Free Cloud Digital Leader practice test: 30 questions with answers

Pub/Sub is the managed messaging service built to ingest high-volume streaming events and fan them out to downstream consumers. Storage Transfer Service suits scheduled bulk copies rather than continuous events. Cloud SQL and Dataproc store or process data, but neither is the ingestion layer for real-time event streams.

BigQuery is a serverless analytics warehouse that uses standard SQL and scales without cluster management. Bigtable and Spanner are operational databases optimised for key-based access and transactions rather than large analytical scans. Dataproc could process the data, but the team would have to size and run clusters, which the requirement rules out.

Looker provides a shared semantic modelling layer and self-service exploration dashboards, so metric definitions are defined once and reused. Cloud Data Fusion moves and transforms data but is not a governed reporting front end. Dataplex governs lakes and Data Catalog helps users find metadata, and neither provides interactive governed dashboards on its own.

Dataplex gives a unified way to organise, discover and govern data across lakes, warehouses and storage, including policy management. Composer and Dataflow orchestrate or execute pipelines, and Pub/Sub transports messages, so none of them is a governance control plane. Picking one of those would solve a processing problem rather than the governance problem described.

Dataflow runs managed Apache Beam pipelines that read from Cloud Storage, transform data and write to BigQuery, scaling workers automatically. Cloud Run serves containers in response to requests rather than running large transformations. A custom script on Compute Engine can do the work, but it puts capacity planning and operations on the team, which the requirement avoids.

Cloud Build and Cloud Deploy automate the build and delivery path, which removes manual hand-offs and supports frequent releases. A manual approval board keeps the quarterly cadence, and a single large VM does not speed up release frequency. Monthly procurement is the opposite of the elastic model the company is trying to adopt.

A rehost onto Compute Engine moves the existing workload with little change, which lowers migration risk for stable applications. Re-architecting into microservices or rewriting as functions are valid later steps but add effort and risk up front. Moving only the reporting database leaves the customer-facing application where it is, so it does not meet the stated goal.

Cloud consumption shifts spending from capital expenditure on hardware to operating expenditure that follows demand. Committed use discounts lower unit prices for a term, but they still require forecasting rather than removing it. The per-employee fee and seven-year depreciation statements describe billing models that do not apply to consumption-based cloud services.

The resource hierarchy lets an enterprise inherit IAM policies, organisation policies and billing structures consistently from the top down. Free-text tags help reporting but do not enforce standards. A single project removes isolation between teams, and separate Workspace domains fragment identity without solving the governance problem.

Managed platforms let teams ship features sooner, and managed services take on routine operations such as database patching so staff can focus on differentiating work. Under the shared responsibility model customers still own their data and access controls. Zero downtime depends on architecture and is not guaranteed, and cloud adoption usually requires new skills rather than fewer.

Document AI provides processors that read forms and scanned documents and return structured fields, which fits the provider's need. Translation and Speech-to-Text handle different input types. Natural Language analyses text that is already digital rather than extracting fields from scanned images, so it would not produce the structured output required.

Vertex AI offers foundation models that can be grounded on enterprise documents through retrieval, which avoids the cost and time of training from scratch. Training a new model before a pilot is usually disproportionate for this goal. A static FAQ and keyword matching do not meet the conversational requirement.

BigQuery ML trains and runs models, including forecasting models, with SQL inside the warehouse, which suits SQL-focused analysts. Vision API analyses images, not sales tables. Dialogflow builds conversational agents and the Healthcare API serves a different domain entirely.

Responsible AI means assessing bias, explaining outcomes and monitoring models after release, which matters for lending decisions that affect customers. Training-set accuracy alone can hide poor generalisation and unfair outcomes. Unrestricted data access and cost-driven region selection both ignore the governance requirement the bank is raising.

Vertex AI can train image classification models on labelled defect images and serve them at an endpoint for low-latency predictions. Speech, Translation and Natural Language operate on audio or text, so they never analyse the images themselves. Applying them to reports after each shift also misses the near-real-time requirement.

Cloud Run runs containers without node management, scales with request volume and can scale to zero. Compute Engine and GKE Standard both leave the team managing machines or node pools. Bare-metal hosting is the most operationally demanding of the four options.

Migrate to Virtual Machines moves existing VMs to Compute Engine with little change, which matches a near-term need to keep the workload running. Refactoring or rebuilding as functions or Cloud Run is modernisation work the company is doing in parallel, not the first step. A BigQuery dataset cannot run a virtual-machine workload.

Anthos provides a consistent application platform, including Kubernetes management, across environments. Cloud Load Balancing and Cloud Armor handle traffic distribution and web protection, and Cloud Storage FUSE provides file access to buckets. None of them manages clusters across environments.

Artifact Registry is the managed repository for images and packages, with IAM access control and vulnerability scanning. Source repositories hold code rather than built artifacts. Cloud Scheduler and Cloud Tasks handle job triggering and task queues, so they do not store or distribute packages.

Extracting one component at a time behind an API keeps each change small and reversible, and a load balancer in front of the monolith lets traffic move to new services gradually. A full rewrite before any deployment delays value and raises risk. Removing tests and changing the database engine in the same window make failures very hard to diagnose.

Identity-Aware Proxy places access control in front of the application and can apply context-aware policies that check identity and device state, so no VPN is needed. A password policy does not verify the device, and an office subnet rule fails once employees work remotely. A secret URL path provides obscurity, not access control.

VPC Service Controls creates a service perimeter around Google-managed services, so data cannot be copied out to resources beyond the perimeter, even by privileged identities. IAM Owner grants broad rights rather than restricting them. Cloud Armor and VM egress firewall rules protect different network paths and do not govern API-level copies to outside projects.

Security Command Center aggregates findings on misconfigurations, vulnerabilities and threats across the organisation. Logging holds the logs that feed investigations but does not rank security posture. Billing reports and Monitoring address cost and performance respectively.

CMEK lets the organisation create and control keys in Cloud KMS for supported services such as Cloud Storage and BigQuery, and separating key administration from data administration strengthens that control. Default Google-managed keys are applied automatically but are not controlled by the customer. Laptop-held keys are unmanaged, and disabling encryption removes protection entirely.

Granting the BigQuery Data Viewer role at the dataset level limits the analyst to that dataset without giving project-wide rights. Project Owner is far broader even with a condition. A shared service account key is a credential that should not be emailed, and disabling audit logging removes accountability.

Cloud Monitoring collects metrics, builds dashboards and creates alerting policies across Google Cloud resources. Cloud Build, Cloud Scheduler and Artifact Registry each support delivery or scheduling rather than operational monitoring. Choosing one of them would miss the visibility and alerting need.

Cloud Logging lets engineers filter and search log entries across services to find the failed requests and their causes. Storage lifecycle rules manage retention and do not help with request failures. Committed use reports and Workspace administration address cost and user management respectively.

A regional managed instance group spread across zones, behind a load balancer that checks health, moves traffic away from a failed zone automatically. Routing to a second region with health checks extends that protection to a wider outage. Larger machines in one zone do not protect against a zone failure, and restore-after-outage or infrequent drills leave recovery times too long for the stated goal.

A Cloud Monitoring SLO tracks the service's reliability against a target, and an alerting policy on error-budget burn rate warns the team when it is using budget too fast. Storage notifications and monthly queries are not tied to reliability targets. Billing alerts track spend, not the reliability goal.

Eventarc and Pub/Sub deliver events from Compute Engine to a handler that can respond immediately, such as restarting the instance. A weekly schedule restarts machines on a timetable rather than responding to a failure. Storage transfer and commit-triggered builds react to unrelated events.