Question 420
A Pod keeps restarting because its liveness probe has an overly aggressive threshold. What should the administrator do?
- Adjust the liveness probe timing or failure threshold based on startup behavior
- Use gcloud compute ssh over a direct reachable address with appropriate access
- Inspect and fix readiness probes for the workload
- Run kubectl describe deployment with the correct namespace
- Run kubectl logs for the selected Pod or container
Explanation
Correct Answer
A:Liveness probes restart containers that appear unhealthy. Misconfigured thresholds can create repeated restarts even when an application would recover.
Incorrect Answers
B:This choice is useful to start a direct CLI SSH session to a VM. The stated task is to avoid premature health-probe restarts, which requires a different mechanism or permission scope.
C:This setting would control Pod traffic readiness. That behavior does not establish the requested ability to avoid premature health-probe restarts; it addresses an adjacent concern instead.
D:Using this action would inspect a Kubernetes controller's conditions, not avoid premature health-probe restarts. Its function is different from the specific change described in the scenario.
E:It can inspect a Pod container's emitted logs; however, it does not provide the specific operation to avoid premature health-probe restarts. The two controls serve different purposes.
Question 421
A platform group offers browser-accessible Cloud Workstations so contractors can use standardized development environments without local credential files. Which TWO controls should it use? (Choose TWO.)
- Grant eligible developers appropriately scoped access to the workstation resources
- Use a Cloud Storage bucket lifecycle policy as the workstation's execution runtime
- Place a shared service account private key inside every developer container image
- Publish an approved workstation configuration with the required image, network and idle-time settings
Explanation
Correct Answers
A:Cloud Workstations relies on IAM authorization for developers and administrators. Scoped role assignments let authorized contractors use their environments without granting unrelated project administration.
D:A workstation configuration defines supported environment and operational parameters that teams can provision consistently. Idle controls can help reduce unnecessary resource use.
Incorrect Answers
B:A lifecycle policy can move or remove objects, but it cannot host a running interactive development workstation or supply its compute and network environment.
C:Embedding a private key in a developer environment exposes a reusable credential across users. Managed identity access and delegated permissions are preferable to copied long-lived keys.
Question 422
A data analyst needs a managed notebook experience integrated with supported BigQuery analytics. How should the team proceed?
- Deploy it through Agent Runtime on Gemini Enterprise Agent Platform using supported tooling
- Use a suitable BigQuery notebook workflow
- Use Cloud Workstations with an approved workstation configuration
- Use the documented Workbench notebook management workflow
Explanation
Correct Answer
B:BigQuery-integrated notebooks support data exploration around analytical datasets. They are not a substitute for the BigQuery warehouse's storage and SQL engine.
Incorrect Answers
A:The tool can deploy a managed AI agent runtime, but it cannot be substituted for the capability needed to explore warehouse data in notebooks. Its operational effect is not equivalent.
C:It can provision managed development workstations; however, it does not provide the specific operation to explore warehouse data in notebooks. The two controls serve different purposes.
D:This mechanism is intended to administer managed ML notebook environments. It does not implement the required action to explore warehouse data in notebooks, even though both are relevant to the same general service.
Question 423
A machine-learning engineer needs to manage notebooks in the supported Gemini Enterprise Agent Platform Workbench offering. What is the most direct solution?
- Use the documented Workbench notebook management workflow
- Install compatible GPU drivers and configure the runtime
- Use a suitable BigQuery notebook workflow
- Use Cloud Workstations with an approved workstation configuration
- Deploy it through Agent Runtime on Gemini Enterprise Agent Platform using supported tooling
Explanation
Correct Answer
A:Workbench provides managed notebook tooling in supported locations and configurations. It does not convert a notebook into a production agent deployment automatically.
Incorrect Answers
B:This alternative addresses compute accelerator operations by allowing a team to prepare guest GPU driver access; The incident, however, requires the different action to operate a hosted notebook workspace, so its central configuration remains unchanged.
C:Its purpose is to work interactively with analytical notebooks. That distinction matters: the scenario calls for an action to operate a hosted notebook workspace, not this separate configuration task.
