Google Professional Cloud Architect Practice Test 96
Question 950
Consider University financial aid and its support access audit workload. The issue is that a troubleshooting engineer needs to inspect errors but receives project Owner privileges; the current observation is that permission request includes unnecessary user data access. The team must ensure that incident support must follow least privilege and logged access. Which continuous monitoring approach best reveals recurrence or changing dependencies?
- Watch stale support grants and unauthorized data reads
- Create scoped temporary diagnostic roles and privileged session recording
- Give all temporary support contractors permanent Owner access
- Attempt an unrelated data export with diagnostic role and verify denial
Explanation
Correct Answer
A:
Watch stale support grants and unauthorized data reads reveals recurrence, drift or changing dependencies after launch. It provides a repeatable signal for the risk (a troubleshooting engineer needs to inspect errors but receives project Owner privileges) rather than treating the first successful test as permanent assurance.
Incorrect Answers
B:
The alternative “Create scoped temporary diagnostic roles and privileged session recording” selects a design, but does not provide the ongoing signal required to detect drift. In the reported observation, permission request includes unnecessary user data access. The option “Watch stale support grants and unauthorized data reads” instead addresses the requested ongoing operational assurance.
C:
Permanent broad privileges are not justified by a narrow debugging need. For this question, it is not a defensible choice for ongoing operational assurance and it does not satisfy the requirement that incident support must follow least privilege and logged access.
D:
The alternative “Attempt an unrelated data export with diagnostic role and verify denial” is a one-time acceptance result rather than ongoing operational assurance. In the reported observation, permission request includes unnecessary user data access. The option “Watch stale support grants and unauthorized data reads” instead addresses the requested ongoing operational assurance.
Question 951
Case study: Public school admissions portal: release version safety
Public school admissions portal relies on release version safety to deliver its operational commitments. The current architectural problem is that frontend rollout references a backend schema that is not yet deployed. The team has recorded this evidence: incident shows new clients send unsupported enrollment fields. Any proposed response must preserve the requirement that admission forms cannot break during deadline week.
At Public school admissions portal, the release version safety workload has a documented risk: frontend rollout references a backend schema that is not yet deployed. The review found that incident shows new clients send unsupported enrollment fields. The architecture must also ensure that admission forms cannot break during deadline week. The team must deliver the approved control safely. Which deployment sequence is most appropriate?
- Design backward-compatible API contracts and staged deployments
- Roll out schema first, then canary clients with traffic rollback
- Track 4xx schema errors and version skew
- Deploy breaking frontend and database schema simultaneously without canary
Explanation
Correct Answer
B:
Roll out schema first, then canary clients with traffic rollback is a controlled means to introduce the approved architecture while meeting the requirement that admission forms cannot break during deadline week. Staged delivery, attributable change evidence and rollback guard against an otherwise avoidable service disruption.
Incorrect Answers
A:
The alternative “Design backward-compatible API contracts and staged deployments” describes the chosen design, but not how it is introduced safely. In the reported observation, incident shows new clients send unsupported enrollment fields. The option “Roll out schema first, then canary clients with traffic rollback” instead addresses the requested controlled implementation.
C:
The alternative “Track 4xx schema errors and version skew” addresses later monitoring rather than the actual controlled deployment. In the reported observation, incident shows new clients send unsupported enrollment fields. The option “Roll out schema first, then canary clients with traffic rollback” instead addresses the requested controlled implementation.
D:
A coupled big-bang change creates untested integration failures. For this question, it is not a defensible choice for controlled implementation and it does not satisfy the requirement that admission forms cannot break during deadline week.
Question 952
Global fleet analytics company plans to update connected vehicle insight platform. Investigators found that forecast shows devices and event attributes will expand across regions, consistent with the reported problem that management expects ten times more telemetry after a merger but the platform allocates fixed partitions. The response must satisfy the requirement that existing fleet alerts cannot lose events during schema evolution. Which target-state design is best justified by the evidence and required business outcome?
