Google Associate Cloud Engineer Practice Test 48
Question 470
A team deploys a new container and then loses the prior image tag because it reused a mutable latest tag, undermining rollback. What should it retain?
- Produce and validate a governed custom image or image family from the approved golden source with version control.
- Deploy using immutable image digests or versioned tags and keep previous proven images available for rollback.
- Increase the VM vCPU quota without changing the failed configuration or resource setting.
- Change the instance display name as a replacement for modifying the actual machine or disk.
Explanation
Correct Answer
B:Container image references must be reproducible; mutable tags can point to different bytes across releases.
Incorrect Answers
A:The proposed setting does not correct the stated machine, disk or scheduling requirement. It leaves the actual compute scheduling or resource condition unaddressed. The appropriate action here is: Deploy using immutable image digests or versioned tags and keep previous proven images available for rollback.
C:More CPU quota affects capacity, not the API or policy decision described. The specific requirement concerns the actual compute scheduling or resource condition. The appropriate action here is: Deploy using immutable image digests or versioned tags and keep previous proven images available for rollback.
D:Display names and labels identify resources; they do not enforce their runtime behavior. The necessary distinction is the actual compute scheduling or resource condition. The appropriate action here is: Deploy using immutable image digests or versioned tags and keep previous proven images available for rollback.
Question 471
Cloud Run revision B returns 5xx only when calling a private internal API. Revision A succeeds against the same API. B changed its VPC egress settings and service identity. Which comparison best isolates the problem?
- Compare Cloud Billing budgets for revision A and B; billing thresholds are the only revision-specific controls that govern private API requests.
- Compare revision-specific network egress, connector or direct-VPC configuration, and the runtime identity's effective access to the private API.
- Recreate all Cloud Storage buckets used by both revisions; object storage automatically repairs private TCP paths.
- Change the Cloud Run service display name, because private DNS and IAM derive permissions from the human-readable service name.
Explanation
Correct Answer
B:Compare revision-specific network egress, connector or direct-VPC configuration, and the runtime identity's effective access to the private API. Cloud Run revisions can have distinct effective network and identity configuration. The failing dependency path should be checked against the healthy revision.
Incorrect Answers
A:Compare Cloud Billing budgets for revision A and B; billing thresholds are the only revision-specific controls that govern private API requests. Cost budgets do not determine whether one revision has a working private VPC path or a permitted identity for an internal dependency.
C:Recreate all Cloud Storage buckets used by both revisions; object storage automatically repairs private TCP paths. Replacing unrelated buckets does not fix an internal service route or IAM permission that differs by Cloud Run revision.
D:Change the Cloud Run service display name, because private DNS and IAM derive permissions from the human-readable service name. Renaming a service is not a substitute for inspecting revision egress configuration, private DNS reachability and service account rights.
Question 472
An API that needs to stream a response for longer than an HTTP request timeout is killed while a background report would be valid. What needs separating?
- Use a request-level idempotency key or stored execution marker and ensure duplicate invocations are safe.
- Create a snapshot only and assume it immediately changes the live VM capacity.
- Give the attached VM service account project Owner instead of fixing resource configuration.
- Keep synchronous calls within supported request limits and move prolonged report work to a job or asynchronous workflow.
Explanation
Correct Answer
D:Cloud Run request timeouts are serving constraints; increasing unrelated VM quota does not make long background work safe inside an HTTP request.
Incorrect Answers
A:The proposed setting does not correct the stated machine, disk or scheduling requirement. The necessary distinction is the actual compute scheduling or resource condition. The appropriate action here is: Keep synchronous calls within supported request limits and move prolonged report work to a job or asynchronous workflow.
B:Snapshots are recovery artifacts and do not directly reconfigure a running resource. The case instead depends on the actual compute scheduling or resource condition. The appropriate action here is: Keep synchronous calls within supported request limits and move prolonged report work to a job or asynchronous workflow.
C:Project Owner broadens authorization and is not an appropriate substitute for the named control. It leaves the actual compute scheduling or resource condition unaddressed. The appropriate action here is: Keep synchronous calls within supported request limits and move prolonged report work to a job or asynchronous workflow.
Question 473
A Cloud Run deployment created revision B with a new environment variable, yet users still observe the previous value. Traffic splitting shows that revision A continues serving 60% of requests. What should the team verify?
- Delete the environment variable from revision A; modifying its immutable revision changes the environment for all active requests.
- Increase regional CPU quota; additional CPU automatically routes all traffic to the newest configuration.
