Your company is a global financial services provider that processes and analyzes a high volume of credit card transactions in real time for fraud detection. Your analytics team must run complex batch queries on the same transaction data for daily reporting. You need to design a data processing solution that can handle both real-time and batch processing of the transaction data while minimizing operational overhead and infrastructure management. What should you do?
Answer(s): B,D
Why option BD is the optimal designDataflow (B) is the recommended service for real-time streaming ingestion and transformation of high-volume transaction data. It offers fully managed, autoscaling pipelines that can apply windowed aggregations, enrichments, and anomaly detection with exactly-once semantics, eliminating the need to operate dedicated streaming servers. BigQuery (D) is the most cost-effective, petabyte-scale analytics engine for batch reporting. It can ingest the processed data from Dataflow (or directly from Cloud Storage) and run complex SQL queries for daily fraud-detection reports with virtually zero administration.Why the other options are less suitableA – Dataprep is primarily a data-preparation tool for cleaning and shaping data before loading; it does not provide a scalable ingestion or processing engine for high-throughput streaming workloads. C – Dataproc requires manual cluster provisioning, scaling, and patching. While it can run Spark jobs for both streaming (via Structured Streaming) and batch, the operational overhead contradicts the requirement to “minimize infrastructure management.” E – Firestore is a NoSQL document store optimized for low-latency reads/writes of user-level data, not for large-scale analytical processing of transactional streams or complex batch queries.Overall solution1. Ingest and stream-process credit-card transactions with Dataflow , leveraging its managed service to handle scaling and fault tolerance. 2. Persist the cleaned, enriched data to a storage tier (e.g., Cloud Storage or BigQuery tables). 3. Run daily batch analytics and reporting on the same dataset using BigQuery , which provides fast, petabyte-scale query performance with minimal operational effort.
Dataflow documentation: https://cloud.google.com/dataflow/docs BigQuery documentation: https://cloud.google.com/bigquery/docs
Your company uses a custom-built application running on a Compute Engine virtual machine (VM). This application processes real-time sales data and writes it to a zonal Persistent Disk. A recent internal audit requires that you implement a backup and recovery plan to protect against zonal failures. Your company has a strict policy that all backup data must be retained for at least 90 days and stored in a separate project with limited access. You need to implement a fully automated backup solution that meets these requirements with minimal operational overhead. What should you do?
Answer(s): D
Why option D is the optimal choiceFully managed backup service – Backup and Disaster Recovery (DR) is a native GCP service that automates snapshot creation, scheduling, and retention without requiring custom scripting or manual intervention. Retention-policy enforcement – The DR service lets you define a 90-day retention period directly on the backup vault, guaranteeing compliance with the audit requirement. Cross-project isolation – Backup vaults can be created in a separate project, enabling strict IAM controls and limiting access to the backup data. Minimal operational overhead – All lifecycle actions (snapshot scheduling, vault management, deletion) are handled by the service, reducing the need for ongoing maintenance of scripts or external storage resources.Why the other options are less suitableOption A – Relies on a custom script and manual copying of snapshots to another zone. It does not provide built-in retention management, requires ongoing script maintenance, and does not enforce cross-project isolation automatically. Option B – Uses Cloud Storage for snapshot storage and a lifecycle rule, but snapshots stored in Cloud Storage are not native GCP snapshots; they must be recreated from exported images, adding complexity and not offering a direct 90-day retention guarantee for snapshots. Option C – Involves a snapshot schedule plus a script to move data to a multi-regional bucket. While it can meet retention via bucket lifecycle, it still requires custom scripting, does not provide a dedicated backup vault with IAM boundaries, and introduces additional operational steps.Conclusion Option D leverages Google’s native Backup and DR service to deliver an automated, policy-driven, cross-project backup solution that meets the 90-day retention and minimal-overhead requirements, making it the most appropriate answer.
Backup and DR service overview: https://cloud.google.com/backup Configuring retention policies for backups: https://cloud.google.com/backup/docs/backup-vaults/retention-policies
Your ecommerce web application includes many products with pictures and videos. You need to improve the application's page load speed and reduce the latency for customer requests. What should you do? (Choose two.)
