Associate Cloud Engineer (Google Cloud) — Free Practice Questions
10 free sample questions from a bank of 276, with the correct answers and explanations. No signup required — start practising right now.
1Every 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?
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.
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.
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 ג€compute.osAdminLoginג€ role to the Google group corresponding to this team.
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.
Answer: C
The short version
C — OS Login with individual keys plus group admin role. Shared private keys destroy auditability because every access looks identical. Per-user keys bound to Google accounts with the compute admin role give efficient deployment plus per-user traceability.
Key concepts in this question
OS Login model: public keys stored on Google accounts propagate to instances without manual key distribution.
compute dot osAdminLogin role: grants administrative SSH access when bound to the team group.
Auditability requirement: security must determine who accessed a given instance from login records.
Operational efficiency: central group binding avoids pushing keys per instance with config tools.
Shared-key anti-pattern: one private key shared across a team prevents attribution and complicates rotation.
Why C is correct
Each team member generates their own SSH pair and attaches the public key to their Google account, so OS Login projects that identity onto every Compute Engine instance automatically. Granting the compute dot osAdminLogin role to the Google group for the team confers admin rights centrally, with no per-instance file pushes. Every SSH session is then tied to an individual Google identity in audit logs, satisfying both efficient credential deployment and who-accessed-what traceability together.
Why the others are wrong
A. Shares one private key across the team, which is inefficient to rotate and makes attribution impossible.
B. Uses per-user keys but deploys them with a config tool per instance, adding operational overhead without central IAM control.
D. Shares one private key as a project-wide key, which repeats the attribution failure and widens blast radius.
ACE exam tip
Need per-user SSH audit on GCP, think OS Login plus osAdminLogin on the group.
2You 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?
0.0.0.0/0
10.0.0.0/8
172.16.0.0/12
192.168.0.0/16
Answer: B
The short version
B — 10 dot 0 dot 0 dot 0 slash 8 is the largest single-subnet range. Among the listed private ranges, the slash 8 block holds far more addresses than slash 12 or slash 16. For one subnet as large as possible, that block maximizes host count.
Key concepts in this question
RFC 1918 private space: 10 slash 8, 172 dot 16 slash 12, and 192 dot 168 slash 16 are the private options.
CIDR size math: smaller prefix number means a larger address count available to the subnet.
Single-subnet constraint: one subnet must contain the whole range, so the largest listed CIDR wins.
Zero slash zero exclusion: 0 dot 0 dot 0 dot 0 slash 0 is not a valid private VPC subnet choice.
GCP VPC practice: custom VPC subnets draw from private ranges sized to growth needs.
Why B is correct
A slash 8 network provides about 16 million host addresses, while slash 12 provides about 1 million and slash 16 provides about 65 thousand. Since the requirement is a single subnet as large as possible, the 10 dot 0 dot 0 dot 0 slash 8 range in settled key B offers the greatest capacity of the valid choices. It remains within private address space and is accepted for a custom VPC subnet, unlike the unspecified 0 dot 0 dot 0 dot 0 slash 0 option.
Why the others are wrong
A. Uses a default-route CIDR that is not a valid private subnet range for this purpose.
C. Offers a medium private range but is roughly sixteen times smaller than the slash 8 block.
D. Offers the smallest listed private range, far below the slash 8 capacity.
ACE exam tip
Largest subnet question, pick the smallest prefix number among valid private ranges.
3You 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?
Select Cloud SQL (MySQL). Verify that the enable binary logging option is selected.
Select Cloud SQL (MySQL). Select the create failover replicas option.
Select Cloud Spanner. Set up your instance with 2 nodes.
Select Cloud Spanner. Set up your instance as multi-regional.
Answer: A
The short version
A — Cloud SQL MySQL with binary logging for cheap point-in-time recovery. Small single-location operational data does not need Spanner scale. MySQL binary logs enable replay to any second at low cost.
Key concepts in this question
Cloud SQL fit: managed MySQL suits small relational operational workloads in one location.
