Google Associate Cloud Engineer Practice Questions with Explanations
Free Google Associate Cloud Engineer practice questions. 94 of them, each with the correct answer, a full explanation, and the reason every other option is wrong.
Google Associate Cloud Engineer practice questions
Question 1
Every employee of your company has a Google account. Your operational team needs to manage a large number of instances on Compute Engine. Each member of this team needs only administrative access to the servers. Your security team wants to ensure that the deployment of credentials is operationally efficient and must be able to determine who accessed a given instance. What should you do?
- A. Generate a new SSH key pair. Give the private key to each member of your team. Configure the public key in the metadata of each instance.
- B. Ask each member of the team to generate a new SSH key pair and to send you their public key. Use a configuration management tool to deploy those keys on each instance.
- C. Ask each member of the team to generate a new SSH key pair and to add the public key to their Google account. Grant the "compute.osAdminLogin" role to the Google group corresponding to this team.
- D. Generate a new SSH key pair. Give the private key to each member of your team. Configure the public key as a project-wide public SSH key in your Cloud Platform project and allow project-wide public SSH keys on each instance.
Show answer and explanation ▾
Correct answer: C
OS Login ties SSH access on Compute Engine to each user's own Google identity, so every member keeps a personal key associated with their Google account instead of shared keys copied into instance metadata. Granting roles/compute.osAdminLogin to the team's Google group gives administrative (sudo) access on the instances, and new members only need to join the group, which keeps credential deployment operationally efficient. Because each login is made with an individual identity, the security team can determine exactly who accessed a given instance through the Linux account name and the audit logs.
Why the other options are wrong:
- A. A single shared private key handed to everyone makes it impossible to tell which person accessed an instance, and sharing private keys is a poor security practice.
- B. Pushing individual keys with a configuration management tool adds ongoing operational overhead for every new member and every new instance, so it is not efficient.
- D. A project-wide key with a shared private key still prevents attributing access to an individual and spreads one credential across the whole team.
Question 2
You need to create a custom VPC with a single subnet. The subnet's range must be as large as possible. Which range should you use?
- A. 0.0.0.0/0
- B. 10.0.0.0/8
- C. 172.16.0.0/12
- D. 192.168.0.0/16
Show answer and explanation ▾
Correct answer: B
A subnet's primary range in a custom VPC must be a valid subnet range, such as an RFC 1918 private block, and of the RFC 1918 blocks the largest is 10.0.0.0/8, which provides roughly 16.7 million addresses. Because the stem asks for the largest possible range for a single subnet, the /8 prefix is the right choice: a smaller prefix number means a larger block, and 10.0.0.0/8 is larger than both 172.16.0.0/12 and 192.168.0.0/16 while still being an accepted subnet range in Google Cloud.
Why the other options are wrong:
- A. 0.0.0.0/0 represents every IPv4 address and is used for routes and firewall rules, not as a subnet range, so it cannot be assigned to a subnet.
- C. 172.16.0.0/12 is a valid private range, but at about one million addresses it is smaller than the 10.0.0.0/8 block.
- D. 192.168.0.0/16 is the smallest of the private blocks, offering only about 65,000 addresses, so it does not meet the largest range requirement.
Question 3
You want to select and configure a cost-effective solution for relational data on Google Cloud Platform. You are working with a small set of operational data in one geographic location. You need to support point-in-time recovery. What should you do?
- A. Select Cloud SQL (MySQL). Verify that the enable binary logging option is selected.
- B. Select Cloud SQL (MySQL). Select the create failover replicas option.
- C. Select Cloud Spanner. Set up your instance with 2 nodes.
- D. Select Cloud Spanner. Set up your instance as multi-regional.
Show answer and explanation ▾
Correct answer: A
The data is relational, small, operational and located in one region, which matches Cloud SQL far better than Cloud Spanner on cost. Point-in-time recovery for Cloud SQL for MySQL relies on binary logging: the binary logs record every change after a backup, so the instance can be restored to a specific moment. Confirming that binary logging is enabled when creating the instance therefore satisfies the recovery requirement while keeping the solution inexpensive.
Why the other options are wrong:
- B. A failover replica improves availability during a zonal outage, but it does not provide point-in-time recovery of the data.
- C. Cloud Spanner is a globally scalable database whose cost is far beyond what a small, single-location operational dataset requires.
- D. A multi-regional Spanner instance is even more expensive and contradicts the requirement that the data stays in one geographic location.
Question 4
You want to configure autohealing for network load balancing for a group of Compute Engine instances that run in multiple zones, using the fewest possible steps.
You need to configure re-creation of VMs if they are unresponsive after 3 attempts of 10 seconds each. What should you do?
- A. Create an HTTP load balancer with a backend configuration that references an existing instance group. Set the health check to healthy (HTTP)
- B. Create an HTTP load balancer with a backend configuration that references an existing instance group. Define a balancing mode and set the maximum RPS to 10.
- C. Create a managed instance group. Set the Autohealing health check to healthy (HTTP)
- D. Create a managed instance group. Verify that the autoscaling setting is on.
Show answer and explanation ▾
Correct answer: C
Re-creating unresponsive VMs is the job of autohealing, which is a feature of managed instance groups. When you create a regional managed instance group you attach an autohealing health check and set its check interval (10 seconds) and unhealthy threshold (3 consecutive failures). When an instance fails the check, the group automatically re-creates it from the instance template. This meets the requirement in the fewest steps, because the health check is configured directly on the group without building a separate load balancer.
Why the other options are wrong:
- A. A load balancer health check only stops sending traffic to unhealthy backends; it does not re-create the VMs, so autohealing is not achieved.
- B. A balancing mode with a maximum RPS controls how traffic is distributed and has nothing to do with detecting and replacing unresponsive VMs.
- D. Autoscaling adds or removes instances based on load, but it does not detect unresponsive VMs or re-create them.
Question 5
You are using multiple configurations for gcloud. You want to review the configured Kubernetes Engine cluster of an inactive configuration using the fewest possible steps. What should you do?
- A. Use gcloud config configurations describe to review the output.
- B. Use gcloud config configurations activate and gcloud config list to review the output.
- C. Use kubectl config get-contexts to review the output.
- D. Use kubectl config use-context and kubectl config view to review the output.
Show answer and explanation ▾
Correct answer: A
The gcloud config configurations describe command prints all the properties stored in a named configuration, including the container/cluster property, whether or not that configuration is currently active. Running it against the inactive configuration shows the configured Kubernetes Engine cluster in a single step, without switching configurations and then switching back, which satisfies the fewest possible steps requirement.
Why the other options are wrong:
- B. Activating the configuration and then listing it works but takes extra steps and changes the active configuration, which then has to be switched back.
- C. kubectl contexts come from the kubeconfig file, not from gcloud configurations, so they do not show what an inactive gcloud configuration has set.
- D. Switching kubectl contexts and viewing kubeconfig changes the current context and still does not read the gcloud configuration's cluster property.
Question 6
Your company uses Cloud Storage to store application backup files for disaster recovery purposes. You want to follow Google's recommended practices. Which storage option should you use?
- A. Multi-Regional Storage
- B. Regional Storage
- C. Nearline Storage
- D. Coldline Storage
Show answer and explanation ▾
Correct answer: D
Disaster recovery backups are written regularly but read only rarely, typically less than once a year when a recovery is needed. Coldline Storage is designed for data accessed less than once a quarter and has a much lower storage price than Standard or Nearline, which makes it the Google-recommended choice among these options for backup and disaster recovery data. Retrieval fees are acceptable because restores are infrequent events.
Why the other options are wrong:
- A. Multi-Regional Storage is optimized for frequently accessed content served worldwide, so it costs far more than needed for rarely read backups.
- B. Regional Storage targets frequently accessed data near compute resources and is priced higher than necessary for backup files.
- C. Nearline suits data read about once a month; backups that are rarely restored are cheaper in Coldline.
Question 7
Several employees at your company have been creating projects with Cloud Platform and paying for it with their personal credit cards, which the company reimburses. The company wants to centralize all these projects under a single, new billing account. What should you do?
- A. Contact cloud-billing@google.com with your bank account details and request a corporate billing account for your company.
- B. Create a ticket with Google Support and wait for their call to share your credit card details over the phone.
- C. In the Google Cloud Platform Console, go to the Resource Manager and move all projects to the root Organization.
- D. In the Google Cloud Platform Console, create a new billing account and set up a payment method.
Show answer and explanation ▾
Correct answer: D
Centralizing costs starts with a new Cloud Billing account, which you create yourself in the Google Cloud Console under Billing by adding a company payment method. Once the account exists, each project can be linked to it in place of the personal credit card accounts, so all spending flows to one corporate account. No contact with Google is required, and the process is entirely self-service.
Why the other options are wrong:
- A. Billing accounts are created self-service in the Console; emailing bank details to Google is not the process and is not a safe practice.
- B. Sharing card details by phone through a support ticket is not how billing accounts are set up and adds unnecessary delay.
- C. Moving projects into the organization changes the resource hierarchy, but it does not change which billing account pays for them.
Question 8
You have an application that looks for its licensing server on the IP 10.0.3.21. You need to deploy the licensing server on Compute Engine. You do not want to change the configuration of the application and want the application to be able to reach the licensing server. What should you do?
- A. Reserve the IP 10.0.3.21 as a static internal IP address using gcloud and assign it to the licensing server.
- B. Reserve the IP 10.0.3.21 as a static public IP address using gcloud and assign it to the licensing server.
- C. Use the IP 10.0.3.21 as a custom ephemeral IP address and assign it to the licensing server.
- D. Start the licensing server with an automatic ephemeral IP address, and then promote it to a static internal IP address.
Show answer and explanation ▾
Correct answer: A
The application expects to find the licensing server at 10.0.3.21, a private address, so the server needs that exact internal IP permanently. Reserving 10.0.3.21 as a static internal IP address in the subnet with gcloud compute addresses create and then assigning it to the licensing server's network interface guarantees the address is held for the server and does not change after restarts. The application keeps its existing configuration and can reach the server over the VPC.
Why the other options are wrong:
- B. 10.0.3.21 is a private RFC 1918 address and cannot be reserved as an external IP; public addresses are allocated from Google's pools.
- C. An ephemeral address is not reserved and can be lost when the instance is deleted or re-created, so the application could lose the server.
- D. An automatically assigned ephemeral address would be whatever the subnet hands out, almost certainly not 10.0.3.21, so promoting it does not help.
Question 9
You are deploying an application to App Engine. You want the number of instances to scale based on request rate. You need at least 3 unoccupied instances at all times. Which scaling type should you use?
- A. Manual Scaling with 3 instances.
- B. Basic Scaling with min_instances set to 3.
- C. Basic Scaling with max_instances set to 3.
- D. Automatic Scaling with min_idle_instances set to 3.
Show answer and explanation ▾
Correct answer: D
Scaling on request rate is the behavior of automatic scaling, which creates instances based on request rate, response latency and other application metrics. Its min_idle_instances setting keeps a specified number of warm instances running and ready to absorb traffic spikes, beyond those actively serving requests. Setting min_idle_instances to 3 meets the requirement for at least 3 unoccupied instances at all times while the rest of the fleet scales with demand.
Why the other options are wrong:
- A. Manual scaling runs a fixed number of instances and does not scale with request rate.
- B. Basic scaling creates instances only when requests arrive and shuts them down when idle; it has no min_instances setting for keeping idle instances available.
- C. A max_instances cap limits growth to 3 instances and does nothing to keep 3 instances idle and ready.
Question 10
You have a development project with appropriate IAM roles defined. You are creating a production project and want to have the same IAM roles on the new project, using the fewest possible steps. What should you do?
- A. Use gcloud iam roles copy and specify the production project as the destination project.
- B. Use gcloud iam roles copy and specify your organization as the destination organization.
- C. In the Google Cloud Platform Console, use the 'create role from role' functionality.
- D. In the Google Cloud Platform Console, use the 'create role' functionality and select all applicable permissions.
Show answer and explanation ▾
Correct answer: A
Custom roles defined in a project exist only in that project, so they have to be recreated in the production project. The gcloud iam roles copy command copies an existing role into a destination project (using the destination project flag) with all its permissions in a single command. Specifying the production project as the destination recreates the same roles there with the fewest steps and without rebuilding permission lists by hand.
Why the other options are wrong:
- B. Copying the roles to the organization makes them organization-level roles, which is broader than the requirement to place them in the production project.
- C. Creating a role from a role in the Console works one role at a time through the UI and needs more manual steps than a gcloud copy.
- D. Building each role by selecting permissions manually is the slowest approach and risks mismatches with the development roles.
Getting these right? Take the first 187 with you.
The free PDF is the first half of the full Google Associate Cloud Engineer pack: 187 questions with every answer explained, as one PDF in your inbox.
Get the free PDFWhat people say about the full Google Associate Cloud Engineer pack 4.8 from 9
Grateful I found this before sitting ACE. It genuinely helped.
Accurate and up to date for ACE. Recommended for anyone preparing.
Passed ace first go. The questions lined up nicely with the actual exam.
Question 11
You have a Google Cloud Platform account with access to both production and development projects. You need to create an automated process to list all compute instances in development and production projects on a daily basis. What should you do?
- A. Create two configurations using gcloud config. Write a script that sets configurations as active, individually. For each configuration, use gcloud compute instances list to get a list of compute resources.
- B. Create two configurations using gsutil config. Write a script that sets configurations as active, individually. For each configuration, use gsutil compute instances list to get a list of compute resources.
- C. Go to Cloud Shell and export this information to Cloud Storage on a daily basis.
- D. Go to GCP Console and export this information to Cloud SQL on a daily basis.
Show answer and explanation ▾
Correct answer: A
gcloud named configurations store settings such as account and project, and you can create one per environment with gcloud config configurations create. A scheduled script can activate each configuration in turn (gcloud config configurations activate) and run gcloud compute instances list, producing a daily inventory of instances in both the development and production projects. This is simple to automate, uses the standard CLI, and needs no extra services.
Why the other options are wrong:
- B. gsutil is the Cloud Storage tool; it has no compute instances list command and its config does not manage project configurations.
- C. Cloud Shell is an interactive session, not a scheduler, and this gives no mechanism to list instances across both projects.
- D. The console has no daily export of instance lists to Cloud SQL, so this is not an automated process.
Question 12
An employee was terminated, but their access to Google Cloud was not removed until 2 weeks later. You need to find out if this employee accessed any sensitive customer information after their termination. What should you do?
- A. View System Event Logs in Cloud Logging. Search for the user's email as the principal.
- B. View System Event Logs in Cloud Logging. Search for the service account associated with the user.
- C. View Data Access audit logs in Cloud Logging. Search for the user's email as the principal.
- D. View the Admin Activity log in Cloud Logging. Search for the service account associated with the user.
Show answer and explanation ▾
Correct answer: C
Reading customer information is a data access operation, and Cloud Audit Logs records such reads, along with reads of resource metadata, in Data Access audit logs. These logs record the principal email of the caller, so searching the Data Access logs in Cloud Logging for the former employee's email shows whether and when that user read sensitive data after the termination date. Data Access logging must be enabled for the relevant services (it is always on for BigQuery) for these entries to exist.
Why the other options are wrong:
- A. System Event audit logs record actions taken by Google systems, such as live migration, not by users, so they will not show the employee's data reads.
- B. System Event logs record Google-initiated actions, and searching for a service account would miss actions the employee performed with their own user identity.
- D. Admin Activity logs record configuration and metadata changes, not data reads, and searching for a service account would not find the user's own activity.
Question 13
You are using Container Registry to centrally store your company's container images in a separate project. In another project, you want to create a Google Kubernetes Engine (GKE) cluster. You want to ensure that Kubernetes can download images from Container Registry. What should you do?
- A. In the project where the images are stored, grant the Storage Object Viewer IAM role to the service account used by the Kubernetes nodes.
- B. When you create the GKE cluster, choose the Allow full access to all Cloud APIs option under 'Access scopes'.
