Google Associate Cloud Engineer Practice Test 25
Question 240
A GKE Autopilot service uses an Arm-only container image in a supported region. Its Pod has no architecture selection and starts on x86 nodes. What should the team specify?
- Set a VPC route priority of arm64; node architecture follows the network route chosen by the Pod.
- Specify a compatible Arm compute class or supported kubernetes.io/arch: arm64 scheduling selector for the cluster version.
- Increase the CPU limit without adding node architecture attributes; Kubernetes converts x86 hardware to Arm automatically.
- Set the Cloud Storage object metadata architecture=arm64; the scheduler uses bucket metadata to select nodes.
Explanation
Correct Answer
B:Specify a compatible Arm compute class or supported kubernetes.io/arch: arm64 scheduling selector for the cluster version. The documented Autopilot Arm deployment paths use architecture selectors or an eligible Arm compute class. Matching the image to an Arm node avoids startup incompatibility.
Incorrect Answers
A:Set a VPC route priority of arm64; node architecture follows the network route chosen by the Pod. Routing determines traffic paths, not instruction-set architecture or kube-scheduler placement.
C:Increase the CPU limit without adding node architecture attributes; Kubernetes converts x86 hardware to Arm automatically. CPU quantity does not specify architecture. An Arm-only image cannot run on an incompatible x86-only node merely by requesting more cores.
D:Set the Cloud Storage object metadata architecture=arm64; the scheduler uses bucket metadata to select nodes. Cloud Storage object labels are not Kubernetes node scheduling constraints and cannot direct the Pod to an Arm node.
Question 241
A GKE Autopilot cluster runs mixed production Pods, and one batch job should use the supported Scale-Out compute class without moving unrelated workloads. What belongs in its manifest?
- Specify the supported Autopilot compute class in the particular workload’s Pod node selector.
- Convert every workload in the cluster to the new class because GKE supports only one class per cluster.
- Change the cluster’s billing account to Scale-Out; billing controls Kubernetes compute class selection.
- Add an HTTP header to the job’s container; the scheduler derives node hardware from request headers.
Explanation
Correct Answer
A:Autopilot supports per-workload placement onto a selected compute class using the documented selector, with compatible resource requests. Update only the job’s Pod template and verify the selected hardware serves that job.
Incorrect Answers
B:Workload-level selection need not force unrelated Pods onto the same class.
C:A billing link controls charges, not which compute class schedules a Pod.
D:Application HTTP request headers are not a Kubernetes scheduling mechanism and cannot select node compute classes.
Question 242
A regular GKE Autopilot Pod requests less CPU than its selected compute class minimum. Admission changes the request automatically. How should the engineer interpret this?
- Keep the below-minimum CPU request; any admission mutation is a transient display error.
- Change only the Pod name, since admission minimums are based on labels.
- Set every CPU request to zero so Autopilot stops adjusting resource specifications.
- Expect Autopilot to raise the unsupported request to the applicable minimum; specify a valid request explicitly.
Explanation
Correct Answer
D:For an eligible Pod without special admission exceptions, Autopilot enforces compute-class minimums and may raise requested CPU. Inspect the admitted spec and choose an explicit supported value to anticipate scheduling and cost.
Incorrect Answers
A:Autopilot can modify undersized requests to meet supported minimums rather than accepting the literal value.
B:Renaming a Pod does not change compute-class minimum resource requirements.
C:A zero request can trigger defaulting or mutation, not an exemption from resource policies.
Question 243
A sensitive batch job must use only Spot capacity, not the standard on-demand compute class. What scheduling feature should be selected?
- Request a supported Autopilot Spot compute class or Spot Pod configuration for that workload.
- Reduce requests to supported bounds or choose a compute class/hardware configuration with sufficient allowed memory.
- Use a public LoadBalancer Service for an internal-only issue instead of fixing its actual boundary.
- Grant the GKE node service account project Owner to bypass the workload-specific issue.
Explanation
Correct Answer
A:Autopilot can schedule interruptible workloads using Spot-specific placement controls.
Incorrect Answers
B:Memory sizing may change capacity but cannot replace the specified platform feature. The specific requirement concerns Spot placement and interruption tolerance.
C:This changes a neighboring Kubernetes feature, not the stated control. The necessary distinction is Spot placement and interruption tolerance.
D:Project Owner broadens authorization and is not an appropriate substitute for the named control. The case instead depends on Spot placement and interruption tolerance.
Question 244
A critical process cannot tolerate Spot preemption but a developer reused a Spot Pod template. What must change?
- Request a supported Autopilot Spot compute class or Spot Pod configuration for that workload.
- Remove the Spot scheduling requirement and run on supported non-Spot capacity with appropriate availability settings.
- Disable the cluster health checks and treat missing readiness as a successful deployment.
- Grant the GKE node service account project Owner to bypass the workload-specific issue.
Explanation
Correct Answer
B:Spot Pods may be interrupted and are unsuitable for stateful operations without recovery tolerance.
Incorrect Answers
A:Changing an unrelated Pod setting leaves the stated cloud feature unresolved. The necessary distinction is Spot placement and interruption tolerance.
C:Deploying a separate workload does not automatically configure the existing feature. The case instead depends on Spot placement and interruption tolerance.
D:Project Owner broadens authorization and is not an appropriate substitute for the named control. It leaves Spot placement and interruption tolerance unaddressed.
Question 245
An operations team uses GKE Standard and assumes node pools are automatically resized and repaired with no node-pool settings or operator responsibility. What configuration principle applies?
- Configure supported node-pool autoscaling and auto-repair policies explicitly; continue managing the Standard cluster's node lifecycle responsibilities.
- Standard clusters behave exactly like Autopilot for every node operation, regardless of their node-pool configuration.
- Create a Cloud Billing budget so the budget service installs cluster autoscaling on every Standard node pool.
- Install a Cloud SQL read replica; replicated databases enable automatic node repair in GKE.
Explanation
Correct Answer
A:Configure supported node-pool autoscaling and auto-repair policies explicitly; continue managing the Standard cluster's node lifecycle responsibilities. Standard mode exposes node-pool management controls. Configured autoscaling and auto-repair automate selected tasks but do not remove the team's responsibility for node settings.
Incorrect Answers
B:Standard clusters behave exactly like Autopilot for every node operation, regardless of their node-pool configuration. Autopilot and Standard differ in responsibility for nodes. Standard node-pool behavior is not identical to fully managed Autopilot capacity.
C:Create a Cloud Billing budget so the budget service installs cluster autoscaling on every Standard node pool. Budgets monitor charges. They do not set the node-pool min/max sizes or enable repair policy features.
D:Install a Cloud SQL read replica; replicated databases enable automatic node repair in GKE. A Cloud SQL replica protects database availability, not the lifecycle state of Kubernetes worker nodes.
Question 246
A GKE Pod is scheduled to a GPU node and requests nvidia.com/gpu correctly, but its CUDA application fails because the container framework does not support the installed CUDA/driver stack. What should be validated?
- Add more CPU limits to the Pod; CPU scheduling automatically upgrades the CUDA framework in the image.
- Use a compatible CUDA-enabled container and verify the node's supported GPU driver/runtime and application framework versions.
- Configure a Cloud DNS A record for the GPU; GPUs are accessed through DNS rather than local device drivers.
- Grant the Pod a broader organization IAM role; GPU driver compatibility is checked by Google Cloud IAM.
Explanation
Correct Answer
B:Use a compatible CUDA-enabled container and verify the node's supported GPU driver/runtime and application framework versions. Accelerator allocation can succeed while the process fails from ABI or CUDA incompatibility. Validating the complete node driver and container stack targets the observed startup error.
Incorrect Answers
A:Add more CPU limits to the Pod; CPU scheduling automatically upgrades the CUDA framework in the image. CPU requests do not install or reconcile device drivers, CUDA libraries, or application frameworks inside the container.
C:Configure a Cloud DNS A record for the GPU; GPUs are accessed through DNS rather than local device drivers. GPU device access is provided through the node/container runtime, not an externally resolved DNS hostname.
D:Grant the Pod a broader organization IAM role; GPU driver compatibility is checked by Google Cloud IAM. IAM grants affect API authorization and do not make a CUDA binary compatible with the GPU driver loaded by the node.
Question 247
A pod runs at high CPU for hours but its Deployment has no horizontal autoscaler. What scaling object is appropriate?
- Read Pod scheduling events, requested memory, node allocatable capacity, constraints and cluster autoscaler diagnostics.
- Configure a HorizontalPodAutoscaler with valid resource requests and a supported utilization or custom metric.
- Increase the Kubernetes Service port without changing the failing Pod or cluster configuration.
- Disable the cluster health checks and treat missing readiness as a successful deployment.
Explanation
Correct Answer
B:HPA manages replica count based on configured measurements; high CPU alone does not create an autoscaler.
Incorrect Answers
A:Memory sizing may change capacity but cannot replace the specified platform feature. The specific requirement concerns horizontal Pod replica scaling.
C:A Kubernetes Service port does not change a Pod’s scheduling, image or identity. The necessary distinction is horizontal Pod replica scaling.
D:Deploying a separate workload does not automatically configure the existing feature. The case instead depends on horizontal Pod replica scaling.
Question 248
A Kubernetes service must be reachable from other Pods inside the cluster, but the developer creates a NodePort exposed on every node. Which service type is usually more appropriate?
- Use ClusterIP for internal-only cluster service discovery where external exposure is unnecessary.
- Make the Service selector match the intended Pods’ labels, and confirm that ready endpoints are discovered.
- Move all Pods to the default namespace and assume this repairs scheduling and authorization.
- Grant the GKE node service account project Owner to bypass the workload-specific issue.
Explanation
Correct Answer
A:ClusterIP offers in-cluster service networking without the broader NodePort exposure model.
Incorrect Answers
B:Changing an unrelated Pod setting leaves the stated cloud feature unresolved. The necessary distinction is the Kubernetes workload or scheduling control.
C:Changing an unrelated Pod setting leaves the stated cloud feature unresolved. The case instead depends on the Kubernetes workload or scheduling control.
D:Project Owner broadens authorization and is not an appropriate substitute for the named control. It leaves the Kubernetes workload or scheduling control unaddressed.
Question 249
A public application requires HTTP path-based routing to two Kubernetes Services, but the team configures only individual ClusterIP services. What ingress resource is relevant?
- Separate application ingress from control-plane access; configure private/restricted cluster access and approved administrator paths.
- Deploy a supported Kubernetes Ingress or Gateway configuration backed by an appropriate load balancer.
- Move all Pods to the default namespace and assume this repairs scheduling and authorization.
- Use a public LoadBalancer Service for an internal-only issue instead of fixing its actual boundary.
Explanation
Correct Answer
B:Path-based external HTTP routing is distinct from creating only internal service endpoints.
Incorrect Answers
A:This changes a neighboring Kubernetes feature, not the stated control. The case instead depends on the intended ingress boundary.
C:Changing an unrelated Pod setting leaves the stated cloud feature unresolved. It leaves the intended ingress boundary unaddressed.
D:This changes a neighboring Kubernetes feature, not the stated control. The specific requirement concerns the intended ingress boundary.
Question 250
A VM instance will run heavy scratch processing whose local data can be lost after termination. Which temporary storage option could be appropriate?
- Use supported Local SSD for ephemeral high-throughput scratch data while persisting required durable output elsewhere.
- Use Spot VMs with durable checkpoints, retry-aware processing and handling for preemption notices.
- Change the instance display name as a replacement for modifying the actual machine or disk.
- Create a snapshot only and assume it immediately changes the live VM capacity.
Explanation
Correct Answer
A:Local SSD is high-performance ephemeral storage and must not hold the only durable copy of critical data.
Incorrect Answers
B:Retries repeat operations; they do not correct missing permissions or invalid configuration. It leaves the actual compute scheduling or resource condition unaddressed.
C:Display names and labels identify resources; they do not enforce their runtime behavior. The specific requirement concerns the actual compute scheduling or resource condition.
D:Snapshots are recovery artifacts and do not directly reconfigure a running resource. The necessary distinction is the actual compute scheduling or resource condition.