Google Professional Cloud Architect Practice Test 85
Question 840
Consider National grocery cooperative and its Gemini Cloud Assist runbook recommendations workload. The issue is that an engineer proposes applying suggested infrastructure edits directly to production; the current observation is that review finds recommendations do not reflect a special pharmacy-workload SLA. The team must ensure that critical pharmacy distribution cannot be disrupted by unreviewed changes. The change committee needs to understand the current system rather than install a fix yet. What should it do?
- Convert approved recommendations to reviewed IaC changes with canary rollout
- Validate AI-assisted suggestions against live architecture, policy and business SLOs
- Define human-reviewed change proposals with policy and unit tests
- Execute every model-generated shell command as project Owner without review
Explanation
Correct Answer
B:
Validate AI-assisted suggestions against live architecture, policy and business SLOs is the appropriate diagnostic because the observation—review finds recommendations do not reflect a special pharmacy-workload SLA—does not by itself establish every dependency, authorization rule or failure boundary. The findings explain the reported risk (an engineer proposes applying suggested infrastructure edits directly to production) before the design is approved.
Incorrect Answers
A:
The alternative “Convert approved recommendations to reviewed IaC changes with canary rollout” is rollout work that may modify the wrong dependency before root-cause evidence is collected. In the reported observation, review finds recommendations do not reflect a special pharmacy-workload SLA. The option “Validate AI-assisted suggestions against live architecture, policy and business SLOs” instead addresses the requested root-cause evidence.
C:
The alternative “Define human-reviewed change proposals with policy and unit tests” is target architecture selection; it does not establish why the observed path failed. In the reported observation, review finds recommendations do not reflect a special pharmacy-workload SLA. The option “Validate AI-assisted suggestions against live architecture, policy and business SLOs” instead addresses the requested root-cause evidence.
D:
An AI suggestion is not an authorized or validated operational change. For this question, it is not a defensible choice for root-cause evidence and it does not satisfy the requirement that critical pharmacy distribution cannot be disrupted by unreviewed changes.
Question 841
Consider Insurance photo assessor and its damaged vehicle classification workload. The issue is that claims teams seek automated detection of damaged panels from uploaded images; the current observation is that validation examples show visual categories differ from text-only customer descriptions. The team must ensure that model outputs are advisory and claims decisions need human review. Select THREE evidence activities to cover (1) diagnosis before design, (2) direct post-change acceptance, and (3) continued operational monitoring. Which THREE fit those stages?
- Accept the deployment ticket as proof that the control has no remaining security or availability risk
- Monitor low-confidence rates, model changes and reviewer overrides
- Determine labels, false-positive risk and annotated evaluation set
- Use a suitable vision-capable model or API with calibrated confidence and human escalation
- Apply only sentiment analysis of the customer's filename
- Measure precision and recall per damage class on withheld claim images
Explanation
Correct Answers
B:
Monitor low-confidence rates, model changes and reviewer overrides reveals recurrence, drift or changing dependencies after launch. It provides a repeatable signal for the risk (claims teams seek automated detection of damaged panels from uploaded images) rather than treating the first successful test as permanent assurance. It supplies the ongoing observation requested.
C:
Determine labels, false-positive risk and annotated evaluation set is the appropriate diagnostic because the observation—validation examples show visual categories differ from text-only customer descriptions—does not by itself establish every dependency, authorization rule or failure boundary. The findings explain the reported risk (claims teams seek automated detection of damaged panels from uploaded images) before the design is approved. It supplies the pre-design diagnostic evidence requested.
F:
Measure precision and recall per damage class on withheld claim images is a direct acceptance test of the path implicated by validation examples show visual categories differ from text-only customer descriptions. It checks that the original risk (claims teams seek automated detection of damaged panels from uploaded images) has actually been controlled and tests that permitted operations still meet the requirement that model outputs are advisory and claims decisions need human review. It supplies the separate post-change acceptance result requested.
Incorrect Answers
A:
A change-record state does not measure the behavior reported by the original evidence, and cannot prove protection or availability. In particular, the observed issue was: validation examples show visual categories differ from text-only customer descriptions.
D:
Designing damaged vehicle classification addresses the root cause, but this specific three-part question asks for evidence to diagnose, accept and continuously verify the response to claims teams seek automated detection of damaged panels from uploaded images, not another design proposal.
E:
Filename sentiment cannot identify visual damage. It cannot satisfy the required evidence stages.