- C. Create a service account, and give it access to Cloud Storage. Create a P12 key for this service account and use it as an imagePullSecrets in Kubernetes.
- D. Configure the ACLs on each image in Cloud Storage to give read-only access to the default Compute Engine service account.
Show answer and explanation ▾
Correct answer: A
Container Registry stores images in Cloud Storage buckets in the project that hosts the registry. GKE nodes pull images using the service account attached to the nodes, so that service account needs read access to those buckets. Granting it the Storage Object Viewer role (roles/storage.objectViewer) in the project where the images are stored lets the nodes in the other project pull images securely, following least privilege.
Why the other options are wrong:
- B. Access scopes only limit what the node service account may do; they do not grant IAM permissions in the project that stores the images.
- C. P12 keys are a legacy format, and managing keys in imagePullSecrets adds unnecessary risk when node IAM access works natively.
- D. Setting ACLs object by object is hard to maintain and is not the recommended IAM-based approach for registry access.
Question 14
You have an application running in Google Kubernetes Engine (GKE) with cluster autoscaling enabled. The application exposes a TCP endpoint. There are several replicas of this application. You have a Compute Engine instance in the same region, but in another Virtual Private Cloud (VPC), called gce-network, that has no overlapping IP ranges with the first VPC. This instance needs to connect to the application on GKE. You want to minimize effort. What should you do?
- A. 1. In GKE, create a Service of type LoadBalancer that uses the application's Pods as backend. 2. Set the service's externalTrafficPolicy to Cluster. 3. Configure the Compute Engine instance to use the address of the load balancer that has been created.
- B. 1. In GKE, create a Service of type NodePort that uses the application's Pods as backend. 2. Create a Compute Engine instance called proxy with 2 network interfaces, one in each VPC. 3. Use iptables on this instance to forward traffic from gce-network to the GKE nodes. 4. Configure the Compute Engine instance to use the address of proxy in gce-network as endpoint.
- C. 1. In GKE, create a Service of type LoadBalancer that uses the application's Pods as backend. 2. Add an annotation to this service: cloud.google.com/load-balancer-type: Internal 3. Peer the two VPCs together. 4. Configure the Compute Engine instance to use the address of the load balancer that has been created.
- D. 1. In GKE, create a Service of type LoadBalancer that uses the application's Pods as backend. 2. Add a Cloud Armor Security Policy to the load balancer that whitelists the internal IPs of the MIG's instances. 3. Configure the Compute Engine instance to use the address of the load balancer that has been created.
Show answer and explanation ▾
Correct answer: C
The client VM is in a different VPC with no overlapping ranges, and the application has several replicas behind cluster autoscaling, so a stable load-balanced private endpoint is needed. A Service of type LoadBalancer with the annotation cloud.google.com/load-balancer-type: Internal provisions an internal passthrough load balancer in the cluster's VPC. Peering the two VPCs makes that internal address reachable from gce-network, since the instance is in the same region. This keeps traffic private and uses managed components with little effort.
Why the other options are wrong:
- A. This creates an external load balancer that exposes the application on a public IP, which is unnecessary for VPC-to-VPC access.
- B. A dual-NIC proxy with iptables is manual, a single point of failure and must be maintained, so it does not minimize effort.
- D. This still exposes a public external endpoint and adds a security policy to maintain, when an internal load balancer reached over VPC peering keeps the traffic private.
Question 15
Your company is moving its entire workload to Compute Engine. Some servers should be accessible through the Internet, and other servers should only be accessible over the internal network. All servers need to be able to talk to each other over specific ports and protocols. The current on-premises network relies on a demilitarized zone (DMZ) for the public servers and a Local Area Network (LAN) for the private servers. You need to design the networking infrastructure on Google Cloud to match these requirements. What should you do?
- A. 1. Create a single VPC with a subnet for the DMZ and a subnet for the LAN. 2. Set up firewall rules to open up relevant traffic between the DMZ and the LAN subnets, and another firewall rule to allow public ingress traffic for the DMZ.
- B. 1. Create a single VPC with a subnet for the DMZ and a subnet for the LAN. 2. Set up firewall rules to open up relevant traffic between the DMZ and the LAN subnets, and another firewall rule to allow public egress traffic for the DMZ.
- C. 1. Create a VPC with a subnet for the DMZ and another VPC with a subnet for the LAN. 2. Set up firewall rules to open up relevant traffic between the DMZ and the LAN subnets, and another firewall rule to allow public ingress traffic for the DMZ.
- D. 1. Create a VPC with a subnet for the DMZ and another VPC with a subnet for the LAN. 2. Set up firewall rules to open up relevant traffic between the DMZ and the LAN subnets, and another firewall rule to allow public egress traffic for the DMZ.
Show answer and explanation ▾
Correct answer: A
All servers must talk to each other on specific ports, some must be reachable from the internet, and the rest only internally. A single VPC with one subnet for the DMZ and one for the LAN mirrors the on-premises design, and because subnets in the same VPC share internal routing, VM-to-VM traffic needs only firewall rules allowing the required ports and protocols between the two subnet ranges. An ingress firewall rule from 0.0.0.0/0 to the DMZ servers then permits public access, while LAN servers remain unreachable from outside.
Why the other options are wrong:
- B. Egress rules control outbound traffic, and egress is already allowed by default; an egress rule does not let internet clients reach the DMZ servers.
- C. Separate VPCs cannot route to each other without peering or VPN, so firewall rules alone would not let the DMZ and LAN servers communicate.
- D. Separate VPCs need peering or VPN to exchange traffic, and an egress rule does not permit inbound public access to the DMZ.
Question 16
Your team maintains the infrastructure for your organization. The current infrastructure requires changes. You need to share your proposed changes with the rest of the team. You want to follow Google's recommended best practices. What should you do?
- A. Use Deployment Manager templates to describe the proposed changes and store them in a Cloud Storage bucket.
- B. Use Deployment Manager templates to describe the proposed changes and store them in Cloud Source Repositories.
- C. Apply the changes in a development environment, run gcloud compute instances list, and then save the output in a shared Storage bucket.
- D. Apply the changes in a development environment, run gcloud compute instances list, and then save the output in Cloud Source Repositories.
Show answer and explanation ▾
Correct answer: B
Google recommends managing infrastructure as code and keeping that code under version control. Describing the proposed changes in Deployment Manager templates makes them declarative and reviewable, and storing the templates in Cloud Source Repositories, a private Git repository, gives the team change history, diffs and a review workflow before anything is applied. This lets colleagues see exactly what will change and roll back to earlier versions if needed.
Why the other options are wrong:
- A. A Cloud Storage bucket stores files but lacks Git features such as diffs, branches and review, so it is a weaker way to collaborate on changes.
- C. Applying changes before review defeats the purpose of proposing them, and an instance list shows little about the actual configuration changes.
- D. Saving command output in a repository still means applying changes before review, and the output does not describe the infrastructure as code.
Question 17
Your organization needs to grant users access to query datasets in BigQuery but prevent them from accidentally deleting the datasets. You want a solution that follows Google-recommended practices. What should you do?
- A. Add users to roles/bigquery.user role only, instead of roles/bigquery.dataOwner.
- B. Add users to roles/bigquery.dataEditor role only, instead of roles/bigquery.dataOwner.
- C. Create a custom role by removing delete permissions, and add users to that role only.
- D. Create a custom role by removing delete permissions. Add users to the group, and then add the group to the custom role.
Show answer and explanation ▾
Correct answer: D
Users must be able to query data while lacking the permission to delete datasets, so a custom role built from the needed BigQuery permissions without bigquery.datasets.delete grants exactly that access. Google recommends granting roles to groups rather than to individual users: membership changes are made once in the group, the IAM policy stays small and readable, and access is easier to audit. Creating the custom role, placing the users in a group and binding the group to the role meets the functional requirement and follows recommended practice.
Why the other options are wrong:
- A. BigQuery User lets users run jobs and create their own datasets, and it grants ownership-level control over datasets they create, so deletions remain possible. Granting it to individuals also skips the recommended group-based approach.
- B. BigQuery Data Editor allows modifying and deleting tables in a dataset, so it does not prevent accidental data loss, and it is granted to individuals instead of a group.
- C. The custom role removes delete permissions correctly, but binding it to each user individually is harder to manage and audit than granting it to a group, which is the recommended practice.
Question 18
You have developed a containerized web application that will serve internal colleagues during business hours. You want to ensure that no costs are incurred outside of the hours the application is used. You have just created a new Google Cloud project and want to deploy the application. What should you do?
- A. Deploy the container on Cloud Run for Anthos, and set the minimum number of instances to zero.
- B. Deploy the container on Cloud Run (fully managed), and set the minimum number of instances to zero.
- C. Deploy the container on App Engine flexible environment with autoscaling, and set the value min_instances to zero in the app.yaml.
- D. Deploy the container on App Engine flexible environment with manual scaling, and set the value instances to zero in the app.yaml.
Show answer and explanation ▾
Correct answer: B
The requirement is that no costs are incurred outside business hours for a containerized web application in a new project. Cloud Run (fully managed) runs containers on demand and, with the minimum number of instances set to zero, scales down to zero when there are no requests. With request-based billing you pay only while requests are being handled, so idle evenings and weekends cost nothing. There is no cluster or underlying infrastructure to provision or pay for.
Why the other options are wrong:
- A. Cloud Run for Anthos runs on a GKE cluster whose nodes keep incurring costs even when the service scales to zero.
- C. App Engine flexible environment requires at least one instance running, so it cannot scale to zero and costs accrue around the clock.
- D. App Engine flexible environment with manual scaling needs at least one instance, so it still incurs costs outside business hours.
Question 19
You want to configure a solution for archiving data in a Cloud Storage bucket. The solution must be cost-effective. Data with multiple versions should be archived after 30 days. Previous versions are accessed once a month for reporting. This archive data is also occasionally updated at month-end. What should you do?
- A. Add a bucket lifecycle rule that archives data with newer versions after 30 days to Coldline Storage.
- B. Add a bucket lifecycle rule that archives data with newer versions after 30 days to Nearline Storage.
- C. Add a bucket lifecycle rule that archives data from regional storage after 30 days to Coldline Storage.
- D. Add a bucket lifecycle rule that archives data from regional storage after 30 days to Nearline Storage.
Show answer and explanation ▾
Correct answer: B
The data to archive is noncurrent versions of objects (those with newer versions) after 30 days, so the lifecycle rule should target objects that have newer versions. Previous versions are read about once a month and occasionally updated at month-end, which fits Nearline Storage: it is designed for data accessed roughly once a month or less and has a 30-day minimum storage duration. That access pattern keeps retrieval and early deletion charges low while reducing storage cost.
Why the other options are wrong:
- A. Coldline suits data accessed about once a quarter; monthly reads and updates would add higher retrieval and early deletion charges.
- C. This rule targets storage class rather than older versions, and Coldline is costlier for monthly access.
- D. Moving all regional data after 30 days ignores the versioning requirement and would also archive current versions.
Question 20
You need to host an application on a Compute Engine instance in a project shared with other teams. You want to prevent the other teams from accidentally causing downtime on that application. Which feature should you use?
- A. Use a Shielded VM.
- B. Use a Preemptible VM.
- C. Use a sole-tenant node.
- D. Enable deletion protection on the instance.
Show answer and explanation ▾
Correct answer: D
The risk in a shared project is that another team member with sufficient permissions deletes the instance by mistake. Deletion protection is an instance property that blocks any delete request until someone explicitly turns the setting off, so an accidental gcloud command or console click fails instead of taking the application down. It costs nothing, needs no change to the machine type or hosting model, and can be set when the instance is created or added later to a running instance, which makes it the direct safeguard against accidental downtime caused by deletion.
Why the other options are wrong:
- A. Shielded VM protects the boot process against rootkits and tampering with secure boot and integrity monitoring. It does nothing to stop another user from deleting or stopping the instance.
- B. A preemptible VM can be stopped by Compute Engine at any time and lasts at most 24 hours, which increases downtime rather than preventing it.
- C. A sole-tenant node isolates VMs on dedicated physical hardware for compliance or licensing. It adds cost and does not stop other project members from deleting the instance.
Question 21
You have a website hosted on App Engine standard environment. You want 1% of your users to see a new test version of the website. You want to minimize complexity. What should you do?
- A. Deploy the new version in the same application and use the --migrate option.
- B. Deploy the new version in the same application and use the --splits option to give a weight of 99 to the current version and a weight of 1 to the new version.
- C. Create a new App Engine application in the same project. Deploy the new version in that application. Use the App Engine library to proxy 1% of the requests to the new version.
- D. Create a new App Engine application in the same project. Deploy the new version in that application. Configure your network load balancer to send 1% of the traffic to that new application.
Show answer and explanation ▾
Correct answer: B
App Engine supports multiple versions of a service within one application and can split incoming traffic between them by weight. After deploying the new version, using the --splits option (for example with gcloud app services set-traffic) to assign 99 to the current version and 1 to the new version routes about 1% of users to the test version. Everything stays in one application with no proxies or load balancer configuration, which meets the requirement to minimize complexity.
Why the other options are wrong:
- A. The --migrate option moves all traffic to the new version rather than sending only 1% of users to it.
- C. A project can hold only one App Engine application, and proxying requests through custom code adds unnecessary complexity.
- D. A project supports only one App Engine application, and a network load balancer is not used to split App Engine traffic.
Question 22
You are running multiple VPC-native Google Kubernetes Engine clusters in the same subnet. The IPs available for the nodes are exhausted, and you want to ensure that the clusters can grow in nodes when needed. What should you do?
- A. Create a new subnet in the same region as the subnet being used.
- B. Add an alias IP range to the subnet used by the GKE clusters.
- C. Create a new VPC, and set up VPC peering with the existing VPC.
- D. Expand the CIDR range of the relevant subnet for the cluster.
Show answer and explanation ▾
Correct answer: D
In VPC-native GKE clusters, node IP addresses come from the subnet's primary IPv4 range, while Pods and Services use secondary ranges. When the node addresses are exhausted, expanding the primary range of that subnet with gcloud compute networks subnets expand-ip-range adds addresses in place, without downtime or recreating anything. All the clusters sharing the subnet can then add nodes, which directly solves the growth requirement.
Why the other options are wrong:
- A. Existing clusters keep drawing node addresses from their current subnet, so a new subnet does not give them room to grow without reconfiguration or migration.
- B. Alias IP ranges are secondary ranges used for Pods and Services, not for node addresses, so they do not relieve node IP exhaustion.
- C. A peered VPC network does not add addresses to the existing clusters' subnet, so their nodes still could not scale.
Question 23
Your organization has user identities in Active Directory. Your organization wants to use Active Directory as their source of truth for identities. Your organization wants to have full control over the Google accounts used by employees for all Google services, including your Google Cloud Platform (GCP) organization. What should you do?
- A. Use Google Cloud Directory Sync (GCDS) to synchronize users into Cloud Identity.
- B. Use the Cloud Identity APIs and write a script to synchronize users to Cloud Identity.
- C. Export users from Active Directory as a CSV and import them to Cloud Identity via the Admin Console.
- D. Ask each employee to create a Google account using self signup. Require that each employee use their company email address and password.
Show answer and explanation ▾
Correct answer: A
Active Directory must remain the source of truth while the company controls the Google accounts used for all Google services. Google Cloud Directory Sync (GCDS) is Google's supported tool for this: it reads users and groups from Active Directory or another LDAP server and provisions, updates and suspends the matching accounts in Cloud Identity on a schedule. Synchronization is one-way from AD, so changes made there flow to Google automatically, and the managed Cloud Identity accounts are then used to access the GCP organization.
Why the other options are wrong:
- B. A custom script against the Cloud Identity APIs could work, but it must be built and maintained in house, which adds effort and risk compared with the supported GCDS tool.
- C. A CSV import is a one-time manual copy. Later changes in Active Directory would not be reflected, so AD would not stay the source of truth.
- D. Self signup creates consumer accounts that the organization does not manage, which removes the required control and does not sync with Active Directory.
