CertSafari

    Free Google Cloud Professional Cloud Network Engineer Sample Questions

    35 free sample questions from our bank of 349+, covering every exam domain, with answers and detailed explanations. Updated November 2026.

    1: Designing and planning a Google Cloud VPC network

    1.1: Designing an overall network architecture

    1.A fintech serves an internal API through an internal passthrough Network Load Balancer in us-central1, with an identical stack in us-east1. Internal clients resolve api.corp.internal from a Cloud DNS private zone. If all us-central1 backends become unhealthy, clients must move to us-east1 automatically without manual changes. What should the network engineer configure?

    1. A.Add a Cloud DNS response policy rule for api.corp.internal returning the us-east1 address, and enable it from a Cloud Monitoring alert on us-central1 backends.
    2. B.Create an A record set for api.corp.internal with a failover routing policy, referencing both forwarding rules and using their health checks to cut over.
    3. C.Add the us-east1 instance groups as backends of the us-central1 load balancer's backend service, so one forwarding rule address serves both regions.
    4. D.Publish both forwarding rule addresses in one A record set with a 30-second TTL, so clients fall back to the other address when a connection fails.
    Show answer & explanation

    Correct answer: B — Create an A record set for api.corp.internal with a failover routing policy, referencing both forwarding rules and using their health checks to cut over.

    • A. Response policy rules are static overrides and Cloud DNS does not evaluate backend health for them. A Monitoring alert only notifies, so enabling the rule still needs manual or custom-built action.
    • B. A Cloud DNS failover routing policy can reference the forwarding rules of internal passthrough Network Load Balancers. It uses their health to return the us-east1 address when us-central1 is down.
    • C. Backends of an internal passthrough Network Load Balancer must be in the same region as its forwarding rule. us-east1 instance groups cannot be added to the us-central1 backend service.
    • D. A plain multi-value record set has no health awareness, so DNS keeps returning the dead us-central1 address. Fallback depends on each client's retry behavior, so cutover is not dependable.

    1.2: Designing VPC networks

    2.A bank hosts a shared-services VPC with DNS and logging. Roughly 30 partner VPCs must reach shared services, but partner VPCs must never reach each other. The team wants this enforced by the connectivity topology itself, using Network Connectivity Center. What should they configure?

    1. A.Create a mesh-topology NCC hub, attach the shared-services and partner VPCs as spokes, and add hierarchical firewall denies between partners.
    2. B.Create a star-topology NCC hub and put the shared-services VPC and the 30 partner VPC spokes together in the hub's edge group.
    3. C.Create a star-topology NCC hub and put the partner VPC spokes in the center group and the shared-services VPC in the edge group.
    4. D.Create a star-topology NCC hub and put the shared-services VPC in the center group and the partner VPC spokes in the edge group.
    Show answer & explanation

    Correct answer: D — Create a star-topology NCC hub and put the shared-services VPC in the center group and the partner VPC spokes in the edge group.

    • A. A mesh topology gives each attached spoke a route to the other spokes, so partner VPCs would be connected to one another. Blocking them with firewall denies enforces isolation in policy rather than in the connectivity topology, which the bank explicitly wants.
    • B. In a star hub, edge-group spokes can exchange traffic only with center-group spokes, not with other edge spokes. With shared services and the partners all in the edge group, the partners could not reach shared services, so the connectivity requirement fails.
    • C. Center-group spokes can reach one another and the edge spokes, so partners placed in the center group could talk to each other. That breaks the requirement that partner VPCs never reach each other. The roles of the two groups are inverted.
    • D. In a star topology, edge-group spokes can communicate only with center-group spokes and never with each other. With the shared-services VPC in the center group, each partner can reach it, and the partner VPCs stay isolated from one another by the topology itself.

    1.4: Designing for Google Kubernetes Engine (GKE)

    3.A SaaS provider runs an internal-LoadBalancer Service on GKE and wants customers in separate VPCs, several using overlapping RFC 1918 ranges, to consume it privately. VPC peering is ruled out because of the overlaps. What design fits?

    1. A.Expose the internal load balancer through a service attachment with a dedicated NAT subnet, and give customers an endpoint in their own VPC to reach it.
    2. B.Peer each customer VPC with the provider VPC, and advertise the load balancer's forwarding rule address to the peers through custom route export.
    3. C.Set up HA VPN tunnels from the provider VPC to the customer VPCs, and advertise the load balancer's forwarding rule address to them over BGP.
    4. D.Front the workload with a global external Application Load Balancer protected by Cloud Armor, and let customers connect to it from their VPCs.
    Show answer & explanation

    Correct answer: A — Expose the internal load balancer through a service attachment with a dedicated NAT subnet, and give customers an endpoint in their own VPC to reach it.