Binary logging: log of changes enables point-in-time recovery by replaying transactions.
Cost-effectiveness: single-zone MySQL costs far less than multi-node Spanner for tiny data.
Spanner overkill: global distribution and multi-node minimums serve huge scale, not this case.
Failover vs recovery: failover replicas aid availability, not historical point-in-time restore.
Why A is correct
The workload is small, relational, and confined to one geography with a point-in-time recovery need, which maps directly to Cloud SQL for MySQL with binary logging enabled. Binary logs record every change so the instance can be restored to a precise moment by replaying logs onto a base backup. Settled key A therefore meets both cost and recovery goals, while Spanner options add unnecessary nodes and multi-regional expense and failover replicas alone do not give temporal replay.
Why the others are wrong
B. Failover replicas improve availability but do not by themselves provide point-in-time replay.
C. Two-node Spanner adds global-scale cost and complexity for a small single-location dataset.
D. Multi-regional Spanner maximizes availability at even higher cost, still missing the logging recovery point.
ACE exam tip
Small plus relational plus point-in-time equals Cloud SQL with binary logging on.
4You 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?
Create an HTTP load balancer with a backend configuration that references an existing instance group. Set the health check to healthy (HTTP)
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.
Create a managed instance group. Set the Autohealing health check to healthy (HTTP)
Create a managed instance group. Verify that the autoscaling setting is on.
Answer: C
The short version
C — Managed instance group autohealing with HTTP health check. Load balancer backend settings do not recreate unhealthy VMs by themselves. The managed group autohealer with a health check does, in the fewest steps.
Key concepts in this question
Managed instance group: maintains identical VMs across zones from a template with self-healing.
Autohealing health check: HTTP check marks instances unhealthy after retries and triggers recreation.
Three by ten logic: three failed probes at ten-second intervals maps to check interval and threshold tuning.
Load balancer role: distributes traffic but does not rebuild failed backends alone.
Fewest-steps goal: one group plus one health check beats separate balancer plus group juggling.
Why C is correct
Autohealing is a managed instance group feature that watches each VM with the attached HTTP health check and recreates the VM when consecutive probes fail. Creating the managed group and setting its autohealing health check to healthy HTTP directly implements recreate-if-unresponsive-after-three-ten-second-attempts across multiple zones. Settled key C is therefore the minimal path, while load-balancer-only options leave no recreation mechanism in place.
Why the others are wrong
A. Creates a load balancer backend without enabling group autohealing, so failed VMs are not recreated.
B. Tunes balancing mode and RPS but still omits the autohealing recreation behavior.
D. Verifies autoscaling, which scales count by load rather than recreating unhealthy instances.
ACE exam tip
Need VM recreation on failure, choose managed group autohealing, not the balancer.
5You 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?
Use gcloud config configurations describe to review the output.
Use gcloud config configurations activate and gcloud config list to review the output.
Use kubectl config get-contexts to review the output.
Use kubectl config use-context and kubectl config view to review the output.
Answer: A
The short version
A — Describe the inactive configuration without activating it. Activating plus listing changes state and adds steps. Describe reads the stored properties of the named inactive config directly.
Key concepts in this question
gcloud configurations: named profiles store project, cluster, zone, and account properties.
Describe verb: shows details of a specific configuration by name without switching to it.
Activate overhead: switching active config then listing requires two commands and mutates shell state.
Fewest-steps rule: a single read-only describe beats activate-plus-list or context-switch flows.
Why A is correct
Multiple gcloud configurations coexist with only one active at a time, each holding its own Kubernetes Engine cluster pointer. The configurations describe command accepts the inactive configuration name and prints its properties, including the cluster, without changing the active selection. Settled key A therefore reviews the inactive cluster in one read-only step, while the alternatives require activation or inspect the wrong layer.
Why the others are wrong
B. Activates then lists, which mutates state and uses more steps than a direct describe.
C. Lists kubectl contexts, which shows cluster access entries rather than the inactive gcloud config.
D. Switches kubectl context plus views config, which is longer and still misses gcloud-level properties.
ACE exam tip
Inspect inactive gcloud config, use configurations describe, never activate first.