- Design partitioned event contracts and independently scalable analytical consumers
- Project growth by region and event class with migration milestones
- Watch partition hotspots, obsolete schemas and capacity forecasts
- Change all event schemas simultaneously without versioning
Explanation
Correct Answer
A:
Design partitioned event contracts and independently scalable analytical consumers targets the underlying issue (management expects ten times more telemetry after a merger but the platform allocates fixed partitions) rather than merely reporting symptoms. It also preserves the required outcome (existing fleet alerts cannot lose events during schema evolution), which a target-state design must satisfy.
Incorrect Answers
B:
The alternative “Project growth by region and event class with migration milestones” investigates the cause without specifying the durable architecture to resolve it. In the reported observation, forecast shows devices and event attributes will expand across regions. The option “Design partitioned event contracts and independently scalable analytical consumers” instead addresses the requested architecture selection.
C:
The alternative “Watch partition hotspots, obsolete schemas and capacity forecasts” measures results without defining the preventive design now required. In the reported observation, forecast shows devices and event attributes will expand across regions. The option “Design partitioned event contracts and independently scalable analytical consumers” instead addresses the requested architecture selection.
D:
A schema cutover without compatibility can strand existing devices. For this question, it is not a defensible choice for architecture selection and it does not satisfy the requirement that existing fleet alerts cannot lose events during schema evolution.
Question 953
The student-data migration environment at Public university registrar faces this issue: bulk import from an aging database overwrites later updates during the cutover. Supporting evidence shows that comparison shows primary-key collisions between initial exports and ongoing edits. The architect must preserve the requirement that graduation records must be correct at the moment of cutover. Select TWO complementary actions: one durable target architecture and one controlled way to introduce it while preserving service. Which TWO are best?
- Import the original snapshot only and ignore writes after its extraction
- Plan full-load plus change-data capture with deterministic conflict handling
- Execute synchronized cutover after low-lag catch-up and final write-freeze
- Map CDC support, key semantics, update order and rollback window
- Track CDC lag, failed writes and migration data quality
Explanation
Correct Answers
B:
Plan full-load plus change-data capture with deterministic conflict handling targets the underlying issue (bulk import from an aging database overwrites later updates during the cutover) rather than merely reporting symptoms. It also preserves the required outcome (graduation records must be correct at the moment of cutover), which a target-state design must satisfy. This supplies the lasting-control half of the two-action package.
C:
Execute synchronized cutover after low-lag catch-up and final write-freeze is a controlled means to introduce the approved architecture while meeting the requirement that graduation records must be correct at the moment of cutover. Staged delivery, attributable change evidence and rollback guard against an otherwise avoidable service disruption. This supplies the controlled-introduction half of the two-action package.
Incorrect Answers
A:
A stale snapshot loses legitimate updates made during the move. The shortcut does not supply the requested design and safe rollout for student-data migration, where comparison shows primary-key collisions between initial exports and ongoing edits.
D:
This diagnoses the original issue (bulk import from an aging database overwrites later updates during the cutover), but the TWO requested actions are the target architecture and its implementation; discovery was already completed.
E:
Monitoring student-data migration for future drift is useful after the rollout, but it does not specify the requested target control and safe implementation for this case: bulk import from an aging database overwrites later updates during the cutover.
Question 954
Case study: Online language tutoring platform: GKE API worker capacity
Online language tutoring platform relies on GKE API worker capacity to deliver its operational commitments. The current architectural problem is that CPU-based scaling alone misses a backlog of asynchronous translation requests. The team has recorded this evidence: pod telemetry shows low CPU but growing Pub/Sub outstanding messages. Any proposed response must preserve the requirement that tutoring sessions require backlogs to clear in under two minutes.
At Online language tutoring platform, the Google Cloud design for GKE API worker capacity is affected because CPU-based scaling alone misses a backlog of asynchronous translation requests. A review found that pod telemetry shows low CPU but growing Pub/Sub outstanding messages. The proposed response must account for the requirement that tutoring sessions require backlogs to clear in under two minutes. Which controlled implementation should be used to preserve operational continuity during the change?