    • A. Private Service Connect publishes the load balancer as a service attachment. Each customer reaches it through an endpoint with an IP from their own subnet, and the producer's NAT subnet translates source addresses, so overlapping RFC 1918 ranges never need to be routed between networks.
    • B. VPC Network Peering cannot be established between networks whose subnet ranges overlap, and custom route export only shares routes that already exist in the peering. It cannot resolve the conflict, so peering is unusable for customers with overlapping RFC 1918 space.
    • C. HA VPN with BGP still routes between the networks at the IP level, so overlapping customer and provider prefixes produce ambiguous or conflicting routes. It also leaves one tunnel pair and BGP configuration to maintain per customer, so it does not solve the overlap problem.
    • D. A global external Application Load Balancer is reached through a public IP address. Cloud Armor restricts who can connect but does not make the path private, so customers would not be consuming the service privately as required.

    1.3: Designing a resilient and performant hybrid and multi-cloud network

    4.A hub-and-spoke design has a hub VPC with a Cloud DNS forwarding zone for corp.example.com targeting on-premises resolvers. Spoke VPCs live in separate projects and also need to resolve corp.example.com. Select TWO correct actions.(Select 2)

    1. A.Create a forwarding zone for corp.example.com in each spoke VPC network, targeting the on-premises resolvers through a dedicated tunnel per spoke
    2. B.Configure VPC Network Peering between the hub and each spoke with custom route exchange, extending the hub's forwarding zone to the spoke networks
    3. C.Create a DNS peering zone for corp.example.com in each spoke VPC network, setting the hub VPC network as the peer network for the zone
    4. D.Grant the DNS Peer role to whoever creates each spoke's peering zone, on the project that owns the hub VPC network targeted by the zone
    5. E.Chain DNS peering zones from spoke A to spoke B and from spoke B to the hub VPC network, sending queries across two hops to the forwarding zone
    Show answer & explanation

    Correct answers: C, D — Create a DNS peering zone for corp.example.com in each spoke VPC network, setting the hub VPC network as the peer network for the zone; Grant the DNS Peer role to whoever creates each spoke's peering zone, on the project that owns the hub VPC network targeted by the zone

    • A. Incorrect: per-spoke forwarding zones need a separate hybrid connection from every spoke to on-premises. That duplicates the resolver targets and connectivity instead of reusing the forwarding zone and the connection already in the hub.
    • B. Incorrect: VPC Network Peering and custom route exchange only exchange routes. They do not make a Cloud DNS zone visible in another network. Spokes need DNS peering zones that point at the hub.
    • C. Correct: a DNS peering zone in each spoke sends queries for corp.example.com to the hub network. The hub's forwarding zone then passes them to the on-premises resolvers.
    • D. Correct: because each spoke lives in a different project from the hub, creating a peering zone that targets the hub network needs the DNS Peer role on the project that owns that network.
    • E. Incorrect: DNS peering is not transitive and is limited to one hop. A peering zone whose target network depends on another peering zone does not resolve the name, so each spoke must peer directly to the hub.

    1.3: Designing a resilient and performant hybrid and multi-cloud network

    5.On-premises DNS servers must resolve names in a Cloud DNS private zone attached to a VPC that is connected over Interconnect. Which configuration should the engineer apply?

    1. A.Attach a response policy to the VPC network with rules for the zone's names, then forward on-premises queries for that domain to the VPC's Cloud DNS resolver
    2. B.Enable an inbound server policy on the VPC network, then configure on-premises conditional forwarding for the domain to the policy's forwarder IP addresses
    3. C.Point on-premises resolvers at the metadata server 169.254.169.254 on the VPC network, then advertise that address over BGP across the Interconnect
    4. D.Create an outbound forwarding zone on the VPC network for the domain targeting the on-premises resolvers, then query that zone from on-premises clients
    Show answer & explanation

    Correct answer: B — Enable an inbound server policy on the VPC network, then configure on-premises conditional forwarding for the domain to the policy's forwarder IP addresses

    • A. Incorrect: a response policy only changes answers for queries that originate inside the VPC. It creates no address that on-premises resolvers can send queries to, and the VPC's internal resolver is not reachable from on-premises.
    • B. Correct: an inbound server policy allocates inbound forwarding addresses in the VPC's subnets. On-premises DNS servers can forward queries for the private zone's domain to those IPs over the Interconnect.
    • C. Incorrect: 169.254.169.254 is a link-local metadata server address that only answers VM instances inside Google Cloud. Advertising it over BGP does not make it reachable or usable as a resolver for on-premises hosts.
    • D. Incorrect: a forwarding zone sends queries from the VPC out to the listed target resolvers. That is the opposite direction from what is needed, because on-premises clients cannot query a forwarding zone to reach VPC records.