Question 24
You have a Compute Engine instance hosting an application used between 9 AM and 6 PM on weekdays. You want to back up this instance daily for disaster recovery purposes. You want to keep the backups for 30 days. You want the Google-recommended solution with the least management overhead and the least number of services. What should you do?
- A. 1. Update your instances' metadata to add the following value: snapshot-schedule: 0 1 * * * 2. Update your instances' metadata to add the following value: snapshot-retention: 30
- B. 1. In the Cloud Console, go to the Compute Engine Disks page and select your instance's disk. 2. In the Snapshot Schedule section, select Create Schedule and configure the following parameters: Schedule frequency: Daily; Start time: 1:00 AM - 2:00 AM; Autodelete snapshots after: 30 days.
- C. 1. Create a Cloud Function that creates a snapshot of your instance's disk. 2. Create a Cloud Function that deletes snapshots that are older than 30 days. 3. Use Cloud Scheduler to trigger both Cloud Functions daily at 1:00 AM.
- D. 1. Create a bash script in the instance that copies the content of the disk to Cloud Storage. 2. Create a bash script in the instance that deletes data older than 30 days in the backup Cloud Storage bucket. 3. Configure the instance's crontab to execute these scripts daily at 1:00 AM.
Show answer and explanation ▾
Correct answer: B
The requirements are daily backups kept for 30 days with the least management overhead and the fewest services. A snapshot schedule is a built-in Compute Engine feature: you attach a schedule to the instance's persistent disk, set a daily frequency with a start window outside business hours, and set snapshots to auto-delete after 30 days. Compute Engine then creates and expires snapshots automatically, with no code, extra services or scripts to maintain.
Why the other options are wrong:
- A. Instance metadata keys such as snapshot-schedule or snapshot-retention are not read by Compute Engine, so no snapshots would be created.
- C. Cloud Functions plus Cloud Scheduler can work but adds two services and custom code to maintain, which the built-in snapshot schedule avoids.
- D. Custom backup scripts and crontab entries on the instance require ongoing maintenance and do not produce consistent disk snapshots, adding overhead.
Question 25
Your company's infrastructure is on-premises, but all machines are running at maximum capacity. You want to burst to Google Cloud. The workloads on Google Cloud must be able to directly communicate to the workloads on-premises using a private IP range. What should you do?
- A. In Google Cloud, configure the VPC as a host for Shared VPC.
- B. In Google Cloud, configure the VPC for VPC Network Peering.
- C. Create bastion hosts both in your on-premises environment and on Google Cloud. Configure both as proxy servers using their public IP addresses.
- D. Set up Cloud VPN between the infrastructure on-premises and Google Cloud.
Show answer and explanation ▾
Correct answer: D
Bursting requires workloads in Google Cloud to reach on-premises workloads directly over private IP addresses. Cloud VPN creates encrypted IPsec tunnels between the on-premises network and a VPC network, extending private routing across both environments. It is quick to set up and is the standard way to connect an on-premises network to Google Cloud privately when a dedicated Interconnect is not required.
Why the other options are wrong:
- A. Shared VPC shares a network between Google Cloud projects; it does not connect to an on-premises network.
- B. VPC Network Peering connects VPC networks within Google Cloud, not an on-premises data center.
- C. Proxying through bastion hosts on public IPs does not provide direct private IP communication and adds management overhead.
Question 26
You need to set up permissions for a set of Compute Engine instances to enable them to write data into a particular Cloud Storage bucket. You want to follow Google-recommended practices. What should you do?
- A. Create a service account with an access scope. Use the access scope 'https://www.googleapis.com/auth/devstorage.write_only'.
- B. Create a service account with an access scope. Use the access scope 'https://www.googleapis.com/auth/cloud-platform'.
- C. Create a service account and add it to the IAM role 'storage.objectCreator' for that bucket.
- D. Create a service account and add it to the IAM role 'storage.objectAdmin' for that bucket.
Show answer and explanation ▾
Correct answer: C
Google recommends controlling what a VM's service account can do with IAM roles, granted with least privilege, rather than relying on access scopes. The VMs only need to write objects into one bucket, so the right approach is to create a dedicated service account, attach it to the instances, and grant it the Storage Object Creator role (roles/storage.objectCreator) on that specific bucket. This role allows creating objects but not reading, overwriting or deleting them, and binding it at the bucket level keeps the permission confined to the single bucket that needs it.
Why the other options are wrong:
- A. Access scopes are a legacy mechanism that only limits API reach; they do not grant permissions on the bucket, and Google recommends IAM roles instead.
- B. The cloud-platform scope just opens all APIs to whatever IAM allows; without an IAM role on the bucket the VMs still cannot write.
- D. Storage Object Admin also allows reading, overwriting and deleting objects, which exceeds the write-only need and breaks least privilege.
Question 27
You are working with a Cloud SQL MySQL database at your company. You need to retain a month-end copy of the database for three years for audit purposes. What should you do?
- A. Set up an export job for the first of the month. Write the export file to an Archive class Cloud Storage bucket.
- B. Save the automatic first-of-the-month backup for three years. Store the backup file in an Archive class Cloud Storage bucket.
- C. Set up an on-demand backup for the first of the month. Write the backup to an Archive class Cloud Storage bucket.
- D. Convert the automatic first-of-the-month backup to an export file. Write the export file to a Coldline class Cloud Storage bucket.
Show answer and explanation ▾
Correct answer: A
Keeping a month-end copy of a Cloud SQL database for three years calls for an export, not a backup. Cloud SQL backups are managed by the service, are tied to the instance and cannot be written to your own Cloud Storage bucket. An export creates a SQL dump or CSV file in a bucket you choose, so a monthly scheduled export written to an Archive class bucket provides an independent, durable copy at the lowest storage cost for data that is rarely read, kept well beyond normal backup retention.
Why the other options are wrong:
- B. Automatic backups are stored and managed by Cloud SQL and cannot be saved as files in a Cloud Storage bucket.
- C. On-demand backups are also stored by Cloud SQL itself and cannot be written to an Archive class bucket.
- D. Backups cannot be converted directly into export files, and Coldline costs more than Archive for data kept three years and rarely read.
Question 28
You are using Google Kubernetes Engine with autoscaling enabled to host a new application. You want to expose this new application to the public, using HTTPS on a public IP address. What should you do?
- A. Create a Kubernetes Service of type NodePort for your application, and a Kubernetes Ingress to expose this Service via a Cloud Load Balancer.
- B. Create a Kubernetes Service of type ClusterIP for your application. Configure the public DNS name of your application using the IP of this Service.
- C. Create a Kubernetes Service of type NodePort to expose the application on port 443 of each node of the Kubernetes cluster. Configure the public DNS name of your application with the IP of every node of the cluster to achieve load-balancing.
- D. Create a HAProxy pod in the cluster to load-balance the traffic to all the pods of the application. Forward the public traffic to HAProxy with an iptable rule. Configure the DNS name of your application using the public IP of the node HAProxy is running on.
Show answer and explanation ▾
Correct answer: A
To serve HTTPS on a public IP from GKE, the standard approach is a Kubernetes Ingress backed by a Service. When you create an Ingress, GKE provisions an external Application Load Balancer with a public IP address, and you can attach TLS certificates to terminate HTTPS. The Service of type NodePort gives the load balancer a backend to send traffic to on each node. Because the load balancer tracks the backends automatically, the setup keeps working as the cluster autoscaler adds or removes nodes.
Why the other options are wrong:
- B. A ClusterIP Service is reachable only from inside the cluster, so its IP cannot be used in a public DNS record.
- C. Pointing DNS at every node IP gives no health checking or HTTPS termination and breaks as autoscaling adds or removes nodes.
- D. A single HAProxy pod with iptables rules is a manual single point of failure and ties the public IP to one node that may be replaced.
Question 29
You need to grant access for three users so that they can view and edit table data on a Cloud Spanner instance. What should you do?
- A. Run gcloud iam roles describe roles/spanner.databaseUser. Add the users to the role.
- B. Run gcloud iam roles describe roles/spanner.databaseUser. Add the users to a new group. Add the group to the role.
- C. Run gcloud iam roles describe roles/spanner.viewer --project my-project. Add the users to the role.
- D. Run gcloud iam roles describe roles/spanner.viewer --project my-project. Add the users to a new group. Add the group to the role.
Show answer and explanation ▾
Correct answer: B
The users need to read and write table data, which the Cloud Spanner Database User role (roles/spanner.databaseUser) provides, and describing the role confirms its permissions. Google recommends granting roles to groups rather than to individual users, so placing the three users in a new group and granting the group the role makes access easier to audit and change later. Adding or removing a person then requires only a group membership change.
Why the other options are wrong:
- A. The role is correct, but granting it to individual users instead of a group is harder to manage and not recommended.
- C. Spanner Viewer allows only viewing instances and databases, not editing table data.
- D. Using a group is good practice, but Spanner Viewer still cannot edit table data.
Question 30
You are about to deploy a new Enterprise Resource Planning (ERP) system on Google Cloud. The application holds the full database in-memory for fast data access, and you need to configure the most appropriate resources on Google Cloud for this application. What should you do?
- A. Provision preemptible Compute Engine instances.
- B. Provision Compute Engine instances with GPUs attached.
- C. Provision Compute Engine instances with local SSDs attached.
- D. Provision Compute Engine instances with M1 machine type.
Show answer and explanation ▾
Correct answer: D
An ERP system that holds its full database in memory needs a very large amount of RAM per VM. The M1 machine series is part of Compute Engine's memory-optimized family, designed for large in-memory databases such as SAP HANA and offering terabytes of memory per instance. Choosing M1 machine types provides the high memory-to-vCPU ratio this workload requires, which general-purpose or accelerator options do not.
Why the other options are wrong:
- A. Preemptible VMs can be stopped at any time, which is unsuitable for a production ERP system, and they add no extra memory.
- B. GPUs accelerate parallel computation such as machine learning and graphics; they do not provide the memory for an in-memory database.
- C. Local SSDs provide fast disk storage, but the database is held in memory, so the deciding need is RAM, not disk.
Question 31
You are monitoring an application and receive user feedback that a specific error is spiking. You notice that the error is caused by a Service Account having insufficient permissions. You are able to solve the problem but want to be notified if the problem recurs. What should you do?
- A. In the Log Viewer, filter the logs on severity 'Error' and the name of the Service Account.
- B. Create a sink to BigQuery to export all the logs. Create a Data Studio dashboard on the exported logs.
- C. Create a custom log-based metric for the specific error to be used in an Alerting Policy.
- D. Grant Project Owner access to the Service Account.
Show answer and explanation ▾
Correct answer: C
The requirement is to be notified automatically if the permission error recurs. A custom log-based metric in Cloud Logging counts log entries that match a filter for the specific error, for example the permission denied message and the service account name. Cloud Monitoring can then use that metric in an alerting policy that fires when the count rises above zero or a chosen threshold, and the policy sends notifications through the configured channels.
Why the other options are wrong:
- A. Filtering logs in the Log Viewer is a manual, one-time search and sends no notification when the error happens again.
- B. A BigQuery sink and Data Studio dashboard only visualize logs; someone must watch the dashboard, so no alert is sent.
- D. Granting Project Owner violates least privilege and does nothing to notify you if a problem recurs.
Question 32
Your company has developed a new application that consists of multiple microservices. You want to deploy the application to Google Kubernetes Engine (GKE), and you want to ensure that the cluster can scale as more applications are deployed in the future. You want to avoid manual intervention when each new application is deployed. What should you do?
- A. Deploy the application on GKE, and add a HorizontalPodAutoscaler to the deployment.
- B. Deploy the application on GKE, and add a VerticalPodAutoscaler to the deployment.
- C. Create a GKE cluster with autoscaling enabled on the node pool. Set a minimum and maximum for the size of the node pool.
- D. Create a separate node pool for each application, and deploy each application to its dedicated node pool.
Show answer and explanation ▾
Correct answer: C
The requirement is for the cluster itself to grow as more applications are deployed, without manual intervention. Enabling the cluster autoscaler on the node pool with a minimum and maximum size lets GKE add nodes automatically when pods cannot be scheduled because of insufficient capacity, and remove underused nodes when demand falls. New applications can then be deployed at any time and GKE supplies the compute they need within the configured bounds, keeping costs controlled by the maximum limit.
Why the other options are wrong:
- A. A HorizontalPodAutoscaler scales pod replicas of one deployment; it does not add nodes, so the cluster can still run out of capacity as new applications arrive.
- B. A VerticalPodAutoscaler adjusts CPU and memory requests of existing pods; it does not add cluster capacity for newly deployed applications.
- D. Creating a dedicated node pool for every application requires manual work for each deployment, which is exactly what the stem wants to avoid.
Question 33
You need to create an autoscaling managed instance group for an HTTPS web application. You want to make sure that unhealthy VMs are recreated. What should you do?
- A. Create a health check on port 443 and use that when creating the Managed Instance Group.
- B. Select Multi-Zone instead of Single-Zone when creating the Managed Instance Group.
- C. In the Instance Template, add the label 'health-check'.
- D. In the Instance Template, add a startup script that sends a heartbeat to the metadata server.
Show answer and explanation ▾
Correct answer: A
Managed instance groups recreate unhealthy VMs through autohealing, and autohealing is driven by an application-based health check attached to the group. Because the application serves HTTPS, a health check that probes port 443 tests whether the web service on each VM is actually responding. When a VM fails the check for the configured number of consecutive probes, the managed instance group deletes and recreates it from the instance template. Configuring that health check when creating the group is the supported way to meet the requirement that unhealthy VMs are recreated automatically.
Why the other options are wrong:
- B. A multi-zone group spreads instances across zones for availability, but zone distribution alone does not detect or recreate an unhealthy VM.
- C. Labels are only key-value metadata for organizing and filtering resources; a label named health-check does not configure any health checking.
- D. The metadata server does not monitor heartbeats or trigger VM recreation, so a heartbeat script has no effect on autohealing.
Question 34
You have a virtual machine that is currently configured with 2 vCPUs and 4 GB of memory. It is running out of memory. You want to upgrade the virtual machine to have 8 GB of memory. What should you do?
- A. Rely on live migration to move the workload to a machine with more memory.
- B. Use gcloud to add metadata to the VM. Set the key to required-memory-size and the value to 8 GB.
- C. Stop the VM, change the machine type to n1-standard-8, and start the VM.
- D. Stop the VM, increase the memory to 8 GB, and start the VM.
Show answer and explanation ▾
Correct answer: D
The VM needs more memory but not more vCPUs. Compute Engine lets you change the machine type of a stopped instance, including switching to a custom machine type with the same 2 vCPUs and 8 GB of memory. Stopping the VM, increasing the memory to 8 GB, and starting it again solves the memory shortage while paying only for the added memory.
Why the other options are wrong:
- A. Live migration moves a VM to another host during maintenance without changing its machine type or memory.
- B. Metadata keys do not change hardware resources, so no metadata value can add memory.
- C. n1-standard-8 has 8 vCPUs and 30 GB of memory, far more than required and much more expensive.
Question 35
Your company has a Google Cloud Platform project that uses BigQuery for data warehousing. Your data science team changes frequently and has few members.
You need to allow members of this team to perform queries. You want to follow Google-recommended practices. What should you do?
- A. 1. Create an IAM entry for each data scientist's user account. 2. Assign the BigQuery jobUser role to the group.
- B. 1. Create an IAM entry for each data scientist's user account. 2. Assign the BigQuery dataViewer user role to the group.
- C. 1. Create a dedicated Google group in Cloud Identity. 2. Add each data scientist's user account to the group. 3. Assign the BigQuery jobUser role to the group.
- D. 1. Create a dedicated Google group in Cloud Identity. 2. Add each data scientist's user account to the group. 3. Assign the BigQuery dataViewer user role to the group.