Question 842
The Cloud Bigtable emulator development environment at Public education agency faces this issue: test suite uses expensive production Bigtable instance for every integration run. Supporting evidence shows that billing shows CI generates thousands of synthetic rows in live student datasets. The architect must preserve the requirement that tests must be repeatable without touching real student data. What test must pass before the business accepts the remediated architecture?
- Review Bigtable client configuration and test fixture requirements
- Grant CI full production Bigtable admin to simplify setup
- Verify fixture cleanup and that no test call reaches production project
- Configure CI to use emulator endpoint with synthetic tables
Explanation
Correct Answer
C:
Verify fixture cleanup and that no test call reaches production project is a direct acceptance test of the path implicated by billing shows CI generates thousands of synthetic rows in live student datasets. It checks that the original risk (test suite uses expensive production Bigtable instance for every integration run) has actually been controlled and tests that permitted operations still meet the requirement that tests must be repeatable without touching real student data.
Incorrect Answers
A:
The alternative “Review Bigtable client configuration and test fixture requirements” collects prechange evidence without confirming that the revised service meets its outcome. In the reported observation, billing shows CI generates thousands of synthetic rows in live student datasets. The option “Verify fixture cleanup and that no test call reaches production project” instead addresses the requested acceptance verification.
B:
Production permissions undermine isolation and create destructive test risk. For this question, it is not a defensible choice for acceptance verification and it does not satisfy the requirement that tests must be repeatable without touching real student data.
D:
The alternative “Configure CI to use emulator endpoint with synthetic tables” documents change execution but not its operational acceptance result. In the reported observation, billing shows CI generates thousands of synthetic rows in live student datasets. The option “Verify fixture cleanup and that no test call reaches production project” instead addresses the requested acceptance verification.
Question 843
The audit log preservation environment at National retail cooperative faces this issue: a project owner can disable or delete local logging after an unauthorized configuration change. Supporting evidence shows that audit exercise shows log sinks owned by the affected project. The architect must preserve the requirement that investigations require independently retained administrative events. Before approving a solution, which analysis most directly clarifies the underlying dependency or control gap?
- Store the only audit copy in a bucket controlled by the investigated project owner
- Stage organization-level sinks and validate destination authorization
- Aggregate audit logs into a protected central project with minimal deletion rights
- Map log routing permissions, sink exclusions and retention requirements
Explanation
Correct Answer
D:
Map log routing permissions, sink exclusions and retention requirements is the appropriate diagnostic because the observation—audit exercise shows log sinks owned by the affected project—does not by itself establish every dependency, authorization rule or failure boundary. The findings explain the reported risk (a project owner can disable or delete local logging after an unauthorized configuration change) before the design is approved.
Incorrect Answers
A:
Shared deletion authority undermines evidentiary independence. For this question, it is not a defensible choice for root-cause evidence and it does not satisfy the requirement that investigations require independently retained administrative events.
B:
The alternative “Stage organization-level sinks and validate destination authorization” is rollout work that may modify the wrong dependency before root-cause evidence is collected. In the reported observation, audit exercise shows log sinks owned by the affected project. The option “Map log routing permissions, sink exclusions and retention requirements” instead addresses the requested root-cause evidence.
C:
The alternative “Aggregate audit logs into a protected central project with minimal deletion rights” is target architecture selection; it does not establish why the observed path failed. In the reported observation, audit exercise shows log sinks owned by the affected project. The option “Map log routing permissions, sink exclusions and retention requirements” instead addresses the requested root-cause evidence.
Question 844
At National meteorological lab, the Google Cloud design for distributed foundation-model training is affected because the team requests ad hoc GPU VMs spread across zones with slow checkpoint synchronization. A review found that training trace shows workers idle waiting for cross-zone gradient transfer. The proposed response must account for the requirement that accelerator capacity is limited and checkpoint recovery is mandatory. The cause has been established and a lasting architecture is being selected. Which design best meets the constraint?
- Spread training workers across unrelated low-bandwidth VMs without topology planning
- Profile interconnect bandwidth, worker coordination and restart behavior
- Track utilization, failed checkpoints and accelerator reservation burn
- Use topology-aware AI Hypercomputer training architecture with durable checkpoints
Explanation
Correct Answer
D:
Use topology-aware AI Hypercomputer training architecture with durable checkpoints targets the underlying issue (the team requests ad hoc GPU VMs spread across zones with slow checkpoint synchronization) rather than merely reporting symptoms. It also preserves the required outcome (accelerator capacity is limited and checkpoint recovery is mandatory), which a target-state design must satisfy.