D:This setting would provision managed development workstations. That behavior does not establish the requested ability to operate a hosted notebook workspace; it addresses an adjacent concern instead.
E:Using this action would deploy a managed AI agent runtime, not operate a hosted notebook workspace. Its function is different from the specific change described in the scenario.
Question 424
A team has built an eligible agent and wants to host it on the managed Agent Runtime service. Which implementation matches this requirement?
- Use Cloud Workstations with an approved workstation configuration
- Use suitable VPA recommendations or explicit resource sizing changes
- Deploy it through Agent Runtime on Gemini Enterprise Agent Platform using supported tooling
- Use a suitable BigQuery notebook workflow
- Use the documented Workbench notebook management workflow
Explanation
Correct Answer
C:Agent Runtime provides managed hosting for eligible AI agent applications. Runtime deployment and agent permissions still require compatible application packaging and authorization.
Incorrect Answers
A:The result of this choice would be to provision managed development workstations. The requested outcome, to host an agent application, is handled at a different layer or through another operation.
B:The technical effect is to right-size individual Pod resources, not to host an agent application; That distinction matters here because the first operation serves adjust CPU and memory allocations, rather than the scenario's required result.
D:This mechanism is intended to work interactively with analytical notebooks. It does not implement the required action to host an agent application, even though both are relevant to the same general service.
E:Administrators use it to administer managed ML notebook environments, rather than to host an agent application. Applying it would change the wrong component of this deployment.
Question 425
A GPU-backed Compute Engine instance must expose the accelerator to a machine-learning framework. Which action is most appropriate?
- Install compatible GPU drivers and configure the runtime
- Validate the framework and accelerator compatibility before provisioning
- Adjust GPU node capacity or Pod GPU resource requests
- Verify accelerator-specific availability and quota before launching
Explanation
Correct Answer
A:An attached GPU needs supported guest drivers and library compatibility to be usable by software. Selecting a GPU machine type alone is not the complete guest setup.
Incorrect Answers
B:Using this action would match accelerators with ML framework support, not enable accelerator execution inside a VM. Its function is different from the specific change described in the scenario.
C:This choice is useful to align Pod GPU requests with node capacity. The stated task is to enable accelerator execution inside a VM, which requires a different mechanism or permission scope.
D:This setting would check accelerator-specific placement constraints. That behavior does not establish the requested ability to enable accelerator execution inside a VM; it addresses an adjacent concern instead.
Question 426
A GKE Pod cannot schedule because it requests more GPU devices than its nodes offer. Which solution should be used?
- Adjust GPU node capacity or Pod GPU resource requests
- Validate the framework and accelerator compatibility before provisioning
- Install compatible GPU drivers and configure the runtime
- Use Cloud Workstations with an approved workstation configuration
- Verify accelerator-specific availability and quota before launching
Explanation
Correct Answer
A:Kubernetes schedules accelerators as extended resources on eligible nodes. CPU requests do not reserve GPU devices.
Incorrect Answers
B:While this can match accelerators with ML framework support, the present task requires an action to resolve insufficient accelerator scheduling. This operation does not supply the behavior described in the requirements.
C:Administrators use it to prepare guest GPU driver access, rather than to resolve insufficient accelerator scheduling. Applying it would change the wrong component of this deployment.
D:This alternative addresses developer and ai platform operations by allowing a team to provision managed development workstations; The incident, however, requires the different action to resolve insufficient accelerator scheduling, so its central configuration remains unchanged.
E:Its purpose is to check accelerator-specific placement constraints. That distinction matters: the scenario calls for an action to resolve insufficient accelerator scheduling, not this separate configuration task.
Question 427
A model training team is choosing between GPU and TPU hardware based on framework support. How should the team proceed?
- Verify accelerator-specific availability and quota before launching
- Install compatible GPU drivers and configure the runtime
- Validate the framework and accelerator compatibility before provisioning
- Adjust Pod resource requests to meet Autopilot constraints
- Adjust GPU node capacity or Pod GPU resource requests
Explanation
Correct Answer
C:GPUs and TPUs differ in supported stacks and workload characteristics. Choosing solely by nominal hardware speed can result in incompatible execution.
