LevelUpWith

Explore / Google Cloud Professional Cloud Architect Exam Prep / Practice questions

Google Cloud Professional Cloud Architect practice questions

10 questions from the bank behind this course's timed mock. Free to read and to answer, no account needed, and each has its answer and an explanation. They belong to Google Cloud Professional Cloud Architect Exam Prep.

Answer them one at a time →Read lesson 1 free

  1. Question 1

    A cloud architect is answering a practice scenario question. The company needs an internal reporting tool. The stated requirements are "lowest possible cost" and "the business can tolerate several hours of downtime." All four answer options meet the functional need. Using the four-step method, which option should the architect most likely cross out as an over-engineered distractor?

    1. A multi-region deployment with automatic failover between regions
    2. A single-zone deployment with scheduled backups
    3. A single-region deployment that uses managed services
    4. A managed service configured with Google's secure-by-default settings
    Show the answer

    Answer: A. A multi-region deployment with automatic failover between regions

    Multi-region failover adds resilience the scenario does not need and raises cost, which breaks the 'lowest possible cost' requirement. That makes it a classic over-engineered distractor. The single-region managed option is the nearest alternative, but it stays within the stated cost and downtime limits, so it should not be crossed out.

  2. Question 2

    An architect is reviewing an order-processing service. The front end runs on a regional managed instance group behind a load balancer, spread across multiple zones, with a 99.99% SLA. It calls a database on a single Compute Engine instance in one zone, with a 99.9% SLA. The business needs the whole service to reach 99.99% availability and to survive a zone outage. It does not need to survive a region outage. What should the architect do at the lowest cost?

    1. Move the database to Cloud SQL Enterprise Plus edition with high availability (regional availability).
    2. Add more instances to the regional managed instance group so the front end has spare capacity.
    3. Migrate the database to a Spanner multi-region instance.
    4. Deploy the front end to a second region and route traffic globally between the two regions.
    Show the answer

    Answer: A. Move the database to Cloud SQL Enterprise Plus edition with high availability (regional availability).

    Availabilities multiply in series, so the 99.9% single-zone database limits the whole service to about 99.89%. A regional Cloud SQL Enterprise Plus HA database (99.99% SLA) removes that weak tier. Spanner multi-region would also meet the target, but it survives a region outage the business does not need, so it costs more than the requirement justifies.

  3. Question 3

    A platform engineering team at a logistics company needs to build an agent that coordinates several specialized sub-agents. It must be defined in code, version-controlled and deployed through the team's existing CI/CD pipeline. The team also wants to be able to switch to a non-Gemini model later. What should the architect recommend?

    1. Reserve TPU capacity on AI Hypercomputer and serve a custom-trained orchestration model
    2. Build the agent with the Agent Development Kit (ADK) and deploy it to Agent Runtime
    3. Build the agent in Agent Studio so business users can assemble it with little code
    4. Make one Gemini API call per request and keep the conversation state in the client application
    Show the answer

    Answer: B. Build the agent with the Agent Development Kit (ADK) and deploy it to Agent Runtime

    ADK is the code-first agent framework. It supports networks of sub-agents and works with any model, and Agent Runtime deploys and runs the result, which fits developers who need full control and CI/CD. Agent Studio is real but is designed for building agents quickly with little code, not for code-defined agents in a pipeline.

  4. Question 4

    A retail company runs a customer-facing API on Cloud Run in us-central1 and europe-west1. The security team requires Cloud Armor and a custom domain in front of the service, and users in both regions should reach their nearest deployment. As the cloud architect, which load balancer configuration should you recommend?

    1. A global external Application Load Balancer with one serverless NEG per region, plus a health check attached to each serverless NEG to detect regional failures
    2. A global external Application Load Balancer with one backend service that contains one serverless NEG for each of the two regions
    3. An external proxy Network Load Balancer with a serverless NEG in each region
    4. A global external Application Load Balancer with one backend service that mixes the serverless NEGs and a zonal GCE_VM_IP_PORT NEG for fallback
    Show the answer

    Answer: B. A global external Application Load Balancer with one backend service that contains one serverless NEG for each of the two regions

    Serverless NEGs work only with Application Load Balancers, and a global external Application Load Balancer allows one serverless NEG per region in a backend service, so deploying to two regions serves both user groups. The health-check option is wrong because serverless NEGs do not support health checks, and you cannot mix serverless NEGs with other NEG types in one backend service.

  5. Question 5

    An insurance company's contact centre records thousands of customer phone calls each week. The claims manager wants a short written summary of each call stored with the claim record. An architect must design the solution using Google AI APIs, keeping it as managed as possible. What is the best design?

    1. Use the Video Intelligence API to analyse the call recordings and extract labels for each conversation
    2. Use Conversational Agents to replay each recorded call as a multi-turn dialogue and record a summary
    3. Use Speech-to-Text to transcribe each recording, then pass the transcript to a separate step that summarises the text
    4. Use Text-to-Speech to process each recording and produce a spoken summary that is attached to the claim
    Show the answer

    Answer: C. Use Speech-to-Text to transcribe each recording, then pass the transcript to a separate step that summarises the text

    Audio work is a pipeline: Speech-to-Text turns the call into a transcript, and a separate step then summarises that text. Text-to-Speech is the closest wrong option because it is also an audio API, but it turns text into speech, which is the opposite direction to what this task needs.

  6. Question 6

    EHR Healthcare runs a Dataflow pipeline inside an enforced VPC Service Controls perimeter that restricts Cloud Storage and BigQuery. The pipeline must now write daily aggregate files to a Cloud Storage bucket in one named partner project outside the perimeter. The security team does not want to weaken protection for any other data flow. What should the architect do?