- Stage HPA/KEDA equivalent scaling configuration with quota safeguards
- Use workload autoscaling based on meaningful request or queue metrics
- Increase CPU limits but keep one replica regardless of queue depth
- Track autoscaler saturation, pending pods and oldest task age
Explanation
Correct Answer
A:
Stage HPA/KEDA equivalent scaling configuration with quota safeguards is a controlled means to introduce the approved architecture while meeting the requirement that tutoring sessions require backlogs to clear in under two minutes. Staged delivery, attributable change evidence and rollback guard against an otherwise avoidable service disruption.
Incorrect Answers
B:
The alternative “Use workload autoscaling based on meaningful request or queue metrics” describes the chosen design, but not how it is introduced safely. In the reported observation, pod telemetry shows low CPU but growing Pub/Sub outstanding messages. The option “Stage HPA/KEDA equivalent scaling configuration with quota safeguards” instead addresses the requested controlled implementation.
C:
A fixed replica count cannot scale with independent queue pressure. For this question, it is not a defensible choice for controlled implementation and it does not satisfy the requirement that tutoring sessions require backlogs to clear in under two minutes.
D:
The alternative “Track autoscaler saturation, pending pods and oldest task age” addresses later monitoring rather than the actual controlled deployment. In the reported observation, pod telemetry shows low CPU but growing Pub/Sub outstanding messages. The option “Stage HPA/KEDA equivalent scaling configuration with quota safeguards” instead addresses the requested controlled implementation.
Question 955
At Regional credit union, the customer deletion obligations workload has a documented risk: data retention jobs preserve marketing identifiers after approved deletion requests. The review found that data catalog shows identifiers copied into backups and staging tables. The architecture must also ensure that statutory retention exceptions and deletion rights must be reconciled. The team must deliver the approved control safely. Which deployment sequence is most appropriate?
- Delete only the primary profile while keeping active analytics identifiers forever
- Track deletion lag, restoration resurrection risk and retained exceptions
- Implement governed erasure or deidentification across eligible datasets with exception tracking
- Roll out deletion propagation by data system after legal review
Explanation
Correct Answer
D:
Roll out deletion propagation by data system after legal review is a controlled means to introduce the approved architecture while meeting the requirement that statutory retention exceptions and deletion rights must be reconciled. Staged delivery, attributable change evidence and rollback guard against an otherwise avoidable service disruption.
Incorrect Answers
A:
A partial deletion fails to address replicated eligible personal data. For this question, it is not a defensible choice for controlled implementation and it does not satisfy the requirement that statutory retention exceptions and deletion rights must be reconciled.
B:
The alternative “Track deletion lag, restoration resurrection risk and retained exceptions” addresses later monitoring rather than the actual controlled deployment. In the reported observation, data catalog shows identifiers copied into backups and staging tables. The option “Roll out deletion propagation by data system after legal review” instead addresses the requested controlled implementation.
C:
The alternative “Implement governed erasure or deidentification across eligible datasets with exception tracking” describes the chosen design, but not how it is introduced safely. In the reported observation, data catalog shows identifiers copied into backups and staging tables. The option “Roll out deletion propagation by data system after legal review” instead addresses the requested controlled implementation.
Question 956
Case study: Medical document classification team: managed model serving
Medical document classification team relies on managed model serving to deliver its operational commitments. The current architectural problem is that training is fast on a multi-GPU cluster but the serving endpoint runs on expensive always-on hardware. The team has recorded this evidence: billing and latency data show low daily inference traffic. Any proposed response must preserve the requirement that classified records require controlled access and predictable response SLO.
Consider Medical document classification team and its managed model serving workload. The issue is that training is fast on a multi-GPU cluster but the serving endpoint runs on expensive always-on hardware; the current observation is that billing and latency data show low daily inference traffic. The team must ensure that classified records require controlled access and predictable response SLO. Select TWO complementary actions: one durable target architecture and one controlled way to introduce it while preserving service. Which TWO are best?
- Keep training GPU fleet online as the only inference endpoint
- Watch model-serving utilization, latency and prediction drift
- Choose independently sized training jobs and autoscaled inference services
- Deploy staging endpoint with monitored inference load and private authorization
- Separate training accelerator use from inference demand and privacy constraints
Explanation
Correct Answers
C:
Choose independently sized training jobs and autoscaled inference services targets the underlying issue (training is fast on a multi-GPU cluster but the serving endpoint runs on expensive always-on hardware) rather than merely reporting symptoms. It also preserves the required outcome (classified records require controlled access and predictable response SLO), which a target-state design must satisfy. This supplies the lasting-control half of the two-action package.