    2: Implementing a VPC network

    2.4: Configuring and maintaining GKE clusters

    6.A namespace has a default-deny policy for both ingress and egress, and NodeLocal DNSCache is enabled. Pods in the namespace can no longer resolve names. Which TWO egress rules restore resolution without opening unrelated traffic? Select TWO.(Select 2)

    1. A.Allow egress to Pods labeled `k8s-app: kube-dns` in the `kube-system` namespace on UDP and TCP port 53.
    2. B.Allow egress to the node-local cache address 169.254.20.10/32 with an `ipBlock` on UDP and TCP port 53.
    3. C.Allow egress to 0.0.0.0/0 on all ports so Pods can reach whichever resolver the node configures for upstream queries.
    4. D.Allow ingress from `kube-system` Pods on port 53 so that DNS responses can return to the application Pods in the namespace.
    5. E.Allow egress to the metadata server at 169.254.169.254/32 on port 80 so that Pods can obtain name servers from the node.
    Show answer & explanation

    Correct answers: A, B — Allow egress to Pods labeled `k8s-app: kube-dns` in the `kube-system` namespace on UDP and TCP port 53.; Allow egress to the node-local cache address 169.254.20.10/32 with an `ipBlock` on UDP and TCP port 53.

    • A. Correct. Cache misses are forwarded to kube-dns, so Pods need egress to the kube-dns Pods on port 53 using both protocols.
    • B. Correct. With NodeLocal DNSCache, Pods query the link-local cache address, so egress to it must be explicitly permitted.
    • C. Incorrect. That opens every destination and port, defeating the default-deny posture the team deliberately implemented.
    • D. Incorrect. Network policies are stateful, so return traffic is allowed automatically and the missing piece is egress, not ingress.
    • E. Incorrect. The metadata server provides instance metadata and credentials, and Pod DNS queries do not depend on it.

    2.4: Configuring and maintaining GKE clusters

    7.A `backend` Pod must accept traffic only from Pods labeled `app: web` in namespaces labeled `env: prod`. An engineer writes a single ingress rule with two `from` entries: one `namespaceSelector` for `env: prod` and one `podSelector` for `app: web`. What does the policy actually allow?

    1. A.Only `app: web` Pods that sit inside namespaces labeled `env: prod`, since entries in the same rule are combined with a logical AND.
    2. B.Only Pods labeled `app: web` in any namespace, because a `podSelector` entry overrides a `namespaceSelector` entry in the same rule.
    3. C.All Pods in `env: prod` namespaces plus `app: web` Pods in the policy's own namespace, as separate `from` entries use OR.
    4. D.Nothing, because a rule mixing selector types is rejected by the API server and the policy is never applied to the Pods.
    Show answer & explanation

    Correct answer: C — All Pods in `env: prod` namespaces plus `app: web` Pods in the policy's own namespace, as separate `from` entries use OR.

    • A. Incorrect. Separate `from` entries are combined with OR, so this reading would only be true if both selectors were in a single entry.
    • B. Incorrect. Neither entry overrides the other, and the namespace selector still contributes its own allowed sources.
    • C. Correct. Each list item is evaluated independently, so the result is broader than intended; one entry holding both selectors would be needed.
    • D. Incorrect. Mixing namespace and Pod selectors is valid syntax, and the policy is applied exactly as written.

    2.1: Configuring VPCs

    8.A shared-services VPC `hub` is peered with VPC `team-a` and VPC `team-b`, and all three use non-overlapping ranges. VMs in `team-a` cannot reach VMs in `team-b`. Which change enables private connectivity with the least extra infrastructure?

    1. A.Add a static route in `team-a` for the `team-b` range, using the peering connection to `hub` as the next hop for that destination.
    2. B.Enable custom route export on the `hub` side of both peerings and custom route import on the `team-a` and `team-b` sides.
    3. C.Create a VPC Network Peering connection directly between `team-a` and `team-b`, with subnet route exchange left at its defaults.
    4. D.Add VPC firewall rules in `hub` that allow ingress and egress between the `team-a` and `team-b` ranges, using the hub as the transit point.
    Show answer & explanation

    Correct answer: C — Create a VPC Network Peering connection directly between `team-a` and `team-b`, with subnet route exchange left at its defaults.