Incorrect Answers
A:Select this when the goal is to check accelerator-specific placement constraints. Here the decisive requirement is to avoid incompatible TPU or GPU deployments, and that outcome is not achieved by this action.
B:This choice is useful to prepare guest GPU driver access. The stated task is to avoid incompatible TPU or GPU deployments, which requires a different mechanism or permission scope.
D:The technical effect is to fit Pod requests to Autopilot rules, not to avoid incompatible TPU or GPU deployments; That distinction matters here because the first operation serves resolve incompatible Autopilot resource settings, rather than the scenario's required result.
E:The tool can align Pod GPU requests with node capacity, but it cannot be substituted for the capability needed to avoid incompatible TPU or GPU deployments. Its operational effect is not equivalent.
Question 428
A training experiment needs dedicated managed accelerator resources within product-supported regions and quota. What is the most direct solution?
- Validate the framework and accelerator compatibility before provisioning
- Use the documented Workbench notebook management workflow
- Verify accelerator-specific availability and quota before launching
- Adjust GPU node capacity or Pod GPU resource requests
- Install compatible GPU drivers and configure the runtime
Explanation
Correct Answer
C:Accelerator capacity depends on regional support and quota as well as machine configuration. Ordinary CPU quota does not guarantee GPU or TPU availability.
Incorrect Answers
A:This setting would match accelerators with ML framework support. That behavior does not establish the requested ability to provision scarce AI compute correctly; it addresses an adjacent concern instead.
B:Choosing this would administer managed ML notebook environments, an appropriate action for developer and ai platform operations operations; It cannot take the place of the separate step needed to provision scarce AI compute correctly in this specific situation.
D:While this can align Pod GPU requests with node capacity, the present task requires an action to provision scarce AI compute correctly. This operation does not supply the behavior described in the requirements.
E:Its purpose is to prepare guest GPU driver access. That distinction matters: the scenario calls for an action to provision scarce AI compute correctly, not this separate configuration task.
Question 429
A stateless VM group must auto-replace members that fail a readiness check. What should be configured?
- Use gcloud compute ssh over a direct reachable address with appropriate access
- Change the VM machine type through a supported stop and update workflow
- Replicate the service across zones behind an appropriate load-balancing design
- Configure scheduled MIG autoscaling for the expected interval
- Enable MIG autohealing with an appropriate health check
Explanation
Correct Answer
E:MIG autohealing recreates unhealthy member VMs against the group template. It is different from autoscaling based on overall traffic volume.
Incorrect Answers
A:It can start a direct CLI SSH session to a VM; however, it does not provide the specific operation to maintain healthy replica count. The two controls serve different purposes.
B:Select this when the goal is to resize a VM machine type. Here the decisive requirement is to maintain healthy replica count, and that outcome is not achieved by this action.
C:This mechanism is intended to deploy a multi-zone application tier. It does not implement the required action to maintain healthy replica count, even though both are relevant to the same general service.
D:The tool can schedule predictable VM fleet capacity, but it cannot be substituted for the capability needed to maintain healthy replica count. Its operational effect is not equivalent.
Question 430
A managed instance group contains VMs with per-instance Persistent Disks holding checkpoint data. Members must survive automated repair and rolling updates without replacing their preserved disks with blank volumes. Which MIG configuration supports this?
- Rely on ordinary scheduled autoscaling to keep the disks attached after an instance is deleted
- Enable only a load balancer health check without specifying any preserved state
- Use Cloud Storage object lifecycle rules as the stateful disk policy for the VM group
- Configure a stateful MIG policy or per-instance preserved-state configuration for the required disks
Explanation
Correct Answer
D:Stateful MIGs retain configured instance-specific disk state through supported recreation, autohealing, and updates. The group must preserve the relevant disks rather than treating them as stateless template resources.
Incorrect Answers
A:A scaling schedule controls group capacity; it does not declare per-instance disks as preserved state. Removing or replacing instances can affect state without the proper stateful configuration.
B:A health check can detect failures but does not tell MIG automation which data disks must be retained across recreation. State preservation needs explicit settings.
C:Bucket object lifecycle rules apply to Cloud Storage objects, not the persistence behavior of Compute Engine disks during a MIG instance recreation.