6Your 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?
Multi-Regional Storage
Regional Storage
Nearline Storage
Coldline Storage
Answer: D
The short version
D — Coldline for rarely accessed disaster-recovery backups. Backup files kept for disaster recovery are accessed infrequently. Coldline pricing rewards rare reads with the lowest at-rest cost among the listed classes.
Key concepts in this question
Access frequency tiers: Standard, Nearline, and Coldline trade storage price against retrieval cost and latency.
Backup pattern: disaster-recovery copies sit idle for months and are read only during tests or disasters.
Google practice: cold backup data belongs in the coldest cost-efficient class that still meets recovery time.
Regional vs multi-regional: location scope is separate from temperature tier and durability.
Settled-key fit: vendor key selects Coldline for this infrequent-access backup use.
Why D is correct
Application backup files for disaster recovery are written once and rarely read, which is the textbook Coldline profile of data accessed at most quarterly with tolerance for slightly higher retrieval cost. Storing them in Coldline minimizes monthly storage spend while keeping the objects available for restore when needed. Settled key D therefore follows Google recommended practice for cold backup retention better than warmer Standard, Regional, or Nearline placements.
Why the others are wrong
A. Multi-regional Standard optimizes for hot global serving, overpaying for idle backups.
B. Regional Standard suits frequently accessed data in one region, not cold backups.
C. Nearline suits monthly reads, still warmer and pricier at rest than Coldline for disaster-only data.
ACE exam tip
Disaster-recovery backups read rarely, choose the coldest listed storage class.
7Several 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?
Contact [email protected] with your bank account details and request a corporate billing account for your company.
Create a ticket with Google Support and wait for their call to share your credit card details over the phone.
In the Google Platform Console, go to the Resource Manage and move all projects to the root Organizarion.
In the Google Cloud Platform Console, create a new billing account and set up a payment method.
Answer: D
The short version
D — Create a new billing account with a payment method in Cloud Console. Scattered personal cards cannot be fixed by emails or tickets. A central billing account lets projects be moved under one payer in one workflow.
Key concepts in this question
Billing account centralization: one payer account funds many projects for consolidated invoicing.
Console creation flow: billing accounts are created in Cloud Console with a payment instrument.
Project linkage: existing projects change their billing account association to the new payer.
Support-channel distractors: emailing details or phoning card numbers is not the documented path.
Org move insufficiency: moving projects in Resource Manager alone does not unify payment.
Why D is correct
The company needs to stop reimbursing personal cards and place all Cloud projects under a single corporate payer. Creating a new billing account in Cloud Console and attaching the corporate payment method establishes that payer in the fewest governed steps. Each existing project is then linked to the new billing account, centralizing charges and reporting. Settled key D describes exactly that Console-based creation flow.
Why the others are wrong
A. Emailing bank details to a generic address is not the billing-account creation process and is insecure.
B. Opening a ticket to share card details by phone adds manual handling and skips Console creation.
C. Moving projects in Resource Manager changes hierarchy, not the payer, so cards remain scattered.
ACE exam tip
Unify scattered cards, create one billing account in Console then relink projects.
8You 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?
Reserve the IP 10.0.3.21 as a static internal IP address using gcloud and assign it to the licensing server.
Reserve the IP 10.0.3.21 as a static public IP address using gcloud and assign it to the licensing server.
Use the IP 10.0.3.21 as a custom ephemeral IP address and assign it to the licensing server.
Start the licensing server with an automatic ephemeral IP address, and then promote it to a static internal IP address.
Answer: A
The short version
A — Reserve 10 dot 0 dot 3 dot 21 as a static internal IP. The app hardcodes that internal address, so the licensing server must own it inside the VPC. A static internal reservation guarantees the binding survives restarts.
Key concepts in this question
Fixed internal dependency: licensing client expects the server at a specific RFC 1918 address.
Static internal IP: reserved address in the subnet that persists across stop, start, and recreation.