Incorrect Answers
A:
Communication overhead can dominate distributed training. For this question, it is not a defensible choice for architecture selection and it does not satisfy the requirement that accelerator capacity is limited and checkpoint recovery is mandatory.
B:
The alternative “Profile interconnect bandwidth, worker coordination and restart behavior” investigates the cause without specifying the durable architecture to resolve it. In the reported observation, training trace shows workers idle waiting for cross-zone gradient transfer. The option “Use topology-aware AI Hypercomputer training architecture with durable checkpoints” instead addresses the requested architecture selection.
C:
The alternative “Track utilization, failed checkpoints and accelerator reservation burn” measures results without defining the preventive design now required. In the reported observation, training trace shows workers idle waiting for cross-zone gradient transfer. The option “Use topology-aware AI Hypercomputer training architecture with durable checkpoints” instead addresses the requested architecture selection.
Question 845
Case study: Medical appointment SaaS: serverless private database access
Medical appointment SaaS relies on serverless private database access to deliver its operational commitments. The current architectural problem is that Cloud Run service has public database IP access though all systems reside in one VPC. The team has recorded this evidence: connection logs show traffic to a globally routable database endpoint. Any proposed response must preserve the requirement that patient database must only accept internal workload connections.
The serverless private database access environment at Medical appointment SaaS faces this issue: Cloud Run service has public database IP access though all systems reside in one VPC. Supporting evidence shows that connection logs show traffic to a globally routable database endpoint. The architect must preserve the requirement that patient database must only accept internal workload connections. Select TWO complementary actions: one durable target architecture and one controlled way to introduce it while preserving service. Which TWO are best?
- Map serverless egress path, database listeners and firewall rules
- Monitor egress routing drift and private connection saturation
- Permit the database to accept all public IPs because TLS is enabled
- Cut over service revision to private egress and remove public listener safely
- Use Cloud Run VPC egress integration and private database access
Explanation
Correct Answers
D:
Cut over service revision to private egress and remove public listener safely is a controlled means to introduce the approved architecture while meeting the requirement that patient database must only accept internal workload connections. 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.
E:
Use Cloud Run VPC egress integration and private database access targets the underlying issue (Cloud Run service has public database IP access though all systems reside in one VPC) rather than merely reporting symptoms. It also preserves the required outcome (patient database must only accept internal workload connections), which a target-state design must satisfy. This supplies the lasting-control half of the two-action package.
Incorrect Answers
A:
This diagnoses the original issue (Cloud Run service has public database IP access though all systems reside in one VPC), but the TWO requested actions are the target architecture and its implementation; discovery was already completed.
B:
Monitoring serverless private database access for future drift is useful after the rollout, but it does not specify the requested target control and safe implementation for this case: Cloud Run service has public database IP access though all systems reside in one VPC.
C:
TLS does not meet the private reachability requirement. The shortcut does not supply the requested design and safe rollout for serverless private database access, where connection logs show traffic to a globally routable database endpoint.
Question 846
Consider Wildlife conservation nonprofit and its field-research collection workload. The issue is that donor funding requires an inexpensive platform but volunteers collect sensitive species locations; the current observation is that team interviews identify intermittent connectivity and controlled-location access. The team must ensure that precise nesting coordinates must remain inaccessible to public donors. The team has completed discovery. Which preventive or corrective architecture is most appropriate?
- Publish exact endangered-species coordinates to prove collection completeness
- Use offline-capable mobile capture with encrypted private storage and aggregated donor reports
- Watch stale devices, data-sharing grants and synchronization conflicts
- Classify location sensitivity, offline sync needs, donation objectives and cost boundaries
Explanation
Correct Answer
B:
Use offline-capable mobile capture with encrypted private storage and aggregated donor reports targets the underlying issue (donor funding requires an inexpensive platform but volunteers collect sensitive species locations) rather than merely reporting symptoms. It also preserves the required outcome (precise nesting coordinates must remain inaccessible to public donors), which a target-state design must satisfy.
Incorrect Answers
A:
Sensitive location disclosure could enable poaching and defeat privacy requirements. For this question, it is not a defensible choice for architecture selection and it does not satisfy the requirement that precise nesting coordinates must remain inaccessible to public donors.