Answer(s): B,C
Why option B and option C are the best choicesB – Use Memorystore for RedisProvides a fully managed, low-latency in-memory data store that can cache product metadata, session state, or frequently accessed catalog information. By reducing round-trips to the primary database, response times for API calls that retrieve product data drop dramatically, directly improving page-load speed for customers. Memorystore scales automatically and offers multi-AZ replication, ensuring high availability without additional operational overhead.C – Implement CDN with the application’s external HTTPS load balancerA Content Delivery Network caches static assets (images, videos, CSS/JS) at edge locations close to end users, cutting the round-trip time for these resources. When paired with the external HTTPS load balancer, the CDN can serve cached content over HTTPS without hitting the backend compute services, further lowering latency for end-user page loads. This is the most direct way to accelerate delivery of the rich media that dominates an ecommerce storefront.Why the other options are less suitableA – Deploy to Cloud Run and enable concurrencyWhile Cloud Run can scale stateless services, concurrency mainly affects throughput, not the latency of static asset delivery. It does not address the primary bottleneck for page-load speed in an ecommerce UI.D – Change the database from Bigtable to SpannerBigtable and Spanner serve different workloads; migrating would not inherently reduce latency for serving images or videos. The issue is not the database type but the delivery of static content and repeated DB fetches, which are better solved with caching/CDN.E – Deploy to a GKE Autopilot cluster instead of a regular GKE clusterAutopilot simplifies cluster management but does not provide a built-in mechanism for accelerating static asset delivery or reducing client-side latency. The performance gain would be indirect and not as targeted as a CDN.
Cloud CDN documentation: https://cloud.google.com/cdn/docs Memorystore for Redis documentation: https://cloud.google.com/memorystore/docs/redis
You need to build and deploy a containerized web application to Google Cloud. The application is very write-heavy and requires a relational database as its data store. The application needs to be highly available in multiple cloud regions. You want to minimize operational overhead while following Google-recommended practices. What should you do?
Answer(s): B
Why option B is the best choiceContainer-native, serverless runtime – Cloud Run executes container images directly, automatically scales to zero and to any load, and requires no cluster management, which minimizes operational overhead. Global, highly available front-end – Deploying Cloud Run services in multiple regions behind a global HTTPS load balancer provides a single, anycast DNS name with health-checked traffic routing across regions, delivering true high availability and low latency for users worldwide. Strong relational semantics – Cloud Spanner is Google’s fully managed, globally distributed relational database that offers ACID transactions, strong consistency, and horizontal scaling. It satisfies the write-heavy workload while preserving relational capabilities that Cloud SQL or Firestore lack at scale. Operational simplicity – All components (Cloud Run, Cloud Spanner, global load balancer) are fully managed services; you do not need to provision or patch VMs, manage storage classes, or configure database replication manually.Why the other options are less suitableOption A – GKE clusters require manual node management, autoscaling policies, and cluster upgrades, increasing operational burden. Cloud SQL is a single-region relational database; to achieve multi-region high availability you would need complex failover setups, which adds overhead. Option C – While Firestore can be multi-region, it is a NoSQL document store, not a relational database, so it does not meet the requirement for a relational data model. GKE adds the same operational overhead as in A. Option D – Bigtable is a wide-column NoSQL store; it does not provide relational features such as joins or ACID transactions, making it inappropriate for a workload that explicitly needs a relational database. Additionally, GKE’s global ingress is less feature-rich and less integrated than the global HTTPS load balancer used with Cloud Run.
Cloud Run documentation: https://cloud.google.com/run/docs Cloud Spanner overview: https://cloud.google.com/spanner/docs/overview
Your machine learning (ML) engineers use self-hosted Jupyter notebooks for tasks such as data preparation, model training, and fine-tuning. The operations team then deploys these models in various environments. You want to provide maximum flexibility for ML engineers, promote collaboration with a common toolset, and leverage Google Cloud's scalability, while following Google-recommended practices. What should you do?
Answer(s): C
JustificationOption C – Vertex AI for ML + MLOps for deployment aligns with Google-recommended best practices.Vertex AI provides a unified, fully-managed environment for training, experimentation, and serving models, giving data scientists the flexibility to use notebooks, custom containers, or pre-built algorithms while maintaining governance and collaboration. Integrated MLOps capabilities (Model Registry, pipelines, automated testing, and continuous monitoring) enable seamless, repeatable deployments across dev, test, and production environments, satisfying the operations team’s need for reliability and scalability. This end-to-end solution leverages Google Cloud’s serverless and auto-scaling infrastructure, ensuring optimal resource utilization and cost-effectiveness.Option A – AutoML + Cloud Deploy is too restrictive for custom model development; AutoML is designed for “no-code” use cases and does not support the full lifecycle of self-hosted notebook workflows.Option B – Colab Enterprise + DevOps mixes a collaboration tool (Colab) with a generic DevOps approach, lacking native MLOps features such as model versioning, automated rollback, and model monitoring that are essential for production ML pipelines.Option D – TensorFlow + Cloud Deploy focuses only on the deployment artifact and ignores the managed training and experiment tracking capabilities that Vertex AI provides, limiting flexibility and collaboration for ML engineers.Conclusion Vertex AI combined with dedicated MLOps practices delivers the required flexibility, collaboration, and Google-scale infrastructure while adhering to the platform’s recommended architecture for end-to-end machine-learning workflows.
Vertex AI Documentation: https://cloud.google.com/vertex-ai/docs MLOps on Google Cloud (Model Registry & Pipelines): https://cloud.google.com/vertex-ai/docs/ml-ops/model-registry-pipeline-overview
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.