Show answer and explanation ▾
Correct answer: C
Google recommends granting roles to groups rather than to individual users, especially when membership changes often: people are simply added to or removed from the group without touching IAM policies. Running queries requires permission to create BigQuery jobs, which the BigQuery Job User role (roles/bigquery.jobUser) provides. Creating a dedicated Google group in Cloud Identity, adding each data scientist to it, and granting that group the Job User role lets every team member run queries while keeping administration simple as the team changes.
Why the other options are wrong:
- A. Creating IAM entries per user is not recommended for a frequently changing team, and the steps are inconsistent because no group is ever created.
- B. Per-user IAM entries add administrative overhead, and Data Viewer only reads data and metadata; it does not grant permission to run query jobs.
- D. Using a group is correct, but BigQuery Data Viewer lacks the job creation permission needed to run queries.
Question 36
You are performing a monthly security check of your Google Cloud environment and want to know who has access to view data stored in your Google Cloud Project. What should you do?
- A. Enable Audit Logs for all APIs that are related to data storage.
- B. Review the IAM permissions for any role that allows for data access.
- C. Review the Identity-Aware Proxy settings for each resource.
- D. Create a Data Loss Prevention job.
Show answer and explanation ▾
Correct answer: B
The question asks who has access to view data, which is determined by IAM. Reviewing the IAM allow policies for the project and its resources, and looking for roles that contain data read permissions such as storage.objects.get or BigQuery data viewer permissions, shows exactly which principals are able to read the data. Tools such as the IAM page or Policy Analyzer help with this review. This answers the question of who can access data, rather than who has already accessed it.
Why the other options are wrong:
- A. Audit logs record access that has already happened; they do not list everyone who currently has permission to view data.
- C. IAP settings cover access to web applications and VMs behind IAP, not overall permissions on stored data.
- D. A DLP job inspects and classifies sensitive content but does not report which identities can access it.
Question 37
You created an instance of SQL Server 2017 on Compute Engine to test features in the new version. You want to connect to this instance using the fewest number of steps. What should you do?
- A. Install an RDP client on your desktop. Verify that a firewall rule for port 3389 exists.
- B. Install an RDP client on your desktop. Set a Windows username and password in the GCP Console. Use the credentials to log in to the instance.
- C. Set a Windows password in the GCP Console. Verify that a firewall rule for port 22 exists. Click the RDP button in the GCP Console and supply the credentials to log in.
- D. Set a Windows username and password in the GCP Console. Verify that a firewall rule for port 3389 exists. Click the RDP button in the GCP Console, and supply the credentials to log in.
Show answer and explanation ▾
Correct answer: D
SQL Server 2017 on Compute Engine runs on a Windows Server image, so access is through Remote Desktop Protocol. In the Console you set a Windows password for a username, which creates the account on the instance and returns its credentials. RDP uses TCP port 3389, so a firewall rule allowing it must exist. Clicking the RDP button in the Console then opens the connection, and you supply the credentials to log in, with nothing extra to install.
Why the other options are wrong:
- A. Installing a client and checking the firewall does not create the Windows credentials needed to log in to the instance.
- B. Installing a separate RDP client is an extra step compared with the Console RDP button, and it skips verifying the port 3389 rule.
- C. Port 22 is used for SSH, not RDP, so a firewall rule for port 22 does not allow a Remote Desktop connection.
Question 38
You need to monitor resources that are distributed over different projects in Google Cloud Platform. You want to consolidate reporting under the same Stackdriver Monitoring dashboard. What should you do?
- A. Use Shared VPC to connect all projects, and link Stackdriver to one of the projects.
- B. For each project, create a Stackdriver account. In each project, create a service account for that project and grant it the role of Stackdriver Account Editor in all other projects.
- C. Configure a single Stackdriver account, and link all projects to the same account.
- D. Configure a single Stackdriver account for one of the projects. In Stackdriver, create a Group and add the other project names as criteria for that Group.
Show answer and explanation ▾
Correct answer: C
Stackdriver Monitoring (now Cloud Monitoring) can view metrics from several projects in one place by using a single Stackdriver account, now called a workspace or metrics scope, that has the other projects linked to it. Once all the projects are linked, one dashboard can chart and alert on resources from every project together. Configuring a single account and linking all projects to it is the supported way to consolidate reporting.
Why the other options are wrong:
- A. Shared VPC shares network resources between projects; it does not combine monitoring data into one dashboard.
- B. Separate accounts with cross-project service accounts add overhead and still leave monitoring split across multiple dashboards.
- D. Groups filter resources within data the account already monitors; they cannot pull in metrics from projects that are not linked.
Question 39
Your VMs are running in a subnet that has a subnet mask of 255.255.255.240. The current subnet has no more free IP addresses and you require an additional 10 IP addresses for new VMs. The existing and new VMs should all be able to reach each other without additional routes. What should you do?
- A. Use gcloud to expand the IP range of the current subnet.
- B. Delete the subnet, and recreate it using a wider range of IP addresses.
- C. Create a new project. Use Shared VPC to share the current network with the new project.
- D. Create a new subnet with the same starting IP but a wider range to overwrite the current subnet.
Show answer and explanation ▾
Correct answer: A
A subnet mask of 255.255.255.240 is a /28, which offers very few usable addresses. VPC subnets can have their primary range expanded in place with gcloud compute networks subnets expand-ip-range, setting a shorter prefix length. Expansion does not disrupt running VMs, keeps existing addresses, and leaves all old and new VMs in the same subnet, so they reach each other through the subnet route without any additional routes.
Why the other options are wrong:
- B. Deleting the subnet requires removing every VM in it first, causing downtime that in-place expansion avoids.
- C. Shared VPC shares the network with another project but does not add any IP addresses to the exhausted subnet.
- D. Subnets cannot overlap within a VPC, so a new subnet with the same starting IP cannot overwrite the current one.
Question 40
Your Dataproc cluster runs in a single Virtual Private Cloud (VPC) network in a single subnet with range 172.16.20.128/25. There are no private IP addresses available in the VPC network. You want to add new VMs to communicate with your cluster using the minimum number of steps. What should you do?
- A. Modify the existing subnet range to 172.16.20.0/24.
- B. Create a new Secondary IP Range in the VPC and configure the VMs to use that range.
- C. Create a new VPC network for the VMs. Enable VPC Peering between the VMs' VPC network and the Dataproc cluster VPC network.
- D. Create a new VPC network for the VMs with a subnet of 172.32.0.0/16. Enable VPC network Peering between the Dataproc VPC network and the VMs VPC network. Configure a custom Route exchange.
Show answer and explanation ▾
Correct answer: A
The subnet 172.16.20.128/25 has run out of addresses, and the goal is the fewest steps. VPC subnets can have their primary IPv4 range expanded in place, without downtime, as long as the new range contains the old one and does not overlap other subnets. The range 172.16.20.0/24 fully contains 172.16.20.128/25 and doubles the available addresses, so a single expansion lets the new VMs join the same subnet and talk to the Dataproc cluster directly, with no peering, new networks or route changes.
Why the other options are wrong:
- B. Secondary ranges are used for alias IPs such as GKE Pods, not as a VM's primary interface address, so new VMs cannot simply be placed in one.
- C. A new VPC network plus peering works but takes several more steps than expanding the existing subnet, violating the minimum-steps requirement.
- D. A new network, peering and custom route exchange add even more steps, and 172.32.0.0/16 is public address space rather than an RFC 1918 private range.
Question 41
Your company has an existing GCP organization with hundreds of projects and a billing account. Your company recently acquired another company that also has hundreds of projects and its own billing account. You would like to consolidate all GCP costs of both GCP organizations onto a single invoice. You would like to consolidate all costs as of tomorrow. What should you do?
- A. Link the acquired company's projects to your company's billing account.
- B. Configure the acquired company's billing account and your company's billing account to export the billing data into the same BigQuery dataset.
- C. Migrate the acquired company's projects into your company's GCP organization. Link the migrated projects to your company's billing account.
- D. Create a new GCP organization and a new billing account. Migrate the acquired company's projects and your company's projects into the new GCP organization and link the projects to the new billing account.
Show answer and explanation ▾
Correct answer: A
A project can be linked to any billing account the user has permission on, regardless of which organization owns the project. Linking the acquired company's projects to your company's billing account immediately places their charges on that account's single invoice, starting from the time of the change. No project migration between organizations is required, so the consolidation can be completed by tomorrow with the least effort and disruption.
Why the other options are wrong:
- B. Exporting both accounts to one BigQuery dataset consolidates reporting only; there would still be two separate invoices.
- C. Migrating hundreds of projects between organizations is slow and unnecessary, since linking billing alone consolidates the invoice.
- D. Creating a new organization and moving every project is a large, disruptive effort that cannot be done reliably by tomorrow.
Question 42
You are hosting an application on bare-metal servers in your own data center. The application needs access to Cloud Storage. However, security policies prevent the servers hosting the application from having public IP addresses or access to the internet. You want to follow Google-recommended practices to provide the application with access to Cloud Storage. What should you do?
- A. 1. Use nslookup to get the IP address for storage.googleapis.com. 2. Negotiate with the security team to be able to give a public IP address to the servers. 3. Only allow egress traffic from those servers to the IP addresses for storage.googleapis.com.
- B. 1. Using Cloud VPN, create a VPN tunnel to a Virtual Private Cloud (VPC) in Google Cloud. 2. In this VPC, create a Compute Engine instance and install the Squid proxy server on this instance. 3. Configure your servers to use that instance as a proxy to access Cloud Storage.
- C. 1. Use Migrate for Compute Engine (formerly known as Velostrata) to migrate those servers to Compute Engine. 2. Create an internal load balancer (ILB) that uses storage.googleapis.com as backend. 3. Configure your new instances to use this ILB as proxy.
- D. 1. Using Cloud VPN or Interconnect, create a tunnel to a VPC in Google Cloud. 2. Use Cloud Router to create a custom route advertisement for 199.36.153.4/30. Announce that network to your on-premises network through the VPN tunnel. 3. In your on-premises network, configure your DNS server to resolve *.googleapis.com as a CNAME to restricted.googleapis.com.
Show answer and explanation ▾
Correct answer: D
On-premises hosts without internet access can reach Google APIs privately through Private Google Access for on-premises hosts. Traffic travels over Cloud VPN or Cloud Interconnect into a VPC, and Cloud Router advertises the restricted.googleapis.com range, 199.36.153.4/30, to the on-premises network as a custom route. The on-premises DNS is configured so that *.googleapis.com resolves as a CNAME to restricted.googleapis.com, sending Cloud Storage requests to those addresses over the private connection. This is the Google-recommended pattern and requires no public IP addresses.
Why the other options are wrong:
- A. Assigning public IPs and reaching the internet directly violates the stated security policy.
- B. A self-managed Squid proxy VM adds maintenance and a single point of failure when Private Google Access provides this natively.
- C. Migrating servers is out of scope, and an internal load balancer cannot use storage.googleapis.com as a backend.
Question 43
Your company uses BigQuery for data warehousing. Over time, many different business units in your company have created 1000+ datasets across hundreds of projects. Your CIO wants you to examine all datasets to find tables that contain an employee_ssn column. You want to minimize effort in performing this task.
What should you do?
- A. Go to Data Catalog and search for employee_ssn in the search box.
- B. Write a shell script that uses the bq command line tool to loop through all the projects in your organization.
- C. Write a script that loops through all the projects in your organization and runs a query on INFORMATION_SCHEMA.COLUMNS view to find the employee_ssn column.
- D. Write a Cloud Dataflow job that loops through all the projects in your organization and runs a query on INFORMATION_SCHEMA.COLUMNS view to find the employee_ssn column.
Show answer and explanation ▾
Correct answer: A
Data Catalog (now part of Dataplex Universal Catalog) automatically indexes metadata for BigQuery datasets, tables and columns across the projects you can access. A single search for employee_ssn, optionally qualified as a column search, returns every table that contains that column across all projects. No scripts or jobs are needed, which meets the goal of minimizing effort for 1000+ datasets.
Why the other options are wrong:
- B. A bq shell script across hundreds of projects means writing and maintaining custom code, far more effort than a catalog search.
- C. Querying INFORMATION_SCHEMA in each project works only through custom looping code, which adds effort the catalog search avoids.
- D. Writing a Dataflow pipeline for a metadata lookup adds development and running cost far beyond a simple catalog search.
Question 44
You need to create a copy of a custom Compute Engine virtual machine (VM) to facilitate an expected increase in application traffic due to a business acquisition.
What should you do?
- A. Create a Compute Engine snapshot of your base VM. Create your images from that snapshot.
- B. Create a Compute Engine snapshot of your base VM. Create your instances from that snapshot.
- C. Create a custom Compute Engine image from a snapshot. Create your images from that image.
- D. Create a custom Compute Engine image from a snapshot. Create your instances from that image.
Show answer and explanation ▾
Correct answer: D
To create several copies of a customized VM for increased traffic, the recommended approach is to capture the configured boot disk as a custom image and launch new instances from it. Creating the image from a snapshot avoids stopping the source VM, and a custom image is the reusable artifact intended for repeated instance creation; it can also back an instance template for a managed instance group. Each new instance starts with the same operating system, software and configuration as the base VM.
Why the other options are wrong:
- A. Creating more images from a snapshot only produces images; it never creates the additional VM instances needed to handle the traffic.
- B. Snapshots are designed for backup and restore, while custom images are the recommended source for creating many identical instances and instance templates.
- C. Creating images from an image only duplicates the image and still does not create any instances to serve the extra traffic.
Question 45
Your company has a large quantity of unstructured data in different file formats. You want to perform ETL transformations on the data. You need to make the data accessible on Google Cloud so it can be processed by a Dataflow job. What should you do?
- A. Upload the data to BigQuery using the bq command line tool.
- B. Upload the data to Cloud Storage using the gsutil command line tool.
- C. Upload the data into Cloud SQL using the import function in the console.
- D. Upload the data into Cloud Spanner using the import function in the console.
Show answer and explanation ▾
Correct answer: B
The data is unstructured and comes in many file formats, so it needs a store that accepts arbitrary objects without a schema. Cloud Storage holds any file type at scale, and Dataflow reads directly from Cloud Storage paths through built-in I/O connectors for text, Avro, Parquet and other formats. Uploading the files with gsutil (or gcloud storage) makes them available for the Dataflow ETL pipeline without forcing a schema or a database load before the transformation.
Why the other options are wrong:
- A. BigQuery stores structured or semi-structured data in tables with a schema, so loading unstructured files of mixed formats would need transformation first.
- C. Cloud SQL is a relational database for structured rows. Importing unstructured files requires a schema and does not suit large mixed-format data.
- D. Cloud Spanner is a relational database for structured, transactional data. Its import expects specific formats and is costly for raw unstructured files.
Question 46
Your company runs one batch process in an on-premises server that takes around 30 hours to complete. The task runs monthly, can be performed offline, and must be restarted if interrupted. You want to migrate this workload to the cloud while minimizing cost. What should you do?
- A. Migrate the workload to a Compute Engine Preemptible VM.
- B. Migrate the workload to a Google Kubernetes Engine cluster with Preemptible nodes.
- C. Migrate the workload to a Compute Engine VM. Start and stop the instance as needed.
- D. Create an Instance Template with Preemptible VMs On. Create a Managed Instance Group from the template and adjust Target CPU Utilization. Migrate the workload.
Show answer and explanation ▾
Correct answer: C
The job runs for about 30 hours and must restart from the beginning if interrupted. Preemptible VMs are always stopped within 24 hours and can be reclaimed at any time, so the job would never finish reliably on them. A standard Compute Engine VM that is started when the monthly run begins and stopped when it ends incurs compute charges only while running, so the cost stays low and the job completes without interruption.
Why the other options are wrong:
- A. Preemptible VMs are terminated after at most 24 hours, so a 30-hour job that must restart when interrupted would never complete.
- B. Preemptible GKE nodes have the same 24-hour limit and add cluster management overhead for a single batch process.
- D. A managed instance group of preemptible VMs still stops each VM within 24 hours, and CPU-based autoscaling does not help a single uninterruptible job.
Question 47
You have an application that receives SSL-encrypted TCP traffic on port 443. Clients for this application are located all over the world. You want to minimize latency for the clients. Which load balancing option should you use?
