Skip to content

Professional Cloud Network Engineer · Domain 1: Designing and planning a Google Cloud VPC network · 21% of the exam

Task 1.1: Designing an overall network architecture.

Network Service Tiers, availability, failover and scale, the DNS topology, choosing a load balancer, GKE address planning and control plane access, IAM roles for network designs, connectivity to managed services, and quotas and limits.

Study it

  • Network architecture: Network Service Tiers, availability and failover, DNS topology, load balancer choice, IAM roles and quotas

    Partly covered by: Load Balancing

  • Connectivity to managed services: private services access, Private Service Connect, Serverless VPC Access and Direct VPC egress

    Lesson coming

Sample questions

Try each one before opening the answer. Every option is explained, with the Google Cloud documentation page that proves it.

Question 1 · choose 1

An internal message broker speaks its own binary protocol over TCP port 5672. Its VMs run in managed instance groups in us-central1 and us-east4, and clients in both regions use one internal address per region. If every broker VM in one region fails, clients in that region must keep working through the same address. The brokers do not need to see client source IP addresses. Which load balancer design meets the requirements?

  1. AAn internal passthrough Network Load Balancer in each region, with global access enabled on both forwarding rules
  2. BA cross-region internal Application Load Balancer with both instance groups in one global backend service
  3. CA regional internal proxy Network Load Balancer in each region, with global access enabled on both forwarding rules
  4. DA cross-region internal proxy Network Load Balancer with both instance groups in one global backend service
Show the answer and why
  • AAn internal passthrough Network Load Balancer in each region, with global access enabled on both forwarding rules

    Incorrect

    Each internal passthrough Network Load Balancer sends traffic only to backends in its own region, so a region whose brokers all fail has no backend left behind its address.

  • BA cross-region internal Application Load Balancer with both instance groups in one global backend service

    Incorrect

    An Application Load Balancer is a layer 7 proxy for HTTP and HTTPS requests; the broker's protocol is not HTTP.

  • CA regional internal proxy Network Load Balancer in each region, with global access enabled on both forwarding rules

    Incorrect

    Global access only lets clients from other regions reach the address; the regional mode still sends traffic only to backends in its own region.

  • DA cross-region internal proxy Network Load Balancer with both instance groups in one global backend service

    Correct

    The cross-region mode proxies TCP, gives each region a VIP that shares the global backend service, and fails over to healthy backends in another region.

Two facts decide this. The protocol is plain TCP, which rules out the HTTP proxies. The failover needs backends in two regions behind one address, which the passthrough load balancer and the regional proxy mode do not offer: global access changes who may connect, not where the backends are. The cross-region internal proxy Network Load Balancer is the internal TCP load balancer whose backend service is global.

Question 2 · choose 1

A Cloud Run service will reach internal databases through Direct VPC egress in a new dedicated subnet. It is capped at 150 container instances. Each release shifts all traffic at once to a new revision, which scales from zero to 150 instances while the old revision scales from 150 to zero. What is the smallest IPv4 subnet size that keeps releases from running out of addresses?

  1. AA /22 range, 1,024 addresses
  2. BA /23 range, 512 addresses
  3. CA /24 range, 256 addresses
  4. DA /26 range, 64 addresses
Show the answer and why
  • AA /22 range, 1,024 addresses

    Correct

    During a release both revisions hold addresses: (150 + 150) instances at two addresses each is 600, which fits in 1,024 but not in 512.

  • BA /23 range, 512 addresses

    Incorrect

    512 covers two addresses per instance for one revision (300), but the old revision keeps its addresses for up to 20 minutes after scaling down, so a release needs about 600.

  • CA /24 range, 256 addresses

    Incorrect

    256 assumes one address per instance; Cloud Run uses about twice as many addresses as container instances at steady state.

  • DA /26 range, 64 addresses

    Incorrect

    /26 is the minimum subnet size the documentation sets for Direct VPC egress, not a size derived from this service's instance count.

At steady state Cloud Run uses about two IP addresses per container instance, and when a revision scales down it keeps its addresses for up to 20 minutes. A release that moves 150 instances from one revision to another therefore needs about (150 + 150) x 2 = 600 addresses for a while, plus a buffer. A /22 is the smallest range here that holds that.

Question 3 · choose 1

In a Shared VPC, the payments team runs the backend service pay-be of a regional internal Application Load Balancer in service project svc-pay. The storefront team, in service project svc-shop, must build its own load balancer frontend (forwarding rule, target proxy and URL map) that sends requests to pay-be. svc-pay holds four other backend services that svc-shop must not use. Which setup follows least privilege and works?

  1. ACompute Load Balancer Services User on svc-pay for the storefront's load balancer admins, with the frontend in the Shared VPC network of pay-be
  2. BCompute Load Balancer Services User on pay-be itself for the storefront's load balancer admins, with the frontend in the Shared VPC network of pay-be
  3. CCompute Load Balancer Services User on pay-be itself for the storefront's load balancer admins, with the frontend in a VPC network of svc-shop peered with the Shared VPC network
  4. DCompute Load Balancer Admin on svc-pay for the storefront's load balancer admins, with the frontend in the Shared VPC network of pay-be
Show the answer and why
  • ACompute Load Balancer Services User on svc-pay for the storefront's load balancer admins, with the frontend in the Shared VPC network of pay-be

    Incorrect

    A project-level grant of the role covers every backend service in svc-pay, including the four that svc-shop must not use.