    • A. A VPC Network Peering connection is not a valid next hop for a static route. Peering also does not relay traffic from one peer to another, so the route would not give team-a a path to team-b.
    • B. Custom route export shares only the hub's own static and dynamic routes. Subnet routes the hub learned from one spoke are not re-exported to the other, because VPC Network Peering is non-transitive. Neither spoke gains a route to the other.
    • C. Peering is non-transitive, so the two spokes need a peering connection of their own. Their ranges do not overlap, so it can be created. It adds no new infrastructure, and subnet routes are exchanged by default.
    • D. Firewall rules only allow or deny traffic that already has a route. Peered networks never route through a middle network, so rules in the hub create no path between team-a and team-b.

    2.1: Configuring VPCs

    9.A team needs Cloud SQL for PostgreSQL instances with private IP, reachable from VMs in a custom VPC, using private services access. What is the correct setup?

    1. A.Run `gcloud compute networks peerings create` against the Cloud SQL tenant project's VPC, then add a custom route for the instance range.
    2. B.Allocate an internal range with purpose `VPC_PEERING`, then create a connection to `servicenetworking.googleapis.com` that uses the range.
    3. C.Create a subnet with purpose `PRIVATE_SERVICE_CONNECT` in the VPC, then select it when creating each instance to host its internal address.
    4. D.Add a secondary range to a regional subnet, then reserve each instance's address as an alias IP from that range for the managed database.
    Show answer & explanation

    Correct answer: B — Allocate an internal range with purpose `VPC_PEERING`, then create a connection to `servicenetworking.googleapis.com` that uses the range.

    • A. The tenant project's VPC that hosts Cloud SQL is managed by Google and is not exposed to customers, so a peering cannot be created to it by hand. Peering with the producer network is only done through a Service Networking connection, and a custom route does not change that.
    • B. Private services access works by reserving an internal range with purpose VPC_PEERING and creating a private connection to servicenetworking.googleapis.com that uses it. This peers the VPC with the Google service producer network, and Cloud SQL then assigns instance private IPs from the reserved range.
    • C. A subnet with purpose PRIVATE_SERVICE_CONNECT is a NAT subnet used by a producer to publish a service attachment. It is not a place where Cloud SQL instances get addresses, and private services access draws instance IPs from an allocated peering range rather than from a subnet.
    • D. Secondary ranges and alias IPs apply to VM and GKE network interfaces inside your subnet. Cloud SQL instances run in a Google-managed tenant network, so their addresses cannot be reserved as alias IPs from a subnet range.

    2.2: Configuring VPC routing

    10.A hub VPC has a Cloud Router learning on-premises prefixes over Interconnect. A spoke VPC peered with the hub must reach those prefixes. Select TWO actions that make the spoke use the learned routes.(Select 2)

    1. A.Add the network tag `onprem` to the hub's routes and the spoke VMs, so tagged instances send on-premises traffic across the peering.
    2. B.Create a Cloud Router in the spoke VPC with a BGP session to the hub Cloud Router, so it relays the on-premises prefixes.
    3. C.Set the spoke VPC's dynamic routing mode to global, so the spoke installs the prefixes that the hub Cloud Router learned.
    4. D.Enable import of custom routes on the spoke's peering, so the hub's dynamic routes are added to the spoke's routing table.
    5. E.Enable export of custom routes on the hub's peering, so the hub's dynamic routes are offered to the peered spoke network.
    Show answer & explanation

    Correct answers: D, E — Enable import of custom routes on the spoke's peering, so the hub's dynamic routes are added to the spoke's routing table.; Enable export of custom routes on the hub's peering, so the hub's dynamic routes are offered to the peered spoke network.

    • A. Incorrect. Routes that carry network tags are not exchanged over VPC peering, and routes learned dynamically by a Cloud Router cannot be tagged, so tagging cannot make the spoke use the on-premises prefixes.
    • B. Incorrect. A Cloud Router establishes BGP sessions with on-premises or appliance routers, not with another VPC's Cloud Router, so the hub's prefixes cannot be relayed this way. Peering route exchange is configured on the peering itself.
    • C. Incorrect. A VPC's dynamic routing mode controls how its own Cloud Routers' learned routes apply across that VPC's regions. It does not affect what a peering exchanges, which is governed by the export and import custom route settings.
    • D. Correct. Custom routes shared over a peering are installed only if the receiving side opts in, so the spoke's peering must have import of custom routes enabled for the hub's offered routes to appear in its routing table.
    • E. Correct. The hub must export custom routes on its side of the peering. This setting covers dynamic routes learned by its Cloud Router, which is what makes the on-premises prefixes available to the spoke.

