# Free PCA practice questions

> 10 free PCA practice questions with the correct answer and an explanation.

Page: https://devopscodelab.com/exams/google-professional-cloud-architect/free-practice-questions/
Practice exams: https://devopscodelab.com/exams/google-professional-cloud-architect/

### Question 1 of 10

A logistics company is designing a shipment-tracking platform. Requirements: globally consistent inventory counts across warehouses in North America, Europe, and Asia; relational schema with SQL; 99.999% availability; and horizontal write scaling as volume grows. Which database should the architect choose?

- A. Cloud SQL for PostgreSQL with cross-region read replicas promoted on failure
- B. AlloyDB for PostgreSQL with a primary cluster and secondary clusters in two other regions
- C. Bigtable with multi-cluster routing across three regional clusters
- D. Spanner with a multi-region instance configuration

**Answer:** D

**Explanation:** Spanner multi-region configurations provide externally consistent reads and writes, SQL, horizontal scaling, and a 99.999% availability SLA. Cloud SQL replicas are read-only and asynchronous, so writes do not scale and failover can lose recent data. AlloyDB secondary clusters are asynchronous read targets for disaster recovery rather than globally consistent multi-writer storage. Bigtable with multi-cluster routing is eventually consistent and is not a relational SQL database.

### Question 2 of 10

A ride-sharing company must store driver location updates (millions of writes per second) and serve time-range lookups per driver with single-digit millisecond latency. Analysts occasionally run SQL over the same data. Which primary datastore is MOST suitable?

- A. Cloud SQL for MySQL with read replicas and an index on driver ID and timestamp
- B. Firestore in Native mode with one document per location update and a composite index
- C. Bigtable with a row key that combines driver ID and a reversed timestamp
- D. BigQuery with streaming inserts and tables partitioned by ingestion time

**Answer:** C

**Explanation:** Bigtable handles very high write throughput with low-latency key-range scans, and a driver-ID plus reversed-timestamp key gives efficient recent-history lookups; analysts can query it through BigQuery external tables. BigQuery is an analytics warehouse, not a millisecond serving store. Firestore has per-database write throughput characteristics and per-document costs unsuited to millions of writes per second. Cloud SQL cannot sustain this write rate on a single primary.

### Question 3 of 10

A financial-news publisher must store 2 PB of raw article archives that are read less than once per year for legal discovery. Retrieval within hours is acceptable, and the CFO wants the lowest storage cost. Which storage choice is MOST appropriate?

- A. Filestore Basic HDD tier exported over NFS to a small VM that serves legal discovery requests
- B. Persistent Disk standard volumes snapshotted monthly to a multi-regional snapshot location
- C. Cloud Storage Archive class in a regional bucket, or dual-region only if geo-redundancy is required
- D. Cloud Storage Nearline class with Object Lifecycle Management deleting objects after one year

**Answer:** C

**Explanation:** Archive class has the lowest storage price for data accessed less than once a year, and choosing regional versus dual-region depends only on the redundancy requirement. Filestore is priced for active file workloads and is far more expensive per GB. Nearline targets roughly monthly access and costs more to store, and deleting after one year breaks the legal retention need. Persistent Disk and snapshots are block storage priced well above archival object storage.

### Question 4 of 10

A company runs microservices on GKE Standard. Some services need GPUs for inference, others are general-purpose, and a few batch jobs can use interruptible capacity. The platform team wants workloads placed correctly without manual node management for each team. What should they configure?

- A. A single general-purpose node pool and DaemonSets that install GPU drivers on demand whenever an inference pod is scheduled
- B. Three separate GKE clusters, one per workload type, connected with Multi Cluster Ingress to share traffic
- C. One large node pool of GPU machines so every workload can run anywhere, with the HorizontalPodAutoscaler controlling replica counts
- D. Separate node pools for GPU, general, and Spot capacity with taints and tolerations, and enable cluster autoscaler with node auto-provisioning

**Answer:** D

**Explanation:** Dedicated node pools with taints and tolerations keep GPU and Spot capacity for the workloads that need it, and autoscaling with node auto-provisioning creates nodes without manual work. Running everything on GPU machines is extremely expensive. Three clusters multiply operations and Multi Cluster Ingress is for routing traffic, not placing workloads by hardware. Drivers cannot add GPU hardware to machines that do not have it.

### Question 5 of 10

A manufacturer connects its data center to Google Cloud with two 10-Gbps Dedicated Interconnect connections in the same metropolitan area, one per edge availability domain. Cloud Router uses BGP. Leadership asks for the 99.99% availability SLA. What must change?

- A. Replace the two connections with a single 100-Gbps Dedicated Interconnect connection to reduce failure points
- B. Add HA VPN tunnels over the internet as a backup path, advertising routes with a lower BGP priority than the interconnect
- C. Upgrade both connections to 100 Gbps and enable global dynamic routing on the VPC network
- D. Add two more connections in a second metropolitan area and attach them to Cloud Routers in a different region

**Answer:** D

**Explanation:** The 99.99% topology for Dedicated Interconnect requires four connections, two in each of two metropolitan areas, with Cloud Routers in two regions. A single large connection lowers availability because it is one point of failure. HA VPN as a backup improves resilience but does not qualify the interconnect for the 99.99% SLA. More bandwidth and global routing do not change the topology the SLA is based on.

