Question 200
A private API hosted in Cloud Run must require authenticated invocations. Which action is most appropriate?
- Set environment variables on the service revision
- Require Cloud Run Invoker IAM authorization rather than allowing public access
- Deploy a Cloud Run service referencing the container image
- Deploy a Cloud Run function with the corresponding event trigger
- Use the Cloud Run source deployment workflow
Explanation
Correct Answer
B:Authenticated invocation relies on appropriate identities and tokens. Merely hiding the URL does not create a security boundary.
Incorrect Answers
A:Its purpose is to configure revision environment variables. That distinction matters: the scenario calls for an action to protect a managed service from anonymous callers, not this separate configuration task.
C:While this can deploy an HTTP service from a container image, the present task requires an action to protect a managed service from anonymous callers. This operation does not supply the behavior described in the requirements.
D:The technical effect is to run an event-triggered function on object changes, not to protect a managed service from anonymous callers; That distinction matters here because the first operation serves respond to object creation events with function code, rather than the scenario's required result.
E:This approach can build and deploy Cloud Run from source. The requirement is to protect a managed service from anonymous callers, so it controls a different part of the workflow and leaves that result unaddressed.
Question 201
An existing Cloud Run service must process native Cloud Storage object-finalized events through managed Eventarc routing, without deploying a new function or creating an intermediate Pub/Sub notification pipeline. Which action is appropriate?
- Deploy a Cloud Run function with the corresponding event trigger
- Use Pub/Sub topics and subscriptions
- Use an Eventarc trigger targeting a compatible Cloud Run destination
- Configure an appropriate Eventarc Pub/Sub trigger for the Cloud Run service
Explanation
Correct Answer
C:Eventarc routes supported provider events to event-driven destinations. Event delivery requires compatible trigger filters and service identities.
Incorrect Answers
A:Select this when the goal is to run an event-triggered function on object changes. Here the decisive requirement is to invoke a handler after an object event, and that outcome is not achieved by this action.
B:This mechanism is intended to fan out asynchronous events to consumers. It does not implement the required action to invoke a handler after an object event, even though both are relevant to the same general service.
D:The tool can connect messaging events to a serverless destination, but it cannot be substituted for the capability needed to invoke a handler after an object event. Its operational effect is not equivalent.
Question 202
A decoupled producer must publish events while several subscribers process them independently. What is the best next step?
- Use Pub/Sub topics and subscriptions
- Use an Eventarc trigger targeting a compatible Cloud Run destination
- Configure an appropriate Eventarc Pub/Sub trigger for the Cloud Run service
- Use the Cloud Run source deployment workflow
- Deploy a Cloud Run function with the corresponding event trigger
Explanation
Correct Answer
A:Pub/Sub separates publishers from consumers through topics and subscriptions. Individual subscriptions maintain independent delivery state.
Incorrect Answers
B:This setting would route storage events to Cloud Run. That behavior does not establish the requested ability to decouple event production from processing; it addresses an adjacent concern instead.
C:Using this action would connect messaging events to a serverless destination, not decouple event production from processing. Its function is different from the specific change described in the scenario.
D:This alternative addresses cloud run application deployment by allowing a team to build and deploy Cloud Run from source; The incident, however, requires the different action to decouple event production from processing, so its central configuration remains unchanged.
E:This choice is useful to run an event-triggered function on object changes. The stated task is to decouple event production from processing, which requires a different mechanism or permission scope.
Question 203
An asynchronous application wants Cloud Run to receive Pub/Sub events through a supported managed trigger. Which configuration best meets the requirement?
- Deploy a Cloud Run function with the corresponding event trigger
- Use an Eventarc trigger targeting a compatible Cloud Run destination
- Deploy a Cloud Run service referencing the container image
- Configure an appropriate Eventarc Pub/Sub trigger for the Cloud Run service
- Use Pub/Sub topics and subscriptions
Explanation
Correct Answer
D:Eventarc integrates supported events with Cloud Run destinations. The trigger and destination service account must be configured for delivery.
Incorrect Answers
A:Administrators use it to run an event-triggered function on object changes, rather than to process Pub/Sub events in Cloud Run. Applying it would change the wrong component of this deployment.