Public vs internal scope: public reservation would expose the server and would not satisfy internal routing.
Ephemeral fragility: ephemeral addresses can change on restart and break the hardcoded client.
Promote-after-start risk: starting ephemeral then promoting still risks initial mismatch and extra steps.
Why A is correct
Because the application cannot be reconfigured and looks for 10 dot 0 dot 3 dot 21 on the internal network, the licensing VM must deterministically hold that internal address. Reserving the address as a static internal IP with gcloud and assigning it to the licensing server ensures VPC routing delivers client traffic to the right host permanently. Settled key A is the only option that pins the exact internal address before launch.
Why the others are wrong
B. Reserves a public address, which does not place the server at the expected internal address.
C. Uses an ephemeral custom address that can change and lacks the static guarantee.
D. Starts ephemeral then promotes, adding steps and risking that the first boot misses the required IP.
ACE exam tip
Hardcoded internal IP, reserve it static internal before assigning to the VM.
9You 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?
Manual Scaling with 3 instances.
Basic Scaling with min_instances set to 3.
Basic Scaling with max_instances set to 3.
Automatic Scaling with min_idle_instances set to 3.
Answer: D
The short version
D — Automatic scaling with min idle instances of 3. Request-rate scaling needs automatic mode, and the always-ready requirement maps to idle-instance minimum. Manual and basic modes cannot express both together.
Key concepts in this question
Automatic scaling: adjusts instance count by request rate, latency, and CPU for request-driven apps.
min idle instances: keeps a floor of warm unoccupied instances ready for spikes.
Basic scaling behavior: scales on pending queue depth, not request rate with idle floor.
Manual scaling rigidity: fixed count cannot scale with request rate at all.
At-least-three rule: the floor must be three warm instances at all times.
Why D is correct
The app must scale on request rate while never dropping below three unoccupied instances, which is precisely what automatic scaling with min idle instances set to 3 provides. Request-driven scaling handles bursts, and the idle floor keeps warm capacity ready without serving traffic yet. Settled key D is the only scaling type that combines request-rate autoscaling with an idle-instance minimum.
Why the others are wrong
A. Manual scaling with 3 fixes the count and never scales with request rate.
B. Basic scaling with min instances reacts to queue depth and lacks the idle-instance request-rate semantic.
C. Basic scaling with max of 3 caps capacity and still misses request-rate plus idle-floor behavior.
ACE exam tip
Request-rate plus warm floor equals automatic scaling with min idle instances.
10You 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?
Use gcloud iam roles copy and specify the production project as the destination project.
Use gcloud iam roles copy and specify your organization as the destination organization.
In the Google Cloud Platform Console, use the ג€˜create role from roleג€™ functionality.
In the Google Cloud Platform Console, use the ג€˜create roleג€™ functionality and select all applicable permissions.
Answer: A
The short version
A — Copy custom roles to the production project with gcloud iam roles copy. Console rebuilds and org-level copies miss the fewest-steps goal. Direct project-to-project role copy replicates definitions exactly.
Key concepts in this question
Custom IAM roles: project-level definitions that must be reproduced in the new project.
gcloud iam roles copy: clones a role from source to destination project in one command.
Destination precision: copying to the production project scopes the roles where they are needed.
Org-copy overbreadth: copying to the organization changes scope and adds governance overhead.
Console manual effort: recreating from role or from scratch requires many clicks and risks drift.
Why A is correct
Development already holds the correct custom role definitions, and production needs identical roles with minimal work. The gcloud iam roles copy command with the production project as destination duplicates each role definition verbatim in seconds without manual permission picking. Settled key A therefore achieves parity in the fewest steps while keeping scope tight to production.
Why the others are wrong
B. Copies to the organization, which broadens scope and does not place roles directly in production.
C. Uses Console create-from-role, which is manual per role and slower than a CLI copy.
D. Recreates roles permission by permission in Console, maximizing effort and drift risk.
ACE exam tip
Clone IAM roles across projects, use gcloud iam roles copy to the target project.