C:
The alternative “Watch stale devices, data-sharing grants and synchronization conflicts” measures results without defining the preventive design now required. In the reported observation, team interviews identify intermittent connectivity and controlled-location access. The option “Use offline-capable mobile capture with encrypted private storage and aggregated donor reports” instead addresses the requested architecture selection.
D:
The alternative “Classify location sensitivity, offline sync needs, donation objectives and cost boundaries” investigates the cause without specifying the durable architecture to resolve it. In the reported observation, team interviews identify intermittent connectivity and controlled-location access. The option “Use offline-capable mobile capture with encrypted private storage and aggregated donor reports” instead addresses the requested architecture selection.
Question 847
At Multinational consulting alliance, the Google Cloud design for shared IAM inheritance is affected because a merger added interns to a group that has editor permission on financial-production resources. A review found that IAM policy audit shows project roles inherited from the organization root. The proposed response must account for the requirement that production access must be limited to verified operational duties. Which analysis will allow the architect to distinguish the reported symptom from the actual failure path?
- Identify inherited member groups and smallest needed job privileges
- Stage grant changes with access simulation and validated rollback
- Create a new group but keep the old wildcard organization binding
- Move broad grants to scoped groups and use conditional role bindings where appropriate
Explanation
Correct Answer
A:
Identify inherited member groups and smallest needed job privileges is the appropriate diagnostic because the observation—IAM policy audit shows project roles inherited from the organization root—does not by itself establish every dependency, authorization rule or failure boundary. The findings explain the reported risk (a merger added interns to a group that has editor permission on financial-production resources) before the design is approved.
Incorrect Answers
B:
The alternative “Stage grant changes with access simulation and validated rollback” is rollout work that may modify the wrong dependency before root-cause evidence is collected. In the reported observation, IAM policy audit shows project roles inherited from the organization root. The option “Identify inherited member groups and smallest needed job privileges” instead addresses the requested root-cause evidence.
C:
A new group does not negate an existing broad inherited permission. For this question, it is not a defensible choice for root-cause evidence and it does not satisfy the requirement that production access must be limited to verified operational duties.
D:
The alternative “Move broad grants to scoped groups and use conditional role bindings where appropriate” is target architecture selection; it does not establish why the observed path failed. In the reported observation, IAM policy audit shows project roles inherited from the organization root. The option “Identify inherited member groups and smallest needed job privileges” instead addresses the requested root-cause evidence.
Question 848
Consider National retail cooperative and its audit log preservation workload. The issue is that a project owner can disable or delete local logging after an unauthorized configuration change; the current observation is that audit exercise shows log sinks owned by the affected project. The team must ensure that investigations require independently retained administrative events. The target architecture is approved; execution without disruption is the remaining decision. Which rollout action is best?
- Store the only audit copy in a bucket controlled by the investigated project owner
- Aggregate audit logs into a protected central project with minimal deletion rights
- Watch logging exclusions, sink failures and retention violations
- Stage organization-level sinks and validate destination authorization
Explanation
Correct Answer
D:
Stage organization-level sinks and validate destination authorization is a controlled means to introduce the approved architecture while meeting the requirement that investigations require independently retained administrative events. Staged delivery, attributable change evidence and rollback guard against an otherwise avoidable service disruption.
Incorrect Answers
A:
Shared deletion authority undermines evidentiary independence. For this question, it is not a defensible choice for controlled implementation and it does not satisfy the requirement that investigations require independently retained administrative events.
B:
The alternative “Aggregate audit logs into a protected central project with minimal deletion rights” describes the chosen design, but not how it is introduced safely. In the reported observation, audit exercise shows log sinks owned by the affected project. The option “Stage organization-level sinks and validate destination authorization” instead addresses the requested controlled implementation.
C:
The alternative “Watch logging exclusions, sink failures and retention violations” addresses later monitoring rather than the actual controlled deployment. In the reported observation, audit exercise shows log sinks owned by the affected project. The option “Stage organization-level sinks and validate destination authorization” instead addresses the requested controlled implementation.
Question 849
Case study: National airport lounge operator: managed Prometheus scraping
National airport lounge operator relies on managed Prometheus scraping to deliver its operational commitments. The current architectural problem is that several GKE services expose metrics but the monitoring system has blind spots after namespace moves. The team has recorded this evidence: scrape health shows targets disappeared when Kubernetes labels changed. Any proposed response must preserve the requirement that capacity alerts must remain tied to actual service discovery.