    1. Add an egress rule that allows the pipeline's service account to call Cloud Storage write methods only in the named partner project
    2. Add the partner project to the existing perimeter so the pipeline's calls no longer cross the boundary
    3. Create an access level for the partner's IP range and attach it to the perimeter
    4. Grant the pipeline's service account the Storage Object Creator role on the partner bucket
    Show the answer

    Answer: A. Add an egress rule that allows the pipeline's service account to call Cloud Storage write methods only in the named partner project

    An egress rule is a narrow from/to exception for traffic leaving the perimeter, scoped to one identity, service, method set and destination project. Adding the partner project to the perimeter is too broad and brings an external project inside the trust boundary, and an IAM grant alone does not get past the perimeter's denial of calls that cross the boundary.

  7. Question 7

    A platform team's Cloud Build pipeline runs gcloud beta terraform vet against a policy library that only allows resources in approved regions. A quarterly audit finds a Compute Engine VM in a region that is not approved. An engineer created it by hand in the Google Cloud console. The security architect wants this to be impossible in future, and wants engineers who use the pipeline to keep getting fast feedback. What should the architect recommend?

    1. Keep the terraform vet check in the pipeline and add an Organization Policy constraint that restricts resource locations, so non-compliant resources are blocked however they are created
    2. Add more location rules to the terraform vet policy library so it covers every resource type the engineers might create
    3. Configure the production target in Cloud Deploy to require approval, so a reviewer must sign off before resources are created
    4. Schedule weekly Cloud Asset Inventory reports so the security team can find and delete VMs in regions that are not approved
    Show the answer

    Answer: A. Keep the terraform vet check in the pipeline and add an Organization Policy constraint that restricts resource locations, so non-compliant resources are blocked however they are created

    Pipeline policy checks give fast feedback but only see changes that go through the pipeline. Organization Policy constraints are enforced by the platform for every creation path, including the console, so you need both. Adding more terraform vet rules still does nothing about resources created in the console, and weekly reports only find violations after they exist.

  8. Question 8

    Six months after a large migration to Google Cloud, a retailer's finance director raises two complaints. Monthly cloud spend has gone over forecast twice without warning. Finance also cannot tell which business unit caused the spend. The CTO asks the architect for the most direct controls that fix both problems. Which two actions should the architect recommend? (Select TWO.)

    More than one answer is right.

    1. Move all production workloads to Spot VMs to bring down the total bill
    2. Buy 3-year resource-based committed use discounts for every VM currently running, to fix the monthly bill
    3. Apply consistent labels (for example, business unit and cost centre) to resources, and enable billing export to BigQuery for cost allocation reporting
    4. Create budgets with budget alerts for the billing account and key projects, so overspend is flagged before the month closes
    5. Rely on sustained use discounts to cap monthly spend automatically
    Show the answer

    Answers: C. Apply consistent labels (for example, business unit and cost centre) to resources, and enable billing export to BigQuery for cost allocation reporting; D. Create budgets with budget alerts for the billing account and key projects, so overspend is flagged before the month closes

    Budgets and alerts deal with unexpected overspend, and labels with billing export to BigQuery deal with cost allocation; these are the controls that OpEx spending needs. Buying CUDs for every current VM before rightsizing locks in any waste for years, and it does not show who is spending.

  9. Question 9

    A platform engineer is moving a nightly job from gsutil to gcloud storage. The job mirrors a local export directory to a Cloud Storage bucket and deletes objects in the bucket that are no longer present locally. The bucket holds the only copy of several years of reports. What is the best way to carry out the migration?

    1. Keep the job on gsutil permanently, because gcloud storage does not offer an rsync command
    2. Replace 'gsutil rsync' with 'gcloud storage rsync' using the same flags, because gcloud storage is the recommended successor and behaves identically
    3. Run 'gsutil version -l' and, if gsutil is a standalone install, switch the job straight to 'gcloud storage rsync' with no further testing
    4. Map the command with Google's migration guide, then test the new 'gcloud storage rsync' job carefully against non-critical data before it runs on the real bucket, because it decides which files to delete differently from gsutil rsync
    Show the answer

    Answer: D. Map the command with Google's migration guide, then test the new 'gcloud storage rsync' job carefully against non-critical data before it runs on the real bucket, because it decides which files to delete differently from gsutil rsync

    gcloud storage rsync decides which files to delete differently from gsutil rsync, so it must be mapped and tested before it touches irreplaceable data. Swapping the command name with the same flags assumes identical behaviour and could delete objects you meant to keep.

  10. Question 10

    A retail team runs a stateless API on Cloud Run. They plan a canary release that sends 10% of traffic to a new revision, and the release must cause no downtime. The new code renames a database column that the current revision still reads. As the architect, what should you require before the canary starts?

    1. Make the schema change backward-compatible, for example by adding the new column and keeping the old one, so both revisions can run against the same database during the rollout
    2. Use a recreate strategy so the old revision is stopped before the new one starts
    3. Switch to a blue/green release, because moving all traffic at once means the two revisions never need to share the database
    4. Raise the first canary step to 50% so the rollout finishes before the schema mismatch causes errors
    Show the answer

    Answer: A. Make the schema change backward-compatible, for example by adding the new column and keeping the old one, so both revisions can run against the same database during the rollout

    Canary and blue/green both depend on backward-compatible data changes, so a breaking migration has to be fixed in the database first. Blue/green is the closest distractor, but it also keeps the old version available for rollback, so the old version still needs the old schema to work.

The full timed mock

50 questions in 120 minutes, with a pass mark of 70%. Each paper is drawn from a bank of questions written for the mock alone, and these 10 are left out of it. Full access opens it, with every level of the course.

See the course →See the plans

LevelUpWith is independent of the organisation that sets this exam. These questions were written for this course; they are not taken from the exam.