    2.3: Configuring Network Connectivity Center

    11.An engineer tries to create a VPC spoke for 'analytics-vpc' on a hub, and creation fails. The VPC has a subnet 172.16.8.0/22, and an existing VPC spoke on the hub contains subnet 172.16.8.0/24. What is the most likely cause and fix?

    1. A.The hub has reached its route quota for the project, so request a higher quota before creating the spoke
    2. B.The new VPC's subnets collide with a range on another spoke of the hub, so renumber one of the conflicting subnets
    3. C.The VPC has no peering with the existing spoke's VPC, so set up VPC Network Peering before adding the spoke
    4. D.The hub has export-psc disabled, so enable PSC propagation on the hub before adding the VPC spoke
    Show answer & explanation

    Correct answer: B — The new VPC's subnets collide with a range on another spoke of the hub, so renumber one of the conflicting subnets

    • A. Incorrect: a higher quota does not help here. The 172.16.8.0/22 subnet contains the 172.16.8.0/24 subnet of a spoke already on the hub, and Network Connectivity Center rejects VPC spokes with overlapping subnet ranges whatever the quota.
    • B. Correct: 172.16.8.0/22 contains 172.16.8.0/24, which belongs to another spoke on the same hub. Network Connectivity Center does not allow VPC spokes on one hub to have overlapping subnet ranges, so one of the conflicting subnets must be renumbered.
    • C. Incorrect: VPC spokes attach to the hub and exchange routes through it, so no peering between the VPCs is needed. Peering would not remove the overlap between the /22 and the /24 either.
    • D. Incorrect: the export-psc setting only controls whether Private Service Connect endpoints are propagated across the hub. It has no effect on whether a VPC spoke can be created, and it does not resolve the overlapping /22 and /24.

    2.3: Configuring Network Connectivity Center

    12.A media company wants two sites, each connected by Cloud Interconnect to a different region, to exchange traffic across Google's backbone using NCC site-to-site data transfer. Select TWO requirements that apply.(Select 2)

    1. A.Each spoke's Cloud Router must tag its advertised prefixes with a BGP community that the hub matches on
    2. B.Each site must add an HA VPN tunnel next to its Interconnect attachment to carry the transferred traffic
    3. C.Both hybrid spokes must attach in the same region, with one regional Cloud Router peering with both sites
    4. D.Each spoke's BGP peers must share one ASN, and that ASN must differ from the ASN used by the other spoke
    5. E.Each hybrid spoke must be created in a project separate from the hub's project and from the other spoke's
    6. F.Both hybrid spokes must attach to one hub, with the same VPC network set as the routing VPC for both
    Show answer & explanation

    Correct answers: D, F — Each spoke's BGP peers must share one ASN, and that ASN must differ from the ASN used by the other spoke; Both hybrid spokes must attach to one hub, with the same VPC network set as the routing VPC for both

    • A. Incorrect: site-to-site data transfer does not depend on tagging prefixes with BGP communities. NCC does not use a community match to decide which spoke re-advertises routes, so this is not a requirement for the two sites.
    • B. Incorrect: Interconnect VLAN attachments are themselves valid hybrid spokes, so the sites already have the connectivity they need. Adding HA VPN tunnels is not required for transfer between them.
    • C. Incorrect: the sites connect to different regions, and backbone transfer between regions is the purpose of the feature. Hybrid spokes are not required to share a region or a single regional Cloud Router.
    • D. Correct: BGP peers within one spoke share an ASN, and different spokes use different ASNs. This lets routes learned from one spoke be re-advertised to the other without loop prevention discarding them.
    • E. Incorrect: NCC sets no requirement that spokes sit in projects separate from the hub or from each other. Spokes can be created in the hub's project, so this is not a condition for site-to-site transfer.
    • F. Correct: site-to-site data transfer needs both hybrid spokes attached to the same hub and using the same VPC network as their routing VPC. NCC then re-advertises dynamic routes learned from one spoke through the sessions of the other.

    3: Configuring managed network services

    3.1: Configuring load balancing

    13.A platform team uses GKE Ingress for a public service and needs to attach a Cloud Armor security policy and set client IP session affinity on the resulting backend service, in a way that survives controller reconciliation. What should they do?

    1. A.Create a GCPBackendPolicy that sets securityPolicy and sessionAffinity, then point its targetRef at the Service behind the Ingress
    2. B.Create a BackendConfig that sets securityPolicy and sessionAffinity, then reference it from the Service with an annotation
    3. C.Create a FