cert
cert-1
cert-2

Pass Google Associate Cloud Engineer Exam in First Attempt Guaranteed!

Get 100% Latest Exam Questions, Accurate & Verified Answers to Pass the Actual Exam!
30 Days Free Updates, Instant Download!

cert-5
cert-6
Associate Cloud Engineer Exam - Verified By Experts
Associate Cloud Engineer Premium Bundle
$29.98

Associate Cloud Engineer Premium Bundle

$79.99
$109.97
  • Premium File 391 Questions & Answers. Last update: Nov 10, 2026
  • Training Course 234 Video Lectures
  • Study Guide 849 Pages
 
$109.97
$79.99
accept 42 downloads in last 7 days
block-screenshots
Purchase Google Associate Cloud Engineer Exam Training Products Individually
Google Associate Cloud Engineer Premium File
Premium File 391 Q&A
$76.99$69.99
Google Associate Cloud Engineer Video Training Course
Training Course 234 Lectures
$21.99 $19.99
Google Associate Cloud Engineer PDF Study Guide
Study Guide 849 Pages
$21.99 $19.99
Google Associate Cloud Engineer Practice Test 22

Question 210

An administrator needs a list of all Pods across namespaces but only checks the default namespace. What command scope should they use?

  1. Run kubectl get pods in the default namespace and assume that is the entire cluster.
  2. Run kubectl get nodes because nodes and Pods are the same Kubernetes objects.
  3. Change every Pod to namespace default before listing them.
  4. Use kubectl get pods --all-namespaces after verifying the active context and read permissions.

Explanation

Correct Answer

D:

The --all-namespaces flag includes namespaces within the current cluster for which the principal has the required access. Confirm the context first so results do not come from staging or another cluster.

Incorrect Answers

A:

The default namespace view omits Pods in all other namespaces, even when they share the cluster.

B:

Node inventory is not a list of running Pods across all namespaces.

C:

Moving workloads changes their namespace isolation and is unnecessary for inventory.

Question 211

A Pod is pending due to insufficient memory requests even though the cluster's other nodes look idle. What scheduling evidence matters?

  1. Read Pod scheduling events, requested memory, node allocatable capacity, constraints and cluster autoscaler diagnostics.
  2. Ignore Pod requests because the scheduler looks only at live CPU utilization averages.
  3. Increase the Pod container image size to create more memory on existing nodes.
  4. Change the Service type from ClusterIP to LoadBalancer to satisfy memory scheduling.

Explanation

Correct Answer

A:

A Pending Pod is scheduled against requested capacity and placement constraints, not just a dashboard of average use. Inspect kubectl describe pod events, available node resources, taints/affinity and autoscaler outcomes before resizing or changing requests.

Incorrect Answers

B:

The scheduler uses requests to reserve placement; low observed usage does not necessarily make a node eligible.

C:

Image size concerns download/storage and does not create node RAM capacity.

D:

Service exposure affects networking and cannot fix an unschedulable Pod’s memory request.

Question 212

A team manually patches a Deployment Pod, but the changed container disappears on the next reconciliation. What resource should be updated?

  1. Edit only the running Pod again; the Deployment will save the change permanently after enough attempts.
  2. Change the Deployment’s Pod template and apply a controlled rollout of the new specification.
  3. Delete the ReplicaSet so no controller can restore the original Pod configuration.
  4. Increase the Service port because service routing determines the Pod’s image and environment.

Explanation

Correct Answer

B:

The Deployment is the desired-state owner. Update its image or other Pod-template fields, apply the manifest, and observe rollout status; future replacements then use the approved configuration.

Incorrect Answers

A:

Pods controlled by a Deployment are replaced from the declared template, undoing one-off Pod edits.

C:

Deleting a controller’s child is not a durable way to change its desired template and can disrupt the workload.

D:

Service ports have no authority over the Pod template’s container configuration.

Question 213

A service's backend Pods have changed labels and the Kubernetes Service has no ready endpoints. What config should be fixed?

  1. Make the Service selector match the intended Pods’ labels, and confirm that ready endpoints are discovered.
  2. Create a new external IP address for the Service but leave its selectors pointing at nonexistent labels.
  3. Change only the namespace name without moving the Pods or updating labels and selectors.
  4. Increase cluster CPU quota because a selector mismatch is a resource scheduling failure.

Explanation

Correct Answer

A:

Kubernetes Services select backend Pods through label selectors. When labels change, the old selector matches none; align it with authorized Pods and verify readiness and EndpointSlices before testing traffic.

Incorrect Answers

B:

A public endpoint cannot discover backend Pods that the Service selector still excludes.

C:

A namespace label edit by itself does not repair the Service-to-Pod selector relationship.

D:

CPU quota is unrelated to discovering healthy endpoint targets when the Pods are already running.

Question 214

A GKE training Pod needs one NVIDIA GPU, but its manifest only requests CPU and memory. A GPU node pool already exists. What Kubernetes resource requirement must the Pod declare?

  1. Request the supported extended accelerator resource such as nvidia.com/gpu: 1 and schedule on compatible GPU nodes.
  2. Increase the CPU limit to 1000m; GKE interprets one CPU as one physical GPU device.
  3. Declare a Cloud Storage bucket as a volume; mounting it makes the GPU visible without requesting an accelerator.
  4. Add an ingress firewall rule for the container; network ingress grants hardware accelerator allocation.

Explanation

Correct Answer

A:

Request the supported extended accelerator resource such as nvidia.com/gpu: 1 and schedule on compatible GPU nodes. Kubernetes schedules GPU resources through supported extended resource requests and compatible nodes. CPU/memory requests alone do not reserve a GPU.

Incorrect Answers

B:

Increase the CPU limit to 1000m; GKE interprets one CPU as one physical GPU device. A CPU core is not an accelerator resource. Raising millicores does not cause the scheduler to reserve nvidia.com/gpu capacity.

C:

Declare a Cloud Storage bucket as a volume; mounting it makes the GPU visible without requesting an accelerator. A storage volume does not allocate accelerator devices. The Pod must request the GPU through its resource specification.

D:

Add an ingress firewall rule for the container; network ingress grants hardware accelerator allocation. Network policy controls connectivity, not GPU device scheduling or node driver availability. Opening a port does not allocate the accelerator.

Question 215

An ML team has a workload built on accelerator libraries that only support NVIDIA CUDA, but a cloud architect suggests Google Cloud TPUs without code changes. What selection criterion dominates?

  1. Select any TPU because CUDA binaries execute without changes on Google tensor processors.
  2. Move to CPUs only, because all accelerator kernels run with identical performance there.
  3. Choose the accelerator with the largest available quota, regardless of its software toolchain.
  4. Use a compatible NVIDIA GPU platform unless the application is ported to supported TPU libraries.

Explanation

Correct Answer

D:

CUDA-specific dependencies require compatible GPU hardware and driver/library support. A move to TPU would first need framework and model adaptation plus performance validation, not just an accelerator swap.

Incorrect Answers

A:

TPUs use a distinct software stack; a CUDA-only workload cannot be assumed to run unchanged.

B:

CPU fallback may require significant rework and does not preserve specialized CUDA execution.

C:

Capacity is irrelevant if the code cannot execute on the selected accelerator platform.

Question 216

Pending GKE Pods require more worker-node capacity, but their Standard-cluster node pool stays at its original fixed size. The pool is not configured to autoscale. What change should be made?

  1. Enable the Horizontal Pod Autoscaler; it will change the number of VM nodes even when node-pool autoscaling is disabled.
  2. Increase a Cloud Billing budget; the approved spending amount automatically resizes fixed node pools.
  3. Enable cluster/node-pool autoscaling with appropriate minimum and maximum nodes and verify quotas and scheduling constraints.
  4. Disable the GKE API so queued Pods can force their own node VM allocations.

Explanation

Correct Answer

C:

Enable cluster/node-pool autoscaling with appropriate minimum and maximum nodes and verify quotas and scheduling constraints. GKE cluster autoscaler can provision or remove nodes within configured pool limits when unschedulable Pods need capacity; quota and placement still constrain the result.

Incorrect Answers

A:

Enable the Horizontal Pod Autoscaler; it will change the number of VM nodes even when node-pool autoscaling is disabled. Horizontal Pod Autoscaler adjusts workload replica counts, not node-pool sizes in isolation. A fixed pool remains bounded unless node autoscaling or manual resize is configured.

B:

Increase a Cloud Billing budget; the approved spending amount automatically resizes fixed node pools. Budget configuration monitors cost but does not create VM nodes on behalf of an unschedulable workload.

D:

Disable the GKE API so queued Pods can force their own node VM allocations. Disabling the service API prevents management operations and cannot cause the scheduler to create missing worker nodes.

Question 217

A workload requires node labels and taints reserved for sensitive data and is scheduled to arbitrary general-purpose nodes. What scheduler controls are required?

  1. Use node selectors or affinity together with required tolerations, and isolate the authorized sensitive node pool.
  2. Set only a larger CPU request; the scheduler will infer data classification from CPU size.
  3. Add a toleration alone and assume it forces the Pod to run only on sensitive nodes.
  4. Use a Kubernetes Service type of LoadBalancer to pin the Pod to the secure node pool.

Explanation

Correct Answer

A:

Select eligible nodes with labels/affinity and tolerate their protection taints when appropriate. The placement rule must constrain where the workload can run; tolerations alone only permit admission to tainted nodes.

Incorrect Answers

B:

CPU requests affect capacity but do not express security placement or reserved node-pool identity.

C:

A toleration permits placement on a tainted node but does not require it, so general-purpose nodes may remain eligible.

D:

Service type controls traffic exposure, not Kubernetes node scheduling policy.

Question 218

A developer deploys a newer Cloud Run container but wants the current production revision to keep all traffic until validation. Which rollout step is correct?

  1. Deploy the new revision with all production traffic and rely on fast rollback if validation fails.
  2. Deploy a no-traffic revision, optionally tag it for restricted testing, and promote only after validation.
  3. Delete the existing production revision before the new container is verified.
  4. Set the new revision as latest and assume “latest” always receives zero traffic by default.

Explanation

Correct Answer

B:

Cloud Run can deploy a revision with zero allocated production traffic. A tagged URL can be used for authorized test requests; retain the stable route until checks pass and then shift traffic deliberately.

Incorrect Answers

A:

An immediate full cutover violates the requirement to keep production traffic on the stable revision during testing.

C:

Deleting the stable revision removes the safe serving path and reduces rollback options.

D:

The latest-revision routing setting can send traffic to a new revision unless explicit traffic allocation is configured.

Question 219

A Cloud Run service needs to split 10% of requests to an evaluated candidate and retain 90% on the stable revision. What mechanism fits?

  1. Create an additional Cloud Run service and assume traffic will split 50/50 through DNS.
  2. Set 10% of requests to a Cloud Storage bucket and send the remainder to production.
  3. Increase the candidate revision’s memory so it automatically receives exactly ten percent of calls.
  4. Configure Cloud Run traffic allocation: 10% to the candidate revision and 90% to stable.

Explanation

Correct Answer

D:

Cloud Run supports percentage-based routing among revisions of a service. Assign explicit percentages to tested revision IDs and monitor errors and latency during the rollout before increasing exposure.

Incorrect Answers

A:

A second service does not automatically implement a measured 10/90 traffic split.

B:

Object storage is not a Cloud Run revision traffic target.

C:

Resource settings do not determine a fixed share of inbound requests.

Question 220

Cloud Run revision B begins returning HTTP 5xx immediately after a release, while revision A remains healthy. The team wants to restore service rapidly without rebuilding a container image. What should it do first?

  1. Increase revision B's CPU allocation and continue sending it all traffic without reviewing the errors.
  2. Delete revision A so every request is forced onto the new failing revision.
  3. Change the Cloud Run service maximum instance limit; that automatically restores an older working revision.
  4. Shift the service traffic allocation back to the previously healthy revision A and investigate revision B off the serving path.

Explanation

Correct Answer

D:

Shift the service traffic allocation back to the previously healthy revision A and investigate revision B off the serving path. Cloud Run supports traffic migration among revisions. Moving production traffic to the healthy revision restores service without a new image build while the faulty revision is investigated.

Incorrect Answers

A:

Increase revision B's CPU allocation and continue sending it all traffic without reviewing the errors. A generic CPU increase is not justified by the given evidence; it keeps the known failing revision exposed to users and delays safe rollback.

B:

Delete revision A so every request is forced onto the new failing revision. Removing the healthy revision would eliminate the known-good rollback target and cannot resolve the newly introduced 5xx regression.

C:

Change the Cloud Run service maximum instance limit; that automatically restores an older working revision. Instance limits control autoscaling, not the code or configuration selected by the service's traffic split. Changing a limit does not roll back to A.

Associate Cloud Engineer Practice Tests