B:This mechanism is intended to route storage events to Cloud Run. It does not implement the required action to process Pub/Sub events in Cloud Run, even though both are relevant to the same general service.
C:The technical effect is to deploy an HTTP service from a container image, not to process Pub/Sub events in Cloud Run; That distinction matters here because the first operation serves publish a managed container service, rather than the scenario's required result.
E:The result of this choice would be to fan out asynchronous events to consumers. The requested outcome, to process Pub/Sub events in Cloud Run, is handled at a different layer or through another operation.
Question 204
A Cloud Run service must start processing whenever an administrator creates a Compute Engine instance. The source event is an audit-logged API call, not a Cloud Storage object event. Which configuration should the engineer use?
- Configure an Eventarc Cloud Audit Logs trigger that matches the Compute Engine API method and targets the Cloud Run service
- Create a Pub/Sub subscription without a publisher and assign it to the Cloud Run service
- Set a Cloud Storage object-finalized trigger on the Compute Engine project
- Configure Cloud Scheduler to invoke the service at the end of each hour
Explanation
Correct Answer
A:Eventarc can route supported audit-log events by service and method to an authorized Cloud Run destination. The required audit-log generation and Eventarc delivery permissions must also be in place.
Incorrect Answers
B:A subscription needs messages from a publisher or event source. Simply creating a subscription does not turn Compute Engine API audit entries into Pub/Sub messages.
C:An object-finalized trigger responds to uploads into a bucket, not to a Compute Engine instance creation API call. Its event type does not match the required source.
D:Cloud Scheduler issues time-based requests and would not react to each specific instance-creation event. It cannot guarantee an invocation for every matching API call.
Question 205
A model-serving workload needs broadly programmable parallel compute accelerated by CUDA-compatible code. Which approach should be selected?
- Select a supported GPU-backed compute option
- Provision a compatible GPU node configuration and request GPU resources for the Pods
- Evaluate available Google Cloud TPUs and their compatible runtime
- Start with CPU-backed serverless or VM compute rather than allocating accelerators unnecessarily
Explanation
Correct Answer
A:GPUs accelerate many parallel numerical workloads, subject to machine and driver support. The actual model stack must be compatible with the accelerator.
Incorrect Answers
B:Its purpose is to schedule GPU workloads in Kubernetes. That distinction matters: the scenario calls for an action to use graphics processors for parallel inference, not this separate configuration task.
C:While this can run TPU-compatible tensor computation, the present task requires an action to use graphics processors for parallel inference. This operation does not supply the behavior described in the requirements.
D:This approach can avoid unnecessary accelerator allocation. The requirement is to use graphics processors for parallel inference, so it controls a different part of the workflow and leaves that result unaddressed.
Question 206
A training workload is written for TPU-supported ML frameworks and tensor computation. Which method addresses the task?
- Provision a compatible GPU node configuration and request GPU resources for the Pods
- Reference a supported image family when creating the VM
- Evaluate available Google Cloud TPUs and their compatible runtime
- Start with CPU-backed serverless or VM compute rather than allocating accelerators unnecessarily
- Select a supported GPU-backed compute option
Explanation
Correct Answer
C:TPUs are specialized accelerators for supported machine-learning tensor workloads. Framework, region, and topology constraints matter to deployment.
Incorrect Answers
A:Select this when the goal is to schedule GPU workloads in Kubernetes. Here the decisive requirement is to use specialized tensor accelerators, and that outcome is not achieved by this action.
B:The technical effect is to choose an image by family, not to use specialized tensor accelerators; That distinction matters here because the first operation serves deploy from a maintained image line, rather than the scenario's required result.
D:The tool can avoid unnecessary accelerator allocation, but it cannot be substituted for the capability needed to use specialized tensor accelerators. Its operational effect is not equivalent.
E:This mechanism is intended to accelerate supported GPU workloads. It does not implement the required action to use specialized tensor accelerators, even though both are relevant to the same general service.
Question 207
An inference service runs well on ordinary CPUs and has modest traffic. What is the best next step?
- Create a ClusterIP Kubernetes Service
- Start with CPU-backed serverless or VM compute rather than allocating accelerators unnecessarily
- Evaluate available Google Cloud TPUs and their compatible runtime
- Provision a compatible GPU node configuration and request GPU resources for the Pods
- Select a supported GPU-backed compute option
Explanation
Correct Answer
B:CPU execution can be cost-effective when latency and throughput goals are met. Accelerators add device-specific setup and potential expense.