D:
Deploy staging endpoint with monitored inference load and private authorization is a controlled means to introduce the approved architecture while meeting the requirement that classified records require controlled access and predictable response SLO. Staged delivery, attributable change evidence and rollback guard against an otherwise avoidable service disruption. This supplies the controlled-introduction half of the two-action package.
Incorrect Answers
A:
Training capacity is not automatically cost-effective for serving. The shortcut does not supply the requested design and safe rollout for managed model serving, where billing and latency data show low daily inference traffic.
B:
Monitoring managed model serving for future drift is useful after the rollout, but it does not specify the requested target control and safe implementation for this case: training is fast on a multi-GPU cluster but the serving endpoint runs on expensive always-on hardware.
E:
This diagnoses the original issue (training is fast on a multi-GPU cluster but the serving endpoint runs on expensive always-on hardware), but the TWO requested actions are the target architecture and its implementation; discovery was already completed.
Question 957
The release regression validation environment at Healthcare payments startup faces this issue: team tests only HTTP status after replacing the production authentication component. Supporting evidence shows that incident reveals token expiry and cross-tenant access tests were omitted. The architect must preserve the requirement that payments and patient data require safe release acceptance. Choose the architecture that resolves the documented weakness without breaching the operational constraint.
- Treat an HTTP 200 response as proof all identity controls work
- Monitor token failure trends and late regression reports
- Build automated integration tests for valid and denied authorization paths
- Map critical auth flows, negative tests and dependency boundaries
Explanation
Correct Answer
C:
Build automated integration tests for valid and denied authorization paths targets the underlying issue (team tests only HTTP status after replacing the production authentication component) rather than merely reporting symptoms. It also preserves the required outcome (payments and patient data require safe release acceptance), which a target-state design must satisfy.
Incorrect Answers
A:
A success code says nothing about unauthorized access paths. For this question, it is not a defensible choice for architecture selection and it does not satisfy the requirement that payments and patient data require safe release acceptance.
B:
The alternative “Monitor token failure trends and late regression reports” measures results without defining the preventive design now required. In the reported observation, incident reveals token expiry and cross-tenant access tests were omitted. The option “Build automated integration tests for valid and denied authorization paths” instead addresses the requested architecture selection.
D:
The alternative “Map critical auth flows, negative tests and dependency boundaries” investigates the cause without specifying the durable architecture to resolve it. In the reported observation, incident reveals token expiry and cross-tenant access tests were omitted. The option “Build automated integration tests for valid and denied authorization paths” instead addresses the requested architecture selection.
Question 958
At Climate modeling consortium, the Google Cloud design for GPU inference profiling is affected because AI inference costs spike although request counts stay flat. A review found that profile shows a new preprocessing step copies tensors repeatedly between CPU and GPU. The proposed response must account for the requirement that cost optimization must not lower model output quality. A deployment ticket has been closed. What direct evidence would justify solution signoff?
- Inspect profiler traces, memory transfer sizes and latency distributions
- Buy more GPUs without measuring the memory-transfer bottleneck
- Compare output parity, GPU utilization and p99 inference latency
- Roll out optimized preprocessing behind a versioned serving endpoint
Explanation
Correct Answer
C:
Compare output parity, GPU utilization and p99 inference latency is a direct acceptance test of the path implicated by profile shows a new preprocessing step copies tensors repeatedly between CPU and GPU. It checks that the original risk (AI inference costs spike although request counts stay flat) has actually been controlled and tests that permitted operations still meet the requirement that cost optimization must not lower model output quality.
Incorrect Answers
A:
The alternative “Inspect profiler traces, memory transfer sizes and latency distributions” collects prechange evidence without confirming that the revised service meets its outcome. In the reported observation, profile shows a new preprocessing step copies tensors repeatedly between CPU and GPU. The option “Compare output parity, GPU utilization and p99 inference latency” instead addresses the requested acceptance verification.