National airport lounge operator plans to update managed Prometheus scraping. Investigators found that scrape health shows targets disappeared when Kubernetes labels changed, consistent with the reported problem that several GKE services expose metrics but the monitoring system has blind spots after namespace moves. The response must satisfy the requirement that capacity alerts must remain tied to actual service discovery. Select THREE evidence activities to cover (1) diagnosis before design, (2) direct post-change acceptance, and (3) continued operational monitoring. Which THREE fit those stages?
- Create a temporary test pod and verify metrics are collected and attributed
- Treat all unlabeled zero metrics as healthy application inactivity
- Configure Managed Service for Prometheus collectors with stable target discovery
- Inspect Kubernetes service selectors, relabeling and scrape permissions
- Watch missing targets, cardinality growth and failed scrapes
- Exclude the affected dependency from all future tests and incident metrics to simplify reporting
Explanation
Correct Answers
A:
Create a temporary test pod and verify metrics are collected and attributed is a direct acceptance test of the path implicated by scrape health shows targets disappeared when Kubernetes labels changed. It checks that the original risk (several GKE services expose metrics but the monitoring system has blind spots after namespace moves) has actually been controlled and tests that permitted operations still meet the requirement that capacity alerts must remain tied to actual service discovery. It supplies the separate post-change acceptance result requested.
D:
Inspect Kubernetes service selectors, relabeling and scrape permissions is the appropriate diagnostic because the observation—scrape health shows targets disappeared when Kubernetes labels changed—does not by itself establish every dependency, authorization rule or failure boundary. The findings explain the reported risk (several GKE services expose metrics but the monitoring system has blind spots after namespace moves) before the design is approved. It supplies the pre-design diagnostic evidence requested.
E:
Watch missing targets, cardinality growth and failed scrapes reveals recurrence, drift or changing dependencies after launch. It provides a repeatable signal for the risk (several GKE services expose metrics but the monitoring system has blind spots after namespace moves) rather than treating the first successful test as permanent assurance. It supplies the ongoing observation requested.
Incorrect Answers
B:
Missing scrape targets can look like falsely healthy zeros. It cannot satisfy the required evidence stages.
C:
Designing managed Prometheus scraping addresses the root cause, but this specific three-part question asks for evidence to diagnose, accept and continuously verify the response to several GKE services expose metrics but the monitoring system has blind spots after namespace moves, not another design proposal.
F:
Omitting the relevant dependency from testing and monitoring would hide, not correct, recurrence of the failure. In particular, the observed issue was: scrape health shows targets disappeared when Kubernetes labels changed.
Question 850
Consider Hospital imaging research unit and its regulated data classification workload. The issue is that an AI training workflow ingests medical reports with accidental patient identifiers; the current observation is that dataset profiler finds patient IDs inside free-text radiology comments. The team must ensure that only approved deidentified inputs may enter model training. Which acceptance test should demonstrate both protection against the failure and continued authorized service?
- Insert deidentification before training and quarantine unresolved findings
- Verify holdout samples contain no prohibited identifiers while clinical utility remains
- Classify identifiable fields and exposure paths with sample inspections
- Trust a generic prompt to remove every patient ID from raw input
Explanation
Correct Answer
B:
Verify holdout samples contain no prohibited identifiers while clinical utility remains is a direct acceptance test of the path implicated by dataset profiler finds patient IDs inside free-text radiology comments. It checks that the original risk (an AI training workflow ingests medical reports with accidental patient identifiers) has actually been controlled and tests that permitted operations still meet the requirement that only approved deidentified inputs may enter model training.
Incorrect Answers
A:
The alternative “Insert deidentification before training and quarantine unresolved findings” documents change execution but not its operational acceptance result. In the reported observation, dataset profiler finds patient IDs inside free-text radiology comments. The option “Verify holdout samples contain no prohibited identifiers while clinical utility remains” instead addresses the requested acceptance verification.
C:
The alternative “Classify identifiable fields and exposure paths with sample inspections” collects prechange evidence without confirming that the revised service meets its outcome. In the reported observation, dataset profiler finds patient IDs inside free-text radiology comments. The option “Verify holdout samples contain no prohibited identifiers while clinical utility remains” instead addresses the requested acceptance verification.
D:
Prompt guidance is not a deterministic privacy-control boundary. For this question, it is not a defensible choice for acceptance verification and it does not satisfy the requirement that only approved deidentified inputs may enter model training.