- A. HTTPS Load Balancer
- B. Network Load Balancer
- C. SSL Proxy Load Balancer
- D. Internal TCP/UDP Load Balancer. Add a firewall rule allowing ingress traffic from 0.0.0.0/0 on the target instances.
Show answer and explanation ▾
Correct answer: A
Encrypted traffic on port 443 from clients around the world is HTTPS traffic, and Google recommends the global external HTTPS Load Balancer for it. The load balancer uses a single anycast IP address and terminates TLS at Google Front Ends close to each client, then carries requests over Google's network to the nearest healthy backend, which minimizes latency for a worldwide user base. It also adds HTTP-aware features such as URL maps and Cloud CDN integration, which a TCP-level proxy cannot provide.
Why the other options are wrong:
- B. The Network Load Balancer is a regional pass-through balancer, so it does not terminate connections near clients worldwide and cannot reduce latency for a global audience.
- C. SSL Proxy Load Balancing is intended for encrypted non-HTTP TCP traffic; for HTTPS on port 443 Google recommends the HTTPS Load Balancer instead.
- D. An Internal TCP/UDP Load Balancer only serves clients inside the VPC network, and a firewall rule cannot make it reachable from the internet.
Question 48
You need to assign a Cloud Identity and Access Management (Cloud IAM) role to an external auditor. The auditor needs to have permissions to review your Google Cloud Platform (GCP) Audit Logs and also to review your Data Access logs. What should you do?
- A. Assign the auditor the IAM role roles/logging.privateLogViewer. Perform the export of logs to Cloud Storage.
- B. Assign the auditor the IAM role roles/logging.privateLogViewer. Direct the auditor to also review the logs for changes to Cloud IAM policy.
- C. Assign the auditor's IAM user to a custom role that has logging.privateLogEntries.list permission. Perform the export of logs to Cloud Storage.
- D. Assign the auditor's IAM user to a custom role that has logging.privateLogEntries.list permission. Direct the auditor to also review the logs for changes to Cloud IAM policy.
Show answer and explanation ▾
Correct answer: B
Data Access audit logs are private logs, so the standard Logs Viewer role is not enough to read them. The predefined Private Logs Viewer role (roles/logging.privateLogViewer) includes everything in Logs Viewer plus permission to read Data Access logs, so one predefined role covers both the Admin Activity audit logs and the Data Access logs. The auditor can then review entries such as SetIamPolicy calls to see changes to IAM policy, with no export or custom role required.
Why the other options are wrong:
- A. The role is right, but exporting logs to Cloud Storage adds a sink, a bucket and extra access control for no benefit, because the auditor can already read the logs in Cloud Logging.
- C. A custom role with only logging.privateLogEntries.list lacks other viewing permissions and must be maintained manually, and the export to Cloud Storage adds unnecessary work.
- D. A custom role with a single permission is harder to maintain than the predefined Private Logs Viewer role, which Google recommends using when it matches the need.
Question 49
You need to verify that a Google Cloud Platform service account was created at a particular time. What should you do?
- A. Filter the Activity log to view the Configuration category. Filter the Resource type to Service Account.
- B. Filter the Activity log to view the Configuration category. Filter the Resource type to Google Project.
- C. Filter the Activity log to view the Data Access category. Filter the Resource type to Service Account.
- D. Filter the Activity log to view the Data Access category. Filter the Resource type to Google Project.
Show answer and explanation ▾
Correct answer: A
Creating a service account is an administrative change, so it is recorded as a configuration event in the Activity log (backed by Admin Activity audit logs). Filtering the Activity log to the Configuration category and setting the resource type to Service Account shows the creation event with its timestamp and the principal who performed it, which confirms exactly when the account was created.
Why the other options are wrong:
- B. Filtering on Google Project shows project-level changes, not the service account creation event you need.
- C. Data Access logs record reads of data and metadata, not administrative actions such as creating a service account.
- D. Data Access is the wrong category and Google Project is the wrong resource type for a service account creation event.
Question 50
You are building an application that processes data files uploaded from thousands of suppliers. Your primary goals for the application are data security and the expiration of aged data. You need to design the application to:
Restrict access so that suppliers can access only their own data.
Give suppliers write access to data only for 30 minutes.
Delete data that is over 45 days old.
You have a very short development cycle, and you need to make sure that the application requires minimal maintenance. Which two strategies should you use? (Choose two.)
- A. Build a lifecycle policy to delete Cloud Storage objects after 45 days.
- B. Use signed URLs to allow suppliers limited time access to store their objects.
- C. Set up an SFTP server for your application, and create a separate user for each supplier.
- D. Build a Cloud Function that triggers a timer of 45 days to delete objects that have expired.
- E. Develop a script that loops through all Cloud Storage buckets and deletes any buckets that are older than 45 days.
Show answer and explanation ▾
Correct answer: A, B
Both requirements can be met with built-in Cloud Storage features and almost no code. Signed URLs grant time-limited access to a specific object without the supplier needing a Google account, so the application can issue each supplier a URL that allows writing only their own object and expires after 30 minutes. An Object Lifecycle Management rule with a Delete action at an age of 45 days removes aged data automatically, with no servers, timers or scripts to maintain. Together they satisfy the security, expiry and minimal maintenance goals.
Why the other options are wrong:
- C. An SFTP server with thousands of per-supplier users is heavy to build and maintain and does not provide automatic 30-minute write windows.
- D. A custom Cloud Function timer reimplements what lifecycle rules already do, adding code to write and maintain.
- E. A custom script adds maintenance, and deleting whole buckets by age would remove newer objects instead of only data older than 45 days.
Question 51
You have one project called proj-sa where you manage all your service accounts. You want to be able to use a service account from this project to take snapshots of VMs running in another project called proj-vm. What should you do?
- A. Download the private key from the service account, and add it to each VM's custom metadata.
- B. Download the private key from the service account, and add the private key to each VM's SSH keys.
- C. Grant the service account the IAM Role of Compute Storage Admin in the project called proj-vm.
- D. When creating the VMs, set the service account's API scope for Compute Engine to read/write.
Show answer and explanation ▾
Correct answer: C
A service account can be granted roles in any project, not just the one where it was created. Granting the service account from proj-sa the Compute Storage Admin role (roles/compute.storageAdmin) in proj-vm gives it permission to manage disks, images and snapshots there, so it can take snapshots of the VMs. This keeps the account centrally managed in proj-sa while access is controlled through IAM in the target project.
Why the other options are wrong:
- A. Putting a private key in metadata exposes the credential and grants no permissions in proj-vm.
- B. SSH keys control login to VMs; a service account key placed there grants no snapshot permission.
- D. Access scopes only limit APIs for the VM's own attached account and grant no IAM permissions to the proj-sa service account.
Question 52
Your company has workloads running on Compute Engine and on-premises. The Google Cloud Virtual Private Cloud (VPC) is connected to your WAN over a Virtual Private Network (VPN). You need to deploy a new Compute Engine instance and ensure that no public Internet traffic can be routed to it. What should you do?
- A. Create the instance without a public IP address.
- B. Create the instance with Private Google Access enabled.
- C. Create a deny-all egress firewall rule on the VPC network.
- D. Create a route on the VPC to route all traffic to the instance over the VPN tunnel.
Show answer and explanation ▾
Correct answer: A
Traffic from the public internet can reach a Compute Engine instance only through an external IP address (or through a load balancer or similar proxy in front of it). Creating the instance without a public IP address means it has only an internal address, which is not routable from the internet, while it remains reachable from on-premises systems over the existing VPN and from other resources in the VPC network.
Why the other options are wrong:
- B. Private Google Access is a subnet setting that lets internal-only VMs reach Google APIs; it does not control whether internet traffic can reach an instance.
- C. A deny-all egress rule blocks outbound traffic, including traffic to on-premises systems, and does not by itself stop inbound internet traffic to an external IP.
- D. Routes control outgoing traffic paths and cannot stop internet traffic from reaching an instance that has a public IP address.
Question 53
You have a Compute Engine instance hosting a production application. You want to receive an email if the instance consumes more than 90% of its CPU resources for more than 15 minutes. You want to use Google services. What should you do?
- A. 1. Create a consumer Gmail account. 2. Write a script that monitors the CPU usage. 3. When the CPU usage exceeds the threshold, have that script send an email using the Gmail account and smtp.gmail.com on port 25 as SMTP server.
- B. 1. Create a Cloud Monitoring Workspace and associate your Google Cloud Platform (GCP) project with it. 2. Create a Cloud Monitoring Alerting Policy that uses the threshold as a trigger condition. 3. Configure your email address in the notification channel.
- C. 1. Create a Cloud Monitoring Workspace and associate your GCP project with it. 2. Write a script that monitors the CPU usage and sends it as a custom metric to Cloud Monitoring. 3. Create an uptime check for the instance in Cloud Monitoring.
- D. 1. In Cloud Logging, create a logs-based metric to extract the CPU usage by using this regular expression: CPU Usage: ([0-9]{1,3})% 2. In Cloud Monitoring, create an Alerting Policy based on this metric. 3. Configure your email address in the notification channel.
Show answer and explanation ▾
Correct answer: B
Cloud Monitoring collects CPU utilization for Compute Engine instances automatically. After the project is associated with a Cloud Monitoring Workspace (now the metrics scope), you create an alerting policy whose condition fires when CPU utilization stays above 90 percent for a 15-minute duration window, and attach an email notification channel. Monitoring then evaluates the metric continuously and sends the email when the condition is met, using Google services only and with no custom scripts.
Why the other options are wrong:
- A. A custom script with a consumer Gmail account is fragile and unmanaged, and Compute Engine blocks outbound traffic on port 25, so the email would not be sent.
- C. CPU is already a built-in metric, so a custom script is unnecessary, and an uptime check tests availability, not CPU usage, and would not alert on the threshold.
- D. A logs-based metric needs CPU usage to be written to logs in that format, which is not done by default, while the built-in CPU metric is already available.
Question 54
You are building an archival solution for your data warehouse and have selected Cloud Storage to archive your data. Your users need to be able to access this archived data once a quarter for some regulatory requirements. You want to select a cost-efficient option. Which storage option should you use?
- A. Cold Storage
- B. Nearline Storage
- C. Regional Storage
- D. Multi-Regional Storage
Show answer and explanation ▾
Correct answer: A
The archived data is read about once a quarter and the solution must be cost-efficient. Coldline Storage is designed for data read or modified at most once a quarter, with a 90-day minimum storage duration. Its storage price is lower than Nearline and Standard, and the higher retrieval cost is acceptable when access happens only a few times a year. That makes Coldline, the cold storage class, the most economical choice for quarterly regulatory access to archived warehouse data.
Why the other options are wrong:
- B. Nearline is aimed at data accessed about once a month; for quarterly access its higher storage price makes it more expensive than Coldline.
- C. Regional Standard storage is meant for frequently accessed data and costs more to store than the colder classes, so it is not cost-efficient for archives.
- D. Multi-Regional Standard storage targets frequently accessed, geo-redundant data and is the most expensive option for rarely read archives.
Question 55
Your company uses a large number of Google Cloud services centralized in a single project. All teams have specific projects for testing and development. The DevOps team needs access to all of the production services in order to perform their job. You want to prevent Google Cloud product changes from broadening their permissions in the future. You want to follow Google-recommended practices. What should you do?
- A. Grant all members of the DevOps team the role of Project Editor on the organization level.
- B. Grant all members of the DevOps team the role of Project Editor on the production project.
- C. Create a custom role that combines the required permissions. Grant the DevOps team the custom role on the production project.
- D. Create a custom role that combines the required permissions. Grant the DevOps team the custom role on the organization level.
Show answer and explanation ▾
Correct answer: C
Two requirements decide this: the DevOps team needs access only to production services, which live in a single project, and their permissions must not grow when Google adds new permissions to products. Basic and predefined roles are updated by Google over time, but a custom role contains only the permissions you list, so it stays fixed until you change it. Granting that custom role on the production project, rather than at the organization level, follows least privilege by limiting access to the resources the team actually needs.
Why the other options are wrong:
- A. Project Editor is a broad basic role that Google expands as products change, and granting it at the organization level covers every project.
- B. Project Editor is a basic role whose permissions Google can broaden over time, which violates the stated requirement.
- D. Granting the custom role at the organization level extends access to all test and development projects, exceeding what the team needs.
Question 56
You are deploying a production application on Compute Engine. You want to prevent anyone from accidentally destroying the instance by clicking the wrong button. What should you do?
- A. Disable the flag "Delete boot disk when instance is deleted."
- B. Enable delete protection on the instance.
- C. Disable Automatic restart on the instance.
- D. Enable Preemptibility on the instance.
Show answer and explanation ▾
Correct answer: B
Compute Engine offers a deletion protection setting on each instance. When it is enabled, any attempt to delete the VM, whether from the console, gcloud or the API, fails until someone deliberately turns the setting off. This directly prevents an accidental click on the delete button from destroying a production instance, while still allowing the VM to be stopped, restarted and managed normally. It can be set when the instance is created or added later on an existing instance.
Why the other options are wrong:
- A. Keeping the boot disk preserves its data, but the instance itself is still deleted by an accidental click.
- C. Automatic restart controls whether the VM restarts after a host failure; it does nothing to block deletion.
- D. Preemptible VMs can be stopped by Google at any time, which makes the instance less durable, not protected from deletion.
Question 57
You need to reduce GCP service costs for a division of your company using the fewest possible steps. You need to turn off all configured services in an existing GCP project. What should you do?
- A. 1. Verify that you are assigned the Project Owners IAM role for this project. 2. Locate the project in the GCP console, click Shut down and then enter the project ID.
- B. 1. Verify that you are assigned the Project Owners IAM role for this project. 2. Switch to the project in the GCP console, locate the resources and delete them.
- C. 1. Verify that you are assigned the Organizational Administrator IAM role for this project. 2. Locate the project in the GCP console, enter the project ID and then click Shut down.
- D. 1. Verify that you are assigned the Organizational Administrators IAM role for this project. 2. Switch to the project in the GCP console, locate the resources and delete them.
Show answer and explanation ▾
Correct answer: A
Shutting down a project stops all billing and turns off every resource and service in it in a single action, which is the fewest possible steps. The resourcemanager.projects.delete permission needed for this is included in the Project Owner role, so you first confirm you hold Owner on the project. In the console you then locate the project, click Shut down and type the project ID to confirm, and the project enters a 30-day pending deletion period during which it can still be restored.
Why the other options are wrong:
- B. Deleting resources one by one takes many steps and can miss services that keep incurring cost, whereas shutting down the project covers everything at once.
- C. Organization Administrator manages IAM policies at the organization level and does not include the permission to delete projects, so it is not the role to verify.
- D. Organization Administrator is the wrong role for deleting project resources, and removing resources individually needs far more steps than shutting down the project.
Question 58
You need to enable traffic between multiple groups of Compute Engine instances that are currently running in two different GCP projects. Each group of Compute Engine instances is running in its own VPC. What should you do?
- A. Verify that both projects are in a GCP Organization. Create a new VPC and add all instances.
- B. Verify that both projects are in a GCP Organization. Share the VPC from one project and request that the Compute Engine instances in the other project use this shared VPC.
- C. Verify that you are the Project Administrator of both projects. Create two new VPCs and add all instances.
- D. Verify that you are the Project Administrator of both projects. Create a new VPC and add all instances.
Show answer and explanation ▾
Correct answer: B
Shared VPC lets a host project share its VPC network with service projects in the same organization, so instances in different projects can attach to the same network and communicate over internal IP addresses. Shared VPC requires both projects to belong to a Google Cloud organization, which is why that check comes first. After the host project shares the network, instances in the service project are created on the shared subnets and can reach the other group directly, with centralized control of the network.
Why the other options are wrong:
- A. A VPC belongs to one project, so you cannot simply create a new VPC and add instances from another project to it without Shared VPC.