Incorrect Answers
A:Choosing this would provide cluster-internal service discovery, an appropriate action for gke service exposure operations; It cannot take the place of the separate step needed to serve a lightweight model using CPU capacity in this specific situation.
C:Using this action would run TPU-compatible tensor computation, not serve a lightweight model using CPU capacity. Its function is different from the specific change described in the scenario.
D:Its purpose is to schedule GPU workloads in Kubernetes. That distinction matters: the scenario calls for an action to serve a lightweight model using CPU capacity, not this separate configuration task.
E:This setting would accelerate supported GPU workloads. That behavior does not establish the requested ability to serve a lightweight model using CPU capacity; it addresses an adjacent concern instead.
Question 208
A containerized training pipeline requires a GPU device visible to its GKE workload. Which configuration best meets the requirement?
- Create the instance with only an internal IP on the chosen VPC subnet
- Evaluate available Google Cloud TPUs and their compatible runtime
- Start with CPU-backed serverless or VM compute rather than allocating accelerators unnecessarily
- Select a supported GPU-backed compute option
- Provision a compatible GPU node configuration and request GPU resources for the Pods
Explanation
Correct Answer
E:GKE can schedule GPU workloads on eligible GPU-equipped nodes when device support and resource requests are configured. Pod CPU limits alone do not allocate accelerators.
Incorrect Answers
A:This proposal can create a private-only VM interface, but the failure or goal described here calls for a way to expose GPU capacity to GKE Pods; It changes the wrong control or scope for the requested outcome.
B:Administrators use it to run TPU-compatible tensor computation, rather than to expose GPU capacity to GKE Pods. Applying it would change the wrong component of this deployment.
C:The result of this choice would be to avoid unnecessary accelerator allocation. The requested outcome, to expose GPU capacity to GKE Pods, is handled at a different layer or through another operation.
D:This mechanism is intended to accelerate supported GPU workloads. It does not implement the required action to expose GPU capacity to GKE Pods, even though both are relevant to the same general service.
Question 209
A VM needs traffic to reach a private database and must not have a public NIC address. Which choice provides the required behavior?
- Create the instance with only an internal IP on the chosen VPC subnet
- Use a targeted VPC ingress firewall rule for the appropriate source ranges and port
- Configure the boot disk auto-delete setting appropriately
- Reference a supported image family when creating the VM
Explanation
Correct Answer
A:Compute Engine instances can omit external addresses while retaining private VPC connectivity. Remote administration then needs IAP, VPN, or another authorized path.
Incorrect Answers
B:Using this action would permit selected inbound load-balancer traffic, not limit a VM to internal networking. Its function is different from the specific change described in the scenario.
C:This setting would retain a disk when a VM is deleted. That behavior does not establish the requested ability to limit a VM to internal networking; it addresses an adjacent concern instead.
D:This choice is useful to choose an image by family. The stated task is to limit a VM to internal networking, which requires a different mechanism or permission scope.
Question 210
A web VM should allow traffic only from a designated HTTP load-balancing source range. Which approach should be selected?
- Use a targeted VPC ingress firewall rule for the appropriate source ranges and port
- Configure the boot disk auto-delete setting appropriately
- Create the instance with only an internal IP on the chosen VPC subnet
- Deploy a Cloud Run service referencing the container image
- Reference a supported image family when creating the VM
Explanation
Correct Answer
A:VPC firewall rules filter packets based on source, target and protocol attributes. Network placement alone does not open an inbound port.
Incorrect Answers
B:Administrators use it to retain a disk when a VM is deleted, rather than to allow only authorized web ingress to VMs. Applying it would change the wrong component of this deployment.
C:Its purpose is to create a private-only VM interface. That distinction matters: the scenario calls for an action to allow only authorized web ingress to VMs, not this separate configuration task.
D:Choosing this would deploy an HTTP service from a container image, an appropriate action for cloud run application deployment operations; It cannot take the place of the separate step needed to allow only authorized web ingress to VMs in this specific situation.
E:The result of this choice would be to choose an image by family. The requested outcome, to allow only authorized web ingress to VMs, is handled at a different layer or through another operation.