B:
More compute may increase spend while copies still dominate. For this question, it is not a defensible choice for acceptance verification and it does not satisfy the requirement that cost optimization must not lower model output quality.
D:
The alternative “Roll out optimized preprocessing behind a versioned serving endpoint” documents change execution but not its operational acceptance result. In the reported observation, profile shows a new preprocessing step copies tensors repeatedly between CPU and GPU. The option “Compare output parity, GPU utilization and p99 inference latency” instead addresses the requested acceptance verification.
Question 959
At National agriculture ministry, the Google Cloud design for sovereign data retention is affected because analysts want one globally replicated bucket for personal farm ownership records. A review found that compliance map lists a specific country-only data-storage rule. The proposed response must account for the requirement that regulatory data must remain within the designated geography. Which analysis will allow the architect to distinguish the reported symptom from the actual failure path?
- Select location-restricted storage and explicit organization location policies
- Rehome eligible object groups with controlled object-copy checks and deny policies
- Verify exact residency constraints, access jurisdictions and lawful transfers
- Use a multi-region spanning prohibited locations for all identifiers
Explanation
Correct Answer
C:
Verify exact residency constraints, access jurisdictions and lawful transfers is the appropriate diagnostic because the observation—compliance map lists a specific country-only data-storage rule—does not by itself establish every dependency, authorization rule or failure boundary. The findings explain the reported risk (analysts want one globally replicated bucket for personal farm ownership records) before the design is approved.
Incorrect Answers
A:
The alternative “Select location-restricted storage and explicit organization location policies” is target architecture selection; it does not establish why the observed path failed. In the reported observation, compliance map lists a specific country-only data-storage rule. The option “Verify exact residency constraints, access jurisdictions and lawful transfers” instead addresses the requested root-cause evidence.
B:
The alternative “Rehome eligible object groups with controlled object-copy checks and deny policies” is rollout work that may modify the wrong dependency before root-cause evidence is collected. In the reported observation, compliance map lists a specific country-only data-storage rule. The option “Verify exact residency constraints, access jurisdictions and lawful transfers” instead addresses the requested root-cause evidence.
D:
A multi-region location cannot be assumed compliant with a country-only restriction. For this question, it is not a defensible choice for root-cause evidence and it does not satisfy the requirement that regulatory data must remain within the designated geography.
Question 960
At Medical lab network, the shared filesystem for containers workload has a documented risk: lab containers need concurrent POSIX file writes but the team selected object-storage semantics. The review found that test suite fails on file-lock operations and atomic renames. The architecture must also ensure that laboratory collaboration requires managed NFS semantics in one region. What follow-up metric or alert will identify emerging failures after the initial acceptance test?
- Select Filestore for shared file workloads consumed by GKE pods
- Run concurrent file-lock and rename tests under a pod restart
- Monitor NFS saturation, share capacity and backup success
- Treat Cloud Storage object writes as POSIX byte-range locking
Explanation
Correct Answer
C:
Monitor NFS saturation, share capacity and backup success reveals recurrence, drift or changing dependencies after launch. It provides a repeatable signal for the risk (lab containers need concurrent POSIX file writes but the team selected object-storage semantics) rather than treating the first successful test as permanent assurance.
Incorrect Answers
A:
The alternative “Select Filestore for shared file workloads consumed by GKE pods” selects a design, but does not provide the ongoing signal required to detect drift. In the reported observation, test suite fails on file-lock operations and atomic renames. The option “Monitor NFS saturation, share capacity and backup success” instead addresses the requested ongoing operational assurance.
B:
The alternative “Run concurrent file-lock and rename tests under a pod restart” is a one-time acceptance result rather than ongoing operational assurance. In the reported observation, test suite fails on file-lock operations and atomic renames. The option “Monitor NFS saturation, share capacity and backup success” instead addresses the requested ongoing operational assurance.
D:
Object storage does not provide identical filesystem locking semantics. For this question, it is not a defensible choice for ongoing operational assurance and it does not satisfy the requirement that laboratory collaboration requires managed NFS semantics in one region.