- Identify the revision handling each request, test revision B, and move the approved traffic allocation away from A when ready.
- Change the project's Cloud Billing account so Cloud Run will stop routing to historical revisions.
Explanation
Correct Answer
C:Identify the revision handling each request, test revision B, and move the approved traffic allocation away from A when ready. Traffic distribution can still send requests to revision A with the old environment variable. A controlled migration to tested B resolves the mixed observed values.
Incorrect Answers
A:Delete the environment variable from revision A; modifying its immutable revision changes the environment for all active requests. Cloud Run revisions are immutable; an older revision retains its previous configuration while receiving traffic.
B:Increase regional CPU quota; additional CPU automatically routes all traffic to the newest configuration. Regional resource quota concerns capacity, not the Cloud Run traffic split between immutable configurations.
D:Change the project's Cloud Billing account so Cloud Run will stop routing to historical revisions. Billing changes do not determine which Cloud Run revision serves individual HTTPS requests. The traffic split must be managed.
Question 474
Cloud Run executes a request-triggered inference service with CPU-intensive processing, but settings throttle CPU outside requests while an unrelated background goroutine must continue after response. What should be redesigned?
- Create a Cloud Run job for finite background processing with appropriate task timeout and retries.
- Move post-response background work to a dedicated job/task queue or adopt an explicitly supported always-allocated CPU design when justified.
- Increase Cloud Run memory alone while leaving the failing routing, identity or event setting unchanged.
- Switch the service to public unauthenticated ingress as a workaround for the stated requirement.
Explanation
Correct Answer
B:Background processing after HTTP completion cannot safely assume request-scoped CPU and instance lifetime guarantees.
Incorrect Answers
A:Timeouts control waiting time, not the missing platform feature or access right. It leaves the Cloud Run service setting unaddressed.
C:Memory sizing may change capacity but cannot replace the specified platform feature. The specific requirement concerns the Cloud Run service setting. The appropriate action here is: Move post-response background work to a dedicated job/task queue or adopt an explicitly supported always-allocated CPU design when justified.
D:The change does not configure the Cloud Run runtime or traffic behavior at issue. The necessary distinction is the Cloud Run service setting. The appropriate action here is: Move post-response background work to a dedicated job/task queue or adopt an explicitly supported always-allocated CPU design when justified.
Question 475
An Agent Runtime endpoint is deployed, but error reports show tool calls are timing out while top-level model inference succeeds. What should operators instrument?
- Test service and dependency availability, identity grants, model routing and data-residency constraints in the standby region before cutover.
- Make every agent tool publicly callable to avoid the required access-policy configuration.
- Enable scoped agent/tool tracing and inspect per-tool latency and failure spans before changing model deployment capacity.
- Give the agent project Owner instead of configuring its supported runtime identity and tool access.
Explanation
Correct Answer
C:An agent can fail in external tools after the model call; separating model and tool execution paths identifies the faulty component.
Incorrect Answers
A:Placement changes do not automatically implement the required resource-level feature. The specific requirement concerns the requested latency measurement.
B:Public access expands exposure without meeting the requested scoped-control requirement. The necessary distinction is the requested latency measurement.
D:Project Owner broadens authorization and is not an appropriate substitute for the named control. The case instead depends on the requested latency measurement.
Question 476
An operations team updates an agent's tool schema, and existing clients still send payloads matching the previous contract. What should the release process include?
- Enable scoped agent/tool tracing and inspect per-tool latency and failure spans before changing model deployment capacity.
- Scale the notebook CPU without changing any managed agent deployment settings.
- Make every agent tool publicly callable to avoid the required access-policy configuration.
- Version the agent interface and conduct backward-compatibility tests before routing existing callers to the updated tool schema.
Explanation
Correct Answer
D:Agent deployments expose application contracts; tool schema changes can break callers without affecting the underlying model.
Incorrect Answers
A:Deploying a separate workload does not automatically configure the existing feature. The necessary distinction is the managed agent feature or authorization. The appropriate action here is: Version the agent interface and conduct backward-compatibility tests before routing existing callers to the updated tool schema.
B:Deploying a separate workload does not automatically configure the existing feature. The case instead depends on the managed agent feature or authorization.
C:Public access expands exposure without meeting the requested scoped-control requirement. It leaves the managed agent feature or authorization unaddressed.
Question 477
An Agent Runtime service reads an internal search index using a development Owner identity. Production policy requires read-only access to one index and no unrelated project administration. What identity change fits?
- Attach a dedicated runtime service account with only the required index-reading permissions and supported private connectivity.