- C. Creating two new VPCs does not connect anything, and instances in separate VPCs still cannot talk to each other privately.
- D. Being Project Administrator does not let a single VPC span two projects; cross-project use of one network requires Shared VPC.
Question 59
You host a static website on Cloud Storage. Recently, you began to include links to PDF files on this site. Currently, when users click on the links to these PDF files, their browsers prompt them to save the file onto their local system. Instead, you want the clicked PDF files to be displayed within the browser window directly, without prompting the user to save the file locally. What should you do?
- A. Enable Cloud CDN on the website frontend.
- B. Enable 'Share publicly' on the PDF file objects.
- C. Set Content-Type metadata to application/pdf on the PDF file objects.
- D. Add a label to the storage bucket with a key of Content-Type and value of application/pdf.
Show answer and explanation ▾
Correct answer: C
Browsers decide whether to display or download a file based on the Content-Type header that Cloud Storage returns, which comes from each object's Content-Type metadata. If the PDF objects carry a generic type such as application/octet-stream, the browser offers to save them. Setting the Content-Type metadata on the PDF objects to application/pdf lets the browser recognize them and render them directly in the window.
Why the other options are wrong:
- A. Cloud CDN caches content closer to users but does not change the Content-Type the browser receives.
- B. Public sharing controls who can access the files, not how the browser handles them.
- D. Bucket labels are for organization and billing; they do not set HTTP headers on objects.
Question 60
You are the team lead of a group of 10 developers. You provided each developer with an individual Google Cloud Project that they can use as their personal sandbox to experiment with different Google Cloud solutions. You want to be notified if any of the developers are spending above $500 per month on their sandbox environment. What should you do?
- A. Create a single budget for all projects and configure budget alerts on this budget.
- B. Create a separate billing account per sandbox project and enable BigQuery billing exports. Create a Data Studio dashboard to plot the spending per billing account.
- C. Create a budget per project and configure budget alerts on all of these budgets.
- D. Create a single billing account for all sandbox projects and enable BigQuery billing exports. Create a Data Studio dashboard to plot the spending per project.
Show answer and explanation ▾
Correct answer: C
The goal is a notification when any single developer's sandbox exceeds $500 in a month. Cloud Billing budgets can be scoped to individual projects, so creating one budget per project with a $500 amount and alert thresholds sends a notification as soon as that specific project crosses the limit. This is a native, low-effort way to get per-project alerting. A single combined budget would only track the total across all ten projects and would not show which developer overspent.
Why the other options are wrong:
- A. One budget across all projects alerts on combined spend, so a single developer exceeding $500 would not trigger a notification on its own.
- B. Separate billing accounts add administrative overhead, and a dashboard only visualizes spending; it does not send notifications.
- D. A Data Studio dashboard shows spending per project but does not alert anyone, so you would have to check it manually.
Question 61
You have created a new project in Google Cloud through the gcloud command line interface (CLI) and linked a billing account. You need to create a new Compute Engine instance using the CLI. You need to perform the prerequisite steps. What should you do?
- A. Create a Cloud Monitoring Workspace.
- B. Create a VPC network in the project.
- C. Enable the compute.googleapis.com API.
- D. Grant yourself the IAM role of Compute Admin.
Show answer and explanation ▾
Correct answer: C
A project created through the gcloud CLI starts with the Compute Engine API disabled, even after a billing account is linked. Before any gcloud compute instances create command can succeed, you must enable the API, for example with gcloud services enable compute.googleapis.com. Enabling it also creates the default VPC network and the default Compute Engine service account in the project, so no separate network setup is needed for a basic instance.
Why the other options are wrong:
- A. Cloud Monitoring is optional observability and is not required to create a Compute Engine instance.
- B. Enabling the Compute Engine API creates the default network automatically, and you cannot create a VPC network until that API is enabled anyway.
- D. The project creator already has the Owner role, which includes Compute Engine permissions; the actual blocker is the disabled API.
Question 62
You significantly changed a complex Deployment Manager template and want to confirm that the dependencies of all defined resources are properly met before committing it to the project. You want the MOST rapid feedback on your changes. What should you do?
- A. Use granular logging statements within a Deployment Manager template authored in Python.
- B. Monitor activity of the Deployment Manager execution on the Stackdriver Logging page of the GCP Console.
- C. Execute the Deployment Manager template against a separate project with the same configuration, and monitor for failures.
- D. Execute the Deployment Manager template using the --preview option in the same project, and observe the state of interdependent resources.
Show answer and explanation ▾
Correct answer: D
Running the deployment with the --preview flag makes Deployment Manager expand the templates and build the full deployment plan, including the resource graph and dependencies, without actually creating or changing any resources. You can inspect the planned state of the interdependent resources right away, then commit the update or cancel the preview. This gives the most rapid feedback without affecting the project.
Why the other options are wrong:
- A. Adding logging statements to a Python template requires rewriting code and still does not validate resource dependencies before deployment.
- B. Watching Stackdriver Logging (now Cloud Logging) only shows results after a real deployment runs, which is slower and affects resources.
- C. Deploying into a separate project actually creates all the resources, which is slower and costlier than a preview.
Question 63
You have a Linux VM that must connect to Cloud SQL. You created a service account with the appropriate access rights. You want to make sure that the VM uses this service account instead of the default Compute Engine service account. What should you do?
- A. When creating the VM via the web console, specify the service account under the 'Identity and API Access' section.
- B. Download a JSON Private Key for the service account. On the Project Metadata, add that JSON as the value for the key compute-engine-service-account.
- C. Download a JSON Private Key for the service account. On the Custom Metadata of the VM, add that JSON as the value for the key compute-engine-service-account.
- D. Download a JSON Private Key for the service account. After creating the VM, ssh into the VM and save the JSON under ~/.gcloud/compute-engine-service-account.json.
Show answer and explanation ▾
Correct answer: A
A Compute Engine VM's identity is set by attaching a service account to the instance, which in the Console is done in the Identity and API access section when the VM is created. The VM then obtains short-lived credentials for that service account automatically from the metadata server, so applications on the VM use it instead of the default Compute Engine service account. No key files need to be downloaded or managed.
Why the other options are wrong:
- B. There is no compute-engine-service-account metadata key that changes the VM identity, and storing private keys in project metadata exposes them.
- C. Custom instance metadata does not attach a service account, and placing a private key there is insecure.
- D. A key file saved at that path is not used by Compute Engine to set the VM identity, and downloaded keys add security risk.
Question 64
You want to verify the IAM users and roles assigned within a GCP project named my-project. What should you do?
- A. Run gcloud iam roles list. Review the output section.
- B. Run gcloud iam service-accounts list. Review the output section.
- C. Navigate to the project and then to the IAM section in the GCP Console. Review the members and roles.
- D. Navigate to the project and then to the Roles section in the GCP Console. Review the roles and status.
Show answer and explanation ▾
Correct answer: C
The users and roles assigned within a project are defined by the project's IAM policy. The IAM page of the console, opened for my-project, lists every principal together with the roles granted to it, so reviewing the members and roles there shows exactly who has which access. The same information is available from gcloud projects get-iam-policy, but none of the listed commands return it.
Why the other options are wrong:
- A. gcloud iam roles list lists available role definitions, not which users hold them in the project.
- B. gcloud iam service-accounts list shows only service accounts, not user role assignments.
- D. The Roles page shows role definitions and their status, not the principals granted those roles.
Question 65
You need to manage a third-party application that will run on a Compute Engine instance. Other Compute Engine instances are already running with default configuration. Application installation files are hosted on Cloud Storage. You need to access these files from the new instance without allowing other virtual machines (VMs) to access these files. What should you do?
- A. Create the instance with the default Compute Engine service account. Grant the service account permissions on Cloud Storage.
- B. Create the instance with the default Compute Engine service account. Add metadata to the objects on Cloud Storage that matches the metadata on the new instance.
- C. Create a new service account and assign this service account to the new instance. Grant the service account permissions on Cloud Storage.
- D. Create a new service account and assign this service account to the new instance. Add metadata to the objects on Cloud Storage that matches the metadata on the new instance.
Show answer and explanation ▾
Correct answer: C
The existing VMs use the default configuration, which means they all run as the default Compute Engine service account. To let only the new instance read the installation files, create a new service account, attach it to the new instance, and grant it a role such as Storage Object Viewer on the bucket. Because no other VM uses this identity, only the new instance can access the files, which follows the principle of least privilege that Google recommends for workload identities.
Why the other options are wrong:
- A. Granting Cloud Storage access to the default Compute Engine service account would give the same access to every existing VM that uses that account.
- B. Object metadata does not control access in Cloud Storage, and the default service account is shared by the other VMs anyway.
- D. Matching metadata between objects and instances is not an access control mechanism; the new service account still needs IAM permissions on the bucket.
Question 66
You need to update a deployment in Deployment Manager without any resource downtime in the deployment. Which command should you use?
- A. gcloud deployment-manager deployments create --config <deployment-config-path>
- B. gcloud deployment-manager deployments update --config <deployment-config-path>
- C. gcloud deployment-manager resources create --config <deployment-config-path>
- D. gcloud deployment-manager resources update --config <deployment-config-path>
Show answer and explanation ▾
Correct answer: B
Changes to an existing deployment are applied with gcloud deployment-manager deployments update and the updated configuration file. Deployment Manager compares the new configuration with the current state and adds, updates or removes only the resources that changed, leaving the rest running, so the deployment is modified in place without taking the existing resources down.
Why the other options are wrong:
- A. The deployments create command creates a new deployment and fails when the deployment name already exists, so it does not update one.
- C. gcloud deployment-manager resources has no create command; resources are managed through their deployment.
- D. The resources command group only lists and describes resources; it cannot update a deployment from a configuration file.
Question 67
Your management has asked an external auditor to review all the resources in a specific project. The security team has enabled the Organization Policy called Domain Restricted Sharing on the organization node by specifying only your Cloud Identity domain.
You want the auditor to only be able to view, but not modify, the resources in that project. What should you do?
- A. Ask the auditor for their Google account, and give them the Viewer role on the project.
- B. Ask the auditor for their Google account, and give them the Security Reviewer role on the project.
- C. Create a temporary account for the auditor in Cloud Identity, and give that account the Viewer role on the project.
- D. Create a temporary account for the auditor in Cloud Identity, and give that account the Security Reviewer role on the project.
Show answer and explanation ▾
Correct answer: C
The Domain Restricted Sharing organization policy only allows IAM bindings for identities from the allowed Cloud Identity domain, so the auditor's external Google account cannot be granted any role. Creating a temporary account for the auditor in your Cloud Identity domain satisfies the constraint. The auditor must view all resources in the project without modifying them, and the basic Viewer role grants read-only access across resources in the project, which covers the full review.
Why the other options are wrong:
- A. Viewer is the right level of access, but Domain Restricted Sharing blocks granting any role to an account outside your Cloud Identity domain.
- B. An external Google account cannot be granted roles under Domain Restricted Sharing, and Security Reviewer lists resources and their IAM policies rather than giving view access to the resources themselves.
- D. The Cloud Identity account satisfies the policy, but Security Reviewer only lists resources and their IAM policies, so it does not let the auditor review the resources themselves.
Question 68
You are the project owner of a GCP project and want to delegate control to colleagues to manage buckets and files in Cloud Storage. You want to follow Google-recommended practices. Which IAM roles should you grant your colleagues?
- A. Project Editor
- B. Storage Admin
- C. Storage Object Admin
- D. Storage Object Creator
Show answer and explanation ▾
Correct answer: B
The colleagues must manage both buckets and the files inside them, and Google recommends granting the narrowest predefined role that covers the job. Storage Admin (roles/storage.admin) gives full control of buckets and objects, including creating, deleting and configuring buckets, without granting access to any other service in the project. That scope matches the requirement exactly, while the broader basic roles and the narrower object-only roles either give too much or too little.
Why the other options are wrong:
- A. Project Editor is a broad basic role that grants modify access to almost every resource in the project, violating least privilege.
- C. Storage Object Admin controls objects only and cannot create, delete or configure buckets.
- D. Storage Object Creator can only upload objects; it cannot manage buckets or read and delete files.
Question 69
You need a dynamic way of provisioning VMs on Compute Engine. The exact specifications will be in a dedicated configuration file. You want to follow Google's recommended practices. Which method should you use?
- A. Deployment Manager
- B. Cloud Composer
- C. Managed Instance Group
- D. Unmanaged Instance Group
Show answer and explanation ▾
Correct answer: A
Deployment Manager is Google Cloud's infrastructure-as-code service: you describe resources such as Compute Engine VMs and their exact specifications in a YAML configuration file, optionally with Jinja or Python templates, and Deployment Manager creates them. Templates and properties make the provisioning dynamic and repeatable, so a dedicated configuration file fully defines what is built. This follows Google's recommended practice of declarative, automated resource deployment.
Why the other options are wrong:
- B. Cloud Composer is a managed Apache Airflow service for orchestrating workflows, not a declarative tool for provisioning VMs from a configuration file.
- C. A managed instance group creates identical VMs from an instance template, but it is not a configuration-file-driven way of provisioning arbitrary infrastructure.
- D. An unmanaged instance group only groups existing, separately created VMs and provides no provisioning from a configuration file.
Question 70
You have a web application deployed as a managed instance group. You have a new version of the application to gradually deploy. Your web application is currently receiving live web traffic. You want to ensure that the available capacity does not decrease during the deployment. What should you do?
- A. Perform a rolling-action start-update with maxSurge set to 0 and maxUnavailable set to 1.
- B. Perform a rolling-action start-update with maxSurge set to 1 and maxUnavailable set to 0.
- C. Create a new managed instance group with an updated instance template. Add the group to the backend service for the load balancer. When all instances in the new managed instance group are healthy, delete the old managed instance group.
- D. Create a new instance template with the new application version. Update the existing managed instance group with the new instance template. Delete the instances in the managed instance group to allow the managed instance group to recreate the instance using the new instance template.
Show answer and explanation ▾
Correct answer: B
A rolling update on a managed instance group is controlled by maxSurge, the number of extra instances that may be created above the target size, and maxUnavailable, the number of instances that may be offline during the update. Setting maxSurge to 1 and maxUnavailable to 0 means the group creates a new instance with the updated template before removing an old one, so serving capacity never falls below the target size. The rollout proceeds gradually while live traffic continues to be handled at full capacity.
Why the other options are wrong:
- A. With maxSurge 0 and maxUnavailable 1, an instance is taken offline before its replacement exists, so capacity temporarily drops.
- C. Building a second group and swapping backends works but is manual, more complex and not a gradual rolling update of the existing group.
- D. Deleting instances so they are recreated removes capacity while the replacements boot, which violates the requirement.
Question 71
Your company is moving from an on-premises environment to Google Cloud. You have multiple development teams that use Cassandra environments as backend databases. They all need a development environment that is isolated from other Cassandra instances. You want to move to Google Cloud quickly and with minimal support effort. What should you do?
- A. 1. Build an instruction guide to install Cassandra on Google Cloud. 2. Make the instruction guide accessible to your developers.
- B. 1. Advise your developers to go to Cloud Marketplace. 2. Ask the developers to launch a Cassandra image for their development work.
- C. 1. Build a Cassandra Compute Engine instance and take a snapshot of it. 2. Use the snapshot to create instances for your developers.
- D. 1. Build a Cassandra Compute Engine instance and take a snapshot of it. 2. Upload the snapshot to Cloud Storage and make it accessible to your developers. 3. Build instructions to create a Compute Engine instance from the snapshot so that developers can do it themselves.
Show answer and explanation ▾
Correct answer: B
Cloud Marketplace provides ready-made Cassandra solutions that deploy preconfigured Compute Engine instances in a few clicks. Each developer can launch their own deployment in their project, giving every team an isolated Cassandra environment. Because the image and deployment are packaged and maintained by the solution provider, there are no installation guides or custom images to build and keep current, which meets the goals of moving quickly with minimal support effort.
Why the other options are wrong:
- A. An installation guide makes every developer install and configure Cassandra by hand, which is slow, error prone and generates support work.