### Question 6 of 10 (choose 2)

A company wants to ensure that workloads can only use Cloud KMS keys from a central, approved key project, and that only CMEK-protected resources can be created for specific services. Which controls should the architect use? (Choose two.)

- A. An IAM deny policy blocking cloudkms.cryptoKeys.create in all projects except the key project
- B. Cloud KMS Autokey enabled on every folder so keys are created alongside each resource
- C. The organization policy constraint requiring CMEK for listed services
- D. The organization policy constraint restricting which projects may supply CMEK keys
- E. Cloud Asset Inventory feeds that alert when a resource is created without a CMEK key

**Answer:** C, D

**Explanation:** The constraint restricting which projects may supply CMEK keys limits key usage to the approved key project, and the constraint requiring CMEK for listed services blocks creation of non-CMEK resources. A deny policy on key creation does not stop resources from using keys elsewhere or from using Google-managed encryption. Asset Inventory alerts are detective, not preventive. Autokey automates key creation but does not by itself block non-CMEK resources or pin keys to one central project.

### Question 7 of 10

An analytics company stores customer data in BigQuery and Cloud Storage. Auditors require that data cannot be copied out to projects outside the company's organization, even by a user with valid IAM permissions using stolen credentials from the internet. What should the architect implement?

- A. Customer-managed encryption keys in Cloud KMS for every dataset and bucket, with automatic 90-day rotation and all key access logged
- B. A VPC Service Controls perimeter around the data projects with access levels limiting requests to corporate networks and devices
- C. IAM deny policies that block the bigquery.tables.export permission for all principals in the organization
- D. Organization policy constraints that enforce uniform bucket-level access and disable public access prevention overrides

**Answer:** B

**Explanation:** VPC Service Controls block data movement across the perimeter, including copies to outside projects, and access levels stop requests from unexpected networks even with valid credentials. CMEK controls key ownership but a user with permission can still read and copy data. Denying export blocks one path but not queries that write results elsewhere or object copies. Uniform access and public access prevention stop public exposure but not copies by authorized identities.

### Question 8 of 10

A company's architecture review finds that a new platform will require developers to learn Kubernetes, but most of the team only knows VM administration. The launch date is fixed in four months. What is the BEST approach to team readiness?

- A. Postpone the launch until every developer has passed the Certified Kubernetes Administrator exam
- B. Hire an external contractor to run the platform permanently so the internal team does not need new skills
- C. Assess skills against the target design, choose a managed option like GKE Autopilot or Cloud Run where possible, and schedule targeted training
- D. Keep the VM-based design to avoid training, even though it fails the scalability requirement

**Answer:** C

**Explanation:** Matching managed services to current skills while closing gaps with targeted training balances risk and timeline. Delaying the launch for certifications ignores the fixed date. Permanent outsourcing removes internal ownership and creates long-term dependency. Ignoring a stated requirement to avoid training produces a solution that fails the business goals.

### Question 9 of 10

A company exposes internal order services to partner developers. They need API keys and OAuth for partners, rate limiting per partner, analytics on usage, and a developer portal. Which solution should the architect recommend?

- A. API Gateway with a single API config and Cloud Monitoring dashboards for usage metrics
- B. An external Application Load Balancer with Cloud Armor rate-based rules per source IP
- C. Cloud Endpoints with an OpenAPI specification and an Extensible Service Proxy in front of each service
- D. Apigee API management with API products, quota policies, and an integrated developer portal

**Answer:** D

**Explanation:** Apigee provides API products, per-developer quotas, monetization-ready analytics, and a developer portal, matching every requirement. Cloud Endpoints and API Gateway offer authentication and basic management but not a full partner portal and rich per-partner product controls. Cloud Armor limits traffic by IP and other attributes, not by partner API product, and offers no portal or API analytics.

### Question 10 of 10

A company wants to ensure its production systems remain reliable while launching new features rapidly. Which SRE concept helps balance these goals?

- A. A fixed quarterly release calendar agreed between product and operations
- B. Change advisory board approval for every production release, regardless of service health
- C. Continuous deployment with automatic rollback, released at whatever pace developers choose
- D. Error budgets derived from SLOs that guide release pace

**Answer:** D

**Explanation:** Error budgets derived from SLOs quantify acceptable unreliability; while budget remains, teams release quickly, and when it is spent they shift effort to reliability. Developer-chosen pace ignores reliability signals even with rollback. Approval boards and fixed calendars slow delivery without tying release pace to measured reliability.

Ready for a timed run? Take the free 15-question PCA mini exam: https://devopscodelab.com/app/exams/google-professional-cloud-architect/mini/

DevOpsCode Lab is not affiliated with Google Cloud. PCA and related names are trademarks of their owners. We write every question ourselves; these are not real exam questions.