- Keep project Owner but add a label stating read-only; resource labels subtract permissions from IAM roles.
- Grant organization Owner to the agent registry service so all runtime calls are automatically limited to one index.
- Make the index anonymously readable so the agent no longer needs to manage its runtime credentials.
Explanation
Correct Answer
A:Attach a dedicated runtime service account with only the required index-reading permissions and supported private connectivity. A service-specific least-privilege identity limits API authorization to the necessary index operations. It avoids retaining broad development privileges.
Incorrect Answers
B:Keep project Owner but add a label stating read-only; resource labels subtract permissions from IAM roles. Labels annotate resources but do not narrow an Owner role. The broad grant would remain effective despite a read-only label.
C:Grant organization Owner to the agent registry service so all runtime calls are automatically limited to one index. An organization-wide Owner grant is even broader than the existing development role and does not restrict index access.
D:Make the index anonymously readable so the agent no longer needs to manage its runtime credentials. Anonymous data access removes authentication protections rather than safely narrowing the agent's production identity.
Question 478
A deployed agent repeatedly emits the same downstream event after a client retry, resulting in duplicate service tickets. Which operational safeguard is needed?
- Use idempotency identifiers and durable deduplication for ticket-creation tools invoked by the agent.
- Enable scoped agent/tool tracing and inspect per-tool latency and failure spans before changing model deployment capacity.
- Scale the notebook CPU without changing any managed agent deployment settings.
- Give the agent project Owner instead of configuring its supported runtime identity and tool access.
Explanation
Correct Answer
A:Retryable agent/tool invocations can repeat side effects, so mutable operations must be designed for at-least-once attempts.
Incorrect Answers
B:Deploying a separate workload does not automatically configure the existing feature. It leaves the managed agent feature or authorization unaddressed. The appropriate action here is: Use idempotency identifiers and durable deduplication for ticket-creation tools invoked by the agent.
C:Deploying a separate workload does not automatically configure the existing feature. The specific requirement concerns the managed agent feature or authorization.
D:Project Owner broadens authorization and is not an appropriate substitute for the named control. The necessary distinction is the managed agent feature or authorization. The appropriate action here is: Use idempotency identifiers and durable deduplication for ticket-creation tools invoked by the agent.
Question 479
A remote agent rollout shows a sudden increase in per-request token cost after a prompt update, but no version-level usage tags exist. What should be added?
- Track model token quotas, concurrent sessions and downstream tool rate limits alongside runtime resource metrics.
- Attribute request, prompt/model version, token usage and downstream tool cost in scoped runtime telemetry.
- Increase the agent service CPU limits while leaving the expanded prompt and model token usage unchanged.
- Give the agent project Owner instead of configuring its supported runtime identity and tool access.
Explanation
Correct Answer
B:Token consumption and tool calls can differ by agent revision; version attribution is needed to detect a cost regression.
Incorrect Answers
A:The suggested action changes another component without meeting the main requirement. The specific requirement concerns the managed agent feature or authorization. The appropriate action here is: Attribute request, prompt/model version, token usage and downstream tool cost in scoped runtime telemetry.
C:CPU limits affect execution capacity, not the number of model input and output tokens billed.
D:Project Owner broadens authorization and is not an appropriate substitute for the named control. The case instead depends on the managed agent feature or authorization.
Question 480
A team plans to delete a prior Agent Runtime deployment immediately after creating a new release, but rollback has not been tested. What should be preserved?
- Retain a known-good deployed version and explicit traffic or endpoint rollback plan through an approved validation window.
- Enable scoped agent/tool tracing and inspect per-tool latency and failure spans before changing model deployment capacity.
- Keep the agent only in the notebook process and assume its session memory is a durable shared store.
- Make every agent tool publicly callable to avoid the required access-policy configuration.
Explanation
Correct Answer
A:A successful deployment command alone does not establish service stability; removing rollback artifacts increases outage risk.
Incorrect Answers
B:Deploying a separate workload does not automatically configure the existing feature. The necessary distinction is the managed agent feature or authorization. The appropriate action here is: Retain a known-good deployed version and explicit traffic or endpoint rollback plan through an approved validation window.
C:Memory sizing may change capacity but cannot replace the specified platform feature. The case instead depends on the managed agent feature or authorization.
D:Public access expands exposure without meeting the requested scoped-control requirement. It leaves the managed agent feature or authorization unaddressed. The appropriate action here is: Retain a known-good deployed version and explicit traffic or endpoint rollback plan through an approved validation window.