- C. Building a custom instance and snapshot means you must create, patch and maintain the image yourself, which adds support effort.
- D. Snapshots are not uploaded to Cloud Storage for reuse, and this approach adds image maintenance and self-service instructions, increasing effort.
Question 72
You are asked to set up application performance monitoring on Google Cloud projects A, B, and C as a single pane of glass. You want to monitor CPU, memory, and disk. What should you do?
- A. Enable API and then share charts from project A, B, and C.
- B. Enable API and then give the metrics.reader role to projects A, B, and C.
- C. Enable API and then use default dashboards to view all projects in sequence.
- D. Enable API, create a workspace under project A, and then add projects B and C.
Show answer and explanation ▾
Correct answer: D
A single pane of glass across projects is achieved by creating a Cloud Monitoring Workspace (now called a metrics scope) in a scoping project and adding the other projects to it. With the workspace hosted in project A and projects B and C added, dashboards, charts and alerts can show CPU, memory and disk metrics for all three projects together. Memory and disk utilization metrics require the monitoring agent (Ops Agent) on the VMs.
Why the other options are wrong:
- A. Sharing individual charts from each project does not combine the metrics in one place and still leaves three separate views.
- B. Granting a metrics reader role to projects does not aggregate monitoring data; roles are granted to identities, not to projects.
- C. Viewing each project's default dashboards in turn is the opposite of a single pane of glass.
Question 73
You need to set up a policy so that videos stored in a specific Cloud Storage Regional bucket are moved to Coldline after 90 days, and then deleted after one year from their creation. How should you set up the policy?
- A. Use Cloud Storage Object Lifecycle Management using Age conditions with SetStorageClass and Delete actions. Set the SetStorageClass action to 90 days and the Delete action to 275 days (365 - 90)
- B. Use Cloud Storage Object Lifecycle Management using Age conditions with SetStorageClass and Delete actions. Set the SetStorageClass action to 90 days and the Delete action to 365 days.
- C. Use gsutil rewrite and set the Delete action to 275 days (365-90).
- D. Use gsutil rewrite and set the Delete action to 365 days.
Show answer and explanation ▾
Correct answer: B
Object Lifecycle Management automates storage class changes and deletion through rules on the bucket. The Age condition is measured from the object's creation time, so one rule with a SetStorageClass action to Coldline at an age of 90 days and another rule with a Delete action at an age of 365 days moves the videos after 90 days and deletes them one year after creation. Changing the storage class does not reset the object's age.
Why the other options are wrong:
- A. Age counts from creation, not from the storage class change, so a Delete age of 275 days would remove the videos after about nine months.
- C. gsutil rewrite changes the storage class of existing objects once; it has no Delete action and does not create an automated policy.
- D. gsutil rewrite cannot schedule deletion, so it cannot implement the one-year delete requirement as a policy.
Question 74
You need to select and configure compute resources for a set of batch processing jobs. These jobs take around 2 hours to complete and are run nightly. You want to minimize service costs. What should you do?
- A. Select Google Kubernetes Engine. Use a single-node cluster with a small instance type.
- B. Select Google Kubernetes Engine. Use a three-node cluster with micro instance types.
- C. Select Compute Engine. Use preemptible VM instances of the appropriate standard machine type.
- D. Select Compute Engine. Use VM instance types that support micro bursting.
Show answer and explanation ▾
Correct answer: C
Nightly batch jobs that take about two hours are fault-tolerant, short-running workloads, which is exactly what preemptible VMs are designed for. Preemptible instances cost far less than standard VMs and run for at most 24 hours, well beyond the two-hour job length. Choosing a standard machine type sized for the jobs on Compute Engine keeps performance adequate while minimizing cost, and jobs can be rerun if an instance is preempted.
Why the other options are wrong:
- A. A small always-on GKE cluster costs more than short-lived preemptible VMs and may be underpowered for the jobs.
- B. Micro instances lack the capacity for heavy batch work, and a running GKE cluster adds ongoing cost.
- D. Shared-core bursting machines suit light, intermittent loads, not sustained two-hour batch processing.
Question 75
Your company wants to standardize the creation and management of multiple Google Cloud resources using Infrastructure as Code. You want to minimize the amount of repetitive code needed to manage the environment. What should you do?
- A. Develop templates for the environment using Cloud Deployment Manager.
- B. Use curl in a terminal to send a REST request to the relevant Google API for each individual resource.
- C. Use the Cloud Console interface to provision and manage all related resources.
- D. Create a bash script that contains all requirement steps as gcloud commands.
Show answer and explanation ▾
Correct answer: A
The requirements are Infrastructure as Code for many resources and minimal repetitive code. Cloud Deployment Manager lets you describe resources declaratively in configuration files and build reusable templates in Jinja or Python. A template defines a resource pattern once and accepts properties as parameters, so the same template can create many similar resources or whole environments. Deployments can then be previewed, updated and deleted as a unit, which standardizes how resources are created and managed.
Why the other options are wrong:
- B. Sending individual REST requests with curl is manual and repetitive, and it provides no declarative, reusable definition of the environment.
- C. Provisioning through the console is manual, not Infrastructure as Code, and cannot be repeated consistently.
- D. A bash script of gcloud commands is imperative and repetitive, and it does not manage resources as a unit or handle updates cleanly.
Question 76
You have designed a solution on Google Cloud that uses multiple Google Cloud products. Your company has asked you to estimate the costs of the solution. You need to provide estimates for the monthly total cost. What should you do?
- A. For each Google Cloud product in the solution, review the pricing details on the product's pricing page. Use the pricing calculator to total the monthly costs for each Google Cloud product.
- B. For each Google Cloud product in the solution, review the pricing details on the product's pricing page. Create a Google Sheet that summarizes the expected monthly costs for each product.
- C. Provision the solution on Google Cloud. Leave the solution provisioned for 1 week. Navigate to the Billing Report page in the Cloud Console. Multiply the 1 week cost to determine the monthly costs.
- D. Provision the solution on Google Cloud. Leave the solution provisioned for 1 week. Use Cloud Monitoring to determine the provisioned and used resource amounts. Multiply the 1 week cost to determine the monthly costs.
Show answer and explanation ▾
Correct answer: A
The requirement is an estimate of the monthly total before committing spend. Each product's pricing page explains the billing dimensions (instance hours, storage, egress, operations), and the Google Cloud Pricing Calculator lets you enter the expected usage for every product in the design and produces a combined monthly estimate. This gives an accurate, shareable figure without provisioning anything, so no money is spent just to learn what the solution will cost, and the estimate can be adjusted easily as the design changes.
Why the other options are wrong:
- B. A hand-built spreadsheet duplicates what the Pricing Calculator already does and is more error-prone, since you must work out every pricing dimension and tier yourself.
- C. Provisioning the full solution for a week incurs real charges just to estimate cost, and one week of usage may not reflect a typical month.
- D. Cloud Monitoring reports resource usage, not costs, and running the solution for a week to estimate spend wastes money before the design is approved.
Question 77
You have a large 5-TB AVRO file stored in a Cloud Storage bucket. Your analysts are proficient only in SQL and need access to the data stored in this file. You want to find a cost-effective way to complete their request as soon as possible. What should you do?
- A. Load data in Cloud Datastore and run a SQL query against it.
- B. Create a BigQuery table and load data in BigQuery. Run a SQL query on this table and drop this table after you complete your request.
- C. Create external tables in BigQuery that point to Cloud Storage buckets and run a SQL query on these external tables to complete your request.
- D. Create a Hadoop cluster and copy the AVRO file to HDFS by compressing it. Load the file in a hive table and provide access to your analysts so that they can run SQL queries.
Show answer and explanation ▾
Correct answer: C
BigQuery external tables can query Avro files directly in Cloud Storage using standard SQL, which suits analysts who only know SQL. Because the data is not loaded, there is no wait for a 5 TB load job and no BigQuery storage charge, so access is available quickly and cheaply. Analysts simply run queries against the external table definition, paying only for the data processed.
Why the other options are wrong:
- A. Datastore is a NoSQL document database, not a SQL analytics engine, and loading 5 TB into it is slow and costly.
- B. Loading 5 TB into BigQuery takes time and adds storage cost, while an external table gives SQL access immediately.
- D. Building a Hadoop cluster and Hive tables adds significant setup time, cost and administration for a simple SQL access need.
Question 78
You are working with a user to set up an application in a new VPC behind a firewall. The user is concerned about data egress. You want to configure the fewest open egress ports. What should you do?
- A. Set up a low-priority (65534) rule that blocks all egress and a high-priority rule (1000) that allows only the appropriate ports.
- B. Set up a high-priority (1000) rule that pairs both ingress and egress ports.
- C. Set up a high-priority (1000) rule that blocks all egress and a low-priority (65534) rule that allows only the appropriate ports.
- D. Set up a high-priority (1000) rule to allow the appropriate ports.
Show answer and explanation ▾
Correct answer: A
VPC firewall rules are evaluated by priority, and a lower number means a higher priority. To keep egress as tight as possible, create a broad deny-all egress rule with a very low priority such as 65534, then add a higher-priority rule (for example 1000) that allows only the specific ports the application needs. Traffic on the approved ports matches the allow rule first, and everything else falls through to the deny rule. This overrides the implied allow-all egress rule that every VPC has, so only the required ports are open.
Why the other options are wrong:
- B. A single firewall rule applies to one direction only, ingress or egress, so it cannot pair both, and it does nothing to block the remaining egress ports.
- C. With the deny-all rule at the higher priority (1000), it is evaluated first and blocks all egress, so the lower-priority allow rule never takes effect.
- D. An allow rule alone does not restrict anything, because the implied allow-all egress rule already permits all outbound traffic on every port.
Question 79
You created a Google Cloud Platform project with an App Engine application inside the project. You initially configured the application to be served from the us-central region. Now you want the application to be served from the asia-northeast1 region. What should you do?
- A. Change the default region property setting in the existing GCP project to asia-northeast1.
- B. Change the region property setting in the existing App Engine application from us-central to asia-northeast1.
- C. Create a second App Engine application in the existing GCP project and specify asia-northeast1 as the region to serve your application.
- D. Create a new GCP project and create an App Engine application inside this new project. Specify asia-northeast1 as the region to serve your application.
Show answer and explanation ▾
Correct answer: D
An App Engine application's region is chosen when the application is created and cannot be changed afterward, and each project can contain only one App Engine application. Serving the application from asia-northeast1 therefore requires a new project with a new App Engine application created in that region, followed by deploying the code there.
Why the other options are wrong:
- A. The project's default region setting does not move an existing App Engine application.
- B. The App Engine application's region is permanent once set and cannot be edited.
- C. A project can host only one App Engine application, so a second one cannot be created.
Question 80
You have a batch workload that runs every night and uses a large number of virtual machines (VMs). It is fault-tolerant and can tolerate some of the VMs being terminated. The current cost of VMs is too high. What should you do?
- A. Run a test using simulated maintenance events. If the test is successful, use preemptible N1 Standard VMs when running future jobs.
- B. Run a test using simulated maintenance events. If the test is successful, use N1 Standard VMs when running future jobs.
- C. Run a test using a managed instance group. If the test is successful, use N1 Standard VMs in the managed instance group when running future jobs.
- D. Run a test using N1 standard VMs instead of N2. If the test is successful, use N1 Standard VMs when running future jobs.
Show answer and explanation ▾
Correct answer: A
The batch workload is fault-tolerant and can lose some VMs, and the goal is lower cost. Preemptible VMs (now Spot VMs) cost far less than standard VMs because Compute Engine can reclaim them at any time. Running a test with simulated maintenance events confirms the job copes with instances being stopped, and once it passes, switching future runs to preemptible N1 Standard VMs gives large savings with acceptable risk.
Why the other options are wrong:
- B. Testing for interruptions and then using standard N1 VMs gains nothing, because standard pricing keeps the cost high.
- C. A managed instance group of standard VMs adds automation but does not lower the price of each VM, so costs stay high.
- D. Switching from N2 to N1 standard VMs may change price slightly but saves far less than preemptible pricing for an interruption-tolerant job.
Question 81
You are running an application on multiple virtual machines within a managed instance group and have autoscaling enabled. The autoscaling policy is configured so that additional instances are added to the group if the CPU utilization of instances goes above 80%. VMs are added until the instance group reaches its maximum limit of five VMs or until CPU utilization of instances lowers to 80%. The initial delay for HTTP health checks against the instances is set to 30 seconds. The virtual machine instances take around three minutes to become available for users. You observe that when the instance group autoscales, it adds more instances than necessary to support the levels of end-user traffic. You want to properly maintain instance group sizes when autoscaling. What should you do?
- A. Set the maximum number of instances to 1.
- B. Decrease the maximum number of instances to 3.
- C. Use a TCP health check instead of an HTTP health check.
- D. Increase the initial delay of the HTTP health check to 200 seconds.
Show answer and explanation ▾
Correct answer: D
The VMs need about three minutes before they can serve users, but the initial delay is only 30 seconds. New instances are therefore judged while still starting up, and because they are not yet sharing the load, CPU on the group stays high and more VMs are added than the traffic needs. Raising the initial delay to 200 seconds, longer than the startup time, gives each new instance time to become ready before it is evaluated, so the group size settles at the level the traffic requires.
Why the other options are wrong:
- A. Capping the group at one instance removes autoscaling and cannot handle higher traffic.
- B. Lowering the maximum only hides the symptom and may leave too few VMs during real peaks.
- C. Changing the health check protocol does not account for the three-minute startup time causing the overscaling.
Question 82
You are building a product on top of Google Kubernetes Engine (GKE). You have a single GKE cluster. For each of your customers, a Pod is running in that cluster, and your customers can run arbitrary code inside their Pod. You want to maximize the isolation between your customers' Pods. What should you do?
- A. Use Binary Authorization and whitelist only the container images used by your customers' Pods.
- B. Use the Container Analysis API to detect vulnerabilities in the containers used by your customers' Pods.
- C. Create a GKE node pool with a sandbox type configured to gvisor. Add the parameter runtimeClassName: gvisor to the specification of your customers' Pods.
- D. Use the cos_containerd image for your GKE nodes. Add a nodeSelector with the value cloud.google.com/gke-os-distribution: cos_containerd to the specification of your customers' Pods.
Show answer and explanation ▾
Correct answer: C
Customers can run arbitrary code, so the goal is a stronger boundary between untrusted workloads and the host kernel. GKE Sandbox uses gVisor, a user-space kernel that intercepts system calls so a container cannot interact directly with the node's kernel. You create a node pool with the sandbox type set to gvisor and add runtimeClassName: gvisor to each customer Pod specification, which schedules those Pods on the sandboxed nodes and gives the strongest isolation of the options.
Why the other options are wrong:
- A. Binary Authorization controls which images can be deployed, but it does not isolate code running inside a Pod from other Pods.
- B. Vulnerability scanning reports known issues in images; it does not isolate arbitrary code customers run at runtime.
- D. The cos_containerd image is a standard node OS and runtime, which provides no extra isolation between Pods.
Question 83
You are developing a new application and are looking for a Jenkins installation to build and deploy your source code. You want to automate the installation as quickly and easily as possible. What should you do?
- A. Deploy Jenkins through the Google Cloud Marketplace.
- B. Create a new Compute Engine instance. Run the Jenkins executable.
- C. Create a new Kubernetes Engine cluster. Create a deployment for the Jenkins image.
- D. Create an instance template with the Jenkins executable. Create a managed instance group with this template.
Show answer and explanation ▾
Correct answer: A
The requirement is to automate a Jenkins installation as quickly and easily as possible. Google Cloud Marketplace offers prepackaged Jenkins solutions that deploy in a few clicks, creating the VM, installing and configuring Jenkins and setting up the needed firewall rules automatically. This removes the manual installation and configuration steps, so you can start building and deploying your source code almost immediately.
Why the other options are wrong:
- B. Creating a VM and installing Jenkins manually is slower and requires configuration work that the Marketplace package already automates.