  • BCompute Load Balancer Services User on pay-be itself for the storefront's load balancer admins, with the frontend in the Shared VPC network of pay-be

    Correct

    The role gives compute.backendServices.use and can be granted on a single backend service, and cross-project referencing needs the frontend in the backend service's network.

  • CCompute Load Balancer Services User on pay-be itself for the storefront's load balancer admins, with the frontend in a VPC network of svc-shop peered with the Shared VPC network

    Incorrect

    For cross-project service referencing, the frontend must use the same Shared VPC network that was used to create the backend service.

  • DCompute Load Balancer Admin on svc-pay for the storefront's load balancer admins, with the frontend in the Shared VPC network of pay-be

    Incorrect

    Load Balancer Admin on svc-pay lets the team manage all of that project's load balancer resources, far beyond using one backend service.

Cross-project service referencing lets a frontend in one service project use a backend service in another. The Load Balancer Admin of the frontend needs compute.backendServices.use, which the Compute Load Balancer Services User role grants; the backend project's Service Project Admin can grant it at the project level, which covers all of the project's backend services, or on an individual backend service. The frontend must use the same Shared VPC network as the backend service.

Question 4 · choose 1

A project's default network tier is Standard Tier. A team now wants to enable Cloud CDN for the backend service of its external Application Load Balancer. Which network tier must the load balancer's IP address use?

  1. AStandard Tier, matching the project's default tier
  2. BEither tier, because Cloud CDN caches content independently of the tier
  3. CPremium Tier, because Cloud CDN is always Premium Tier
  4. DStandard Tier, if the backend is a Cloud Storage bucket
Show the answer and why
  • AStandard Tier, matching the project's default tier

    Incorrect

    Cloud CDN can't be used with Standard Tier, whatever the project default is.

  • BEither tier, because Cloud CDN caches content independently of the tier

    Incorrect

    The tier matters: Cloud CDN works only with Premium Tier.

  • CPremium Tier, because Cloud CDN is always Premium Tier

    Correct

    The Network Service Tiers overview states that Cloud CDN is always Premium Tier and can't be used with Standard Tier.

  • DStandard Tier, if the backend is a Cloud Storage bucket

    Incorrect

    A bucket backend allows Standard Tier for the load balancer itself, but not together with Cloud CDN.

Network Service Tiers can be set per project and per resource. Some products only run on Premium Tier: Cloud CDN is one of them, so a load balancer that uses Cloud CDN needs a Premium Tier address even in a project whose default is Standard Tier.

Question 5 · choose 1

An administrator grants an engineer the Compute Load Balancer Admin role. One minute later, the engineer's attempt to create a backend service is still denied. What explains this?

  1. AThe role doesn't include permissions on backend services
  2. BLoad balancing roles take effect only after the engineer signs out and signs in again
  3. CThe role must be granted on the organization, not on the project
  4. DCloud Load Balancing caches IAM permissions for five minutes
Show the answer and why
  • AThe role doesn't include permissions on backend services

    Incorrect

    Backend services are load balancing resources, which is what this predefined role is for.

  • BLoad balancing roles take effect only after the engineer signs out and signs in again

    Incorrect

    The delay comes from permission caching on the service side, not from the engineer's session.

  • CThe role must be granted on the organization, not on the project

    Incorrect

    Load balancing roles work when granted on the project that holds the resources.

  • DCloud Load Balancing caches IAM permissions for five minutes

    Correct

    Because of that cache, a role change can take up to five minutes to become effective.

Cloud Load Balancing caches IAM permissions for five minutes, so a newly granted or removed role can take up to five minutes to take effect for load balancing API requests.

Question 6 · choose 1

After a data center started advertising many more prefixes, some of them never appear as dynamic routes in the VPC network, while others do. The network engineer suspects the Cloud Router dynamic route quotas. Where can the engineer confirm which prefixes were dropped?

  1. AIn the VPC Flow Logs of the subnets that send traffic to those prefixes
  2. BIn the Cloud Router's BGP session status, which lists each rejected prefix
  3. CIn the route_quota_events entries that Cloud Logging records
  4. DIn the Firewall Insights recommendations for the network
Show the answer and why
  • AIn the VPC Flow Logs of the subnets that send traffic to those prefixes

    Incorrect

    Flow logs record sampled traffic, not the routes the dynamic route control plane dropped.

  • BIn the Cloud Router's BGP session status, which lists each rejected prefix

    Incorrect

    The dropping happens in the dynamic route control plane, and the documentation points to Cloud Logging for it.

  • CIn the route_quota_events entries that Cloud Logging records

    Correct

    When the quotas are exceeded, the dropped route information can be identified through route_quota_events in Cloud Logging.

  • DIn the Firewall Insights recommendations for the network

    Incorrect

    Firewall Insights analyzes firewall rules, not dynamic routes.

Each region's dynamic route control plane enforces quotas on unique Cloud Router dynamic route prefixes. It drops the route information for prefixes beyond the limits, deterministically, and those drops are visible as route_quota_events in Cloud Logging.

Practice domain 1 →Practice all domains →