are workstations used by developers Monday through Friday, 9:00 AM to 6:00 PM local time. Currently, the environments run on a fleet of nl-standard-4 Compute Engine instances that operate 24/7, leading to a high monthly cost for resources that are idle more than 70% of the time. You need to implement a solution that significantly reduces the monthly cost of these non-production environments without impacting the development team's productivity during work hours. What should you do?
JustificationOption A – Schedule VM start/stop to match the team’s work scheduleBy configuring the nl-standard-4 instances to start at 09:00 and stop at 18:00 on weekdays, the VMs run only during the 9-hour daily window and are completely shut down overnight and on weekends. This directly eliminates the ~70 % idle time, cutting the monthly compute charge proportionally (≈ 30 % of the original cost). The approach requires only a simple instance-schedule policy in the Compute Engine console or Deployment Manager, preserving developer productivity during the required hours and incurring no architectural redesign.Option B – Purchase 3-year Committed Use Discounts (CUDs)CUDs provide up to 70 % discount on the per-hour rate only if the VMs remain running for the full commitment period . Since the current workload is idle most of the time, committing to sustained usage would lock in payment for capacity that is never fully utilized, negating the cost-saving intent.Option C – Upgrade to N2 machine seriesMoving to N2 instances may improve CPU-to-memory ratios or pricing per vCPU, but the cost reduction is modest and does not address the fundamental issue of over-provisioned, always-on resources. Without a schedule or rightsizing, the same idle time persists, so the expected cost benefit is limited.Option D – Re-architect to a regional managed instance group with autoscalingAutoscaling is valuable for workloads with variable demand, but dev/staging environments have a predictable, fixed schedule. Introducing a regional MIG adds operational complexity, requires additional networking and load-balancing configuration, and does not inherently reduce cost unless combined with pre-emptible VMs or strict scheduling—both of which are not mentioned. For a simple start/stop use case, this option is unnecessarily heavyweight.Conclusion Implementing a start/stop schedule (Option A) directly aligns VM availability with the developers’ working hours, delivering the greatest cost reduction while maintaining productivity and requiring minimal operational overhead. The other options either fail to address the idle-time problem or introduce unnecessary complexity and cost.
Scheduling VM instances: https://cloud.google.com/compute/docs/instances/scheduling-vm-instances Committed use discounts: https://cloud.google.com/products/committed-use-discounts
A retail company s most critical application is its online payment processing system. The business has a requirement that the system must be able to survive a complete zonal outage while minimizing cost. You need a design solution that can handle a zonal failure. What should you do?
Answer(s): B
B: Deploy the application on a regional MIG to provide high availability across multiple zones in the primary region. ChoiceSurvives Zonal A regional Managed Instance Group (MIG) automatically distributes application instances evenly across multiple availability zones within a single region. If a complete zonal outage occurs, the underlying health checks automatically redirect traffic to instances running in the surviving zones.Minimizes Cost: This solution perfectly balances the requirement to survive a zonal failure while minimizing cost. Keeping traffic and computing footprints constrained to a single region eliminates the massive cross-region data transfer fees, multi-region database replication costs, and idle dual-region compute resources required by multi-regional setups.
Post your Comments and Discuss Google Google Cloud Architect Professional exam prep with other Community members:
Our website is free, but we have to fight against AI bots and content theft. We're sorry for the inconvenience caused by these security measures. You can access the rest of the Google Cloud Architect Professional content, but please register or login to continue.