- C. Building a GKE cluster and writing a Deployment for Jenkins involves far more setup than a one-click Marketplace deployment.
- D. Preparing an instance template and managed instance group adds steps and complexity that a single Jenkins server does not need.
Question 84
You deployed an App Engine application using gcloud app deploy, but it did not deploy to the intended project. You want to find out why this happened and where the application deployed. What should you do?
- A. Check the app.yaml file for your application and check project settings.
- B. Check the web-application.xml file for your application and check project settings.
- C. Go to Deployment Manager and review settings for deployment of applications.
- D. Go to Cloud Shell and run gcloud config list to review the Google Cloud configuration used for deployment.
Show answer and explanation ▾
Correct answer: D
gcloud app deploy deploys to the project set in the active gcloud configuration unless the --project flag is given. Running gcloud config list shows the active account and core project, revealing which project the deployment actually targeted and why it went there. Correcting the project property, or passing --project, then sends future deployments to the intended project.
Why the other options are wrong:
- A. app.yaml defines runtime and service settings; the target project comes from the gcloud configuration, not this file.
- B. A web-application.xml file does not set the deployment project for gcloud app deploy.
- C. gcloud app deploy does not use Deployment Manager, so it holds no record of this deployment.
Question 85
You need to manage a Cloud Spanner instance for best query performance. Your instance in production runs in a single Google Cloud region. You need to improve performance in the shortest amount of time. You want to follow Google best practices for service configuration. What should you do?
- A. Create an alert in Cloud Monitoring to alert when the percentage of high priority CPU utilization reaches 45%. If you exceed this threshold, add nodes to your instance.
- B. Create an alert in Cloud Monitoring to alert when the percentage of high priority CPU utilization reaches 45%. Use database query statistics to identify queries that result in high CPU usage, and then rewrite those queries to optimize their resource usage.
- C. Create an alert in Cloud Monitoring to alert when the percentage of high priority CPU utilization reaches 65%. If you exceed this threshold, add nodes to your instance.
- D. Create an alert in Cloud Monitoring to alert when the percentage of high priority CPU utilization reaches 65%. Use database query statistics to identify queries that result in high CPU usage, and then rewrite those queries to optimize their resource usage.
Show answer and explanation ▾
Correct answer: C
Google recommends keeping high priority CPU utilization below 65 percent for a regional (single-region) Cloud Spanner instance; the stricter 45 percent guidance applies to multi-region configurations. An alert at 65 percent therefore matches best practice for this instance. When the threshold is exceeded, adding nodes (compute capacity) is the fastest fix, because capacity increases take effect quickly without code changes, which satisfies the requirement to improve performance in the shortest amount of time.
Why the other options are wrong:
- A. The 45 percent threshold is the guidance for multi-region instances; for a single-region instance it triggers scaling earlier than Google recommends.
- B. The 45 percent threshold suits multi-region instances, and rewriting queries takes far longer than adding nodes.
- D. The threshold is correct for a regional instance, but analyzing and rewriting queries is slower than adding nodes when performance must improve quickly.
Question 86
You are using Deployment Manager to create a Google Kubernetes Engine cluster. Using the same Deployment Manager deployment, you also want to create a DaemonSet in the kube-system namespace of the cluster. You want a solution that uses the fewest possible services. What should you do?
- A. Add the cluster's API as a new Type Provider in Deployment Manager, and use the new type to create the DaemonSet.
- B. Use the Deployment Manager Runtime Configurator to create a new Config resource that contains the DaemonSet definition.
- C. With Deployment Manager, create a Compute Engine instance with a startup script that uses kubectl to create the DaemonSet.
- D. In the cluster's definition in Deployment Manager, add a metadata that has kube-system as key and the DaemonSet manifest as value.
Show answer and explanation ▾
Correct answer: A
Deployment Manager can manage resources from any API that exposes an OpenAPI or Google Discovery document by registering it as a Type Provider. Adding the GKE cluster's Kubernetes API as a type provider lets the same deployment create Kubernetes objects, such as a DaemonSet in the kube-system namespace, as ordinary Deployment Manager resources alongside the cluster. This keeps everything inside Deployment Manager without introducing additional services, meeting the requirement to use the fewest possible services.
Why the other options are wrong:
- B. Runtime Configurator stores configuration values and variables; it does not create Kubernetes objects in a cluster.
- C. An extra Compute Engine instance running kubectl adds another service and resource to manage, contradicting the fewest-services requirement.
- D. Cluster metadata does not deploy Kubernetes manifests, so a DaemonSet in metadata is never created.
Question 87
You built an application on your development laptop that uses Google Cloud services.
Your application uses Application Default Credentials for authentication and works fine on your development laptop.
You want to migrate this application to a Compute Engine virtual machine (VM) and set up authentication using Google-recommended practices and minimal changes.
What should you do?
- A. Assign appropriate access for Google services to the service account used by the Compute Engine VM.
- B. Create a service account with appropriate access for Google services, and configure the application to use this account.
- C. Store credentials for service accounts with appropriate access for Google services in a config file, and deploy this config file with your application.
- D. Store credentials for your user account with appropriate access for Google services in a config file, and deploy this config file with your application.
Show answer and explanation ▾
Correct answer: A
Application Default Credentials automatically use the attached service account when code runs on Compute Engine, retrieving short-lived tokens from the metadata server. Because the application already uses ADC, no code change is needed: you only give the VM's service account the IAM roles the application requires. This avoids downloading or distributing key files, which is the Google-recommended way to authenticate workloads running on Google Cloud and the option with the fewest changes.
Why the other options are wrong:
- B. Configuring the application to use a specific account implies code or configuration changes, often a key file, when ADC already picks up the VM's attached service account.
- C. Deploying service account key files with the application creates long-lived secrets that can leak, which Google advises against when an attached service account is available.
- D. Shipping user account credentials ties the application to a personal identity and exposes those credentials, which is poor security practice.
Question 88
You are building an application that stores relational data from users. Users across the globe will use this application. Your CTO is concerned about the scaling requirements because the size of the user base is unknown. You need to implement a database solution that can scale with your user growth with minimum configuration changes. Which storage solution should you use?
- A. Cloud SQL
- B. Cloud Spanner
- C. Cloud Firestore
- D. Cloud Datastore
Show answer and explanation ▾
Correct answer: B
The data is relational, users are distributed across the globe, and growth is unpredictable. Cloud Spanner is a fully managed relational database with SQL and strong consistency that scales horizontally and can be deployed in multi-region configurations. Capacity is increased by adding compute capacity (nodes or processing units) without schema changes, sharding or application changes, so it grows with the user base with minimal configuration effort. That combination of relational semantics and global horizontal scale makes it the right choice.
Why the other options are wrong:
- A. Cloud SQL is relational but scales mainly vertically within a region, so handling unknown global growth would require significant reconfiguration.
- C. Cloud Firestore is a NoSQL document database, not a relational store for the application's relational data.
- D. Cloud Datastore is a NoSQL document database and does not provide the relational model the application needs.
Question 89
You are deploying an application to a Compute Engine VM in a managed instance group. The application must be running at all times, but only a single instance of the VM should run per GCP project. How should you configure the instance group?
- A. Set autoscaling to On, set the minimum number of instances to 1, and then set the maximum number of instances to 1.
- B. Set autoscaling to Off, set the minimum number of instances to 1, and then set the maximum number of instances to 1.
- C. Set autoscaling to On, set the minimum number of instances to 1, and then set the maximum number of instances to 2.
- D. Set autoscaling to Off, set the minimum number of instances to 1, and then set the maximum number of instances to 2.
Show answer and explanation ▾
Correct answer: B
The application must always be running but only one VM may exist per project, so the group should hold exactly one instance with no scaling. Turning autoscaling off and fixing the group at one instance keeps the size constant, while the managed instance group still recreates the VM if it fails or is deleted, keeping the application available. With no autoscaler, there is nothing that could ever add a second instance.
Why the other options are wrong:
- A. Turning autoscaling on with a 1 to 1 range adds an autoscaler that serves no purpose when the size can never change.
- C. A maximum of 2 lets the autoscaler run a second VM, breaking the single-instance requirement.
- D. Allowing a maximum of 2 permits a second instance, which violates the one-VM-per-project requirement.
Question 90
You have created a code snippet that should be triggered whenever a new file is uploaded to a Cloud Storage bucket. You want to deploy this code snippet. What should you do?
- A. Use App Engine and configure Cloud Scheduler to trigger the application using Pub/Sub.
- B. Use Cloud Functions and configure the bucket as a trigger resource.
- C. Use Google Kubernetes Engine and configure a CronJob to trigger the application using Pub/Sub.
- D. Use Dataflow as a batch job, and configure the bucket as a data source.
Show answer and explanation ▾
Correct answer: B
The requirement is to run a small piece of code each time a new file is uploaded to a bucket. Cloud Functions is built for this event-driven pattern: you deploy the function with the bucket as its trigger resource and the object finalize event as the trigger type, and Google invokes the function automatically for every new object. There are no servers or schedules to manage, and you pay only when the function runs, which makes it the simplest way to deploy a code snippet that reacts to uploads.
Why the other options are wrong:
- A. Cloud Scheduler fires on a fixed timetable, not when a file is uploaded, and App Engine adds unnecessary hosting for a small snippet.
- C. A Kubernetes CronJob runs on a schedule rather than on upload events and requires running and managing a GKE cluster.
- D. A Dataflow batch job processes a data source when it is launched; it is not an event trigger that runs on each new upload.
Question 91
Your company publishes large files on an Apache web server that runs on a Compute Engine instance. The Apache web server is not the only application running in the project. You want to receive an email when the egress network costs for the server exceed 100 dollars for the current month as measured by Google Cloud.
What should you do?
- A. Set up a budget alert on the project with an amount of 100 dollars, a threshold of 100%, and notification type of "email."
- B. Set up a budget alert on the billing account with an amount of 100 dollars, a threshold of 100%, and notification type of "email."
- C. Export the billing data to BigQuery. Create a Cloud Function that uses BigQuery to sum the egress network costs of the exported billing data for the Apache web server for the current month and sends an email if it is over 100 dollars. Schedule the Cloud Function using Cloud Scheduler to run hourly.
- D. Use the Cloud Logging Agent to export the Apache web server logs to Cloud Logging. Create a Cloud Function that uses BigQuery to parse the HTTP response log data in Cloud Logging for the current month and sends an email if the size of all HTTP responses, multiplied by current Google Cloud egress prices, totals over 100 dollars. Schedule the Cloud Function using Cloud Scheduler to run hourly.
Show answer and explanation ▾
Correct answer: C
The alert must track only the egress cost of one server, while other applications share the project. Exporting Cloud Billing data to BigQuery provides detailed cost rows by service, SKU and resource, so a query can sum the network egress charges attributed to the Apache instance for the current month as measured by Google Cloud. A Cloud Function that runs this query hourly through Cloud Scheduler can send an email as soon as the total passes 100 dollars, which a standard budget cannot do at that granularity.
Why the other options are wrong:
- A. A project budget counts all costs in the project, including other applications, so it cannot isolate the egress cost of the Apache server.
- B. A billing account budget covers every project on the account, which is even broader and cannot single out one server's egress charges.
- D. Estimating cost from HTTP response sizes and list prices is an approximation, not the cost measured by Google Cloud, and adds complex log processing.
Question 92
You have one GCP account running in your default region and zone and another account running in a non-default region and zone. You want to start a new Compute Engine instance in these two Google Cloud Platform accounts using the command line interface. What should you do?
- A. Create two configurations using gcloud config configurations create [NAME]. Run gcloud config configurations activate [NAME] to switch between accounts when running the commands to start the Compute Engine instances.
- B. Create two configurations using gcloud config configurations create [NAME]. Run gcloud configurations list to start the Compute Engine instances.
- C. Activate two configurations using gcloud configurations activate [NAME]. Run gcloud config list to start the Compute Engine instances.
- D. Activate two configurations using gcloud configurations activate [NAME]. Run gcloud configurations list to start the Compute Engine instances.
Show answer and explanation ▾
Correct answer: A
A named gcloud configuration stores a set of properties such as the account, project, region and zone. Creating one configuration for each account with gcloud config configurations create, setting its properties, and then running gcloud config configurations activate before each command switches the active account and location, so each gcloud compute instances create command runs against the correct account with the right region and zone.
Why the other options are wrong:
- B. gcloud configurations list is not a valid command, and listing configurations would not start any instances.
- C. The activate command belongs under gcloud config configurations, only one configuration can be active at a time, and gcloud config list does not start instances.
- D. gcloud configurations is not a valid command group, and listing configurations does not start Compute Engine instances.
Question 93
You have a workload running on Compute Engine that is critical to your business. You want to ensure that the data on the boot disk of this workload is backed up regularly. You need to be able to restore a backup as quickly as possible in case of disaster. You also want older backups to be cleaned automatically to save on cost. You want to follow Google-recommended practices. What should you do?
- A. Create a Cloud Function to create an instance template.
- B. Create a snapshot schedule for the disk using the desired interval.
- C. Create a cron job to create a new disk from the disk using gcloud.
- D. Create a Cloud Task to create an image and export it to Cloud Storage.
Show answer and explanation ▾
Correct answer: B
A snapshot schedule is a resource policy attached to a persistent disk that takes snapshots automatically at the chosen interval and applies a retention policy that deletes snapshots older than a set number of days. Snapshots are incremental, so they are cost-efficient, and a new boot disk or instance can be created from a snapshot quickly during a disaster. This managed feature meets the requirements for regular backups, fast restore and automatic cleanup without any custom code, which is Google's recommended approach.
Why the other options are wrong:
- A. An instance template only stores VM configuration such as machine type and image. It does not back up the data on the existing boot disk.
- C. A cron job that clones disks is custom tooling to maintain, full disk copies cost more than incremental snapshots, and nothing removes old copies automatically.
- D. Cloud Tasks queues HTTP work items and is not a backup scheduler. Exporting images to Cloud Storage is slow to restore and has no built-in retention.
Question 94
You are the organization and billing administrator for your company. The engineering team has the Project Creator role on the organization. You do not want the engineering team to be able to link projects to the billing account. Only the finance team should be able to link a project to a billing account, but they should not be able to make any other changes to projects. What should you do?
- A. Assign the finance team only the Billing Account User role on the billing account.
- B. Assign the engineering team only the Billing Account User role on the billing account.
- C. Assign the finance team the Billing Account User role on the billing account and the Project Billing Manager role on the organization.
- D. Assign the engineering team the Billing Account User role on the billing account and the Project Billing Manager role on the organization.
Show answer and explanation ▾
Correct answer: C
Linking a project to a billing account requires permission on both sides: the Billing Account User role (roles/billing.user) on the billing account and permission to change the project's billing, which the Project Billing Manager role (roles/billing.projectManager) grants without any other project modification rights. Granting the finance team both roles, with Project Billing Manager at the organization level, lets them link any project while preventing other project changes. The engineering team receives no billing roles, so its Project Creator role alone cannot link projects to the billing account.
Why the other options are wrong:
- A. Billing Account User alone covers only the billing account side; the finance team would lack permission on the projects to change their billing.
- B. Granting the engineering team a billing role does the opposite of the requirement by enabling them to link projects.
- D. This gives the engineering team full ability to link projects, which is exactly what must be prevented.
That was 94 of 375.
The full Google Associate Cloud Engineer pack has all 375 questions, each with the answer, the explanation and why the other options are wrong, plus a questions-only copy for timed runs. $39.00 USD, paid once, with free monthly updates and a pass-or-your-money-back guarantee.
Get the full packNot ready yet? Get the first 187 free as one PDF by email. Get the Half Pack →
About the Google Associate Cloud Engineer exam
These 94 Google Associate Cloud Engineer questions are free here. The free PDF, the first half of the pack, has 187 free by email, and the full pack has all 375. Each step includes the one before it.
They are drawn from the same bank as the full Google Associate Cloud Engineer pack, which has 375 questions in total.
Get the full Google Associate Cloud Engineer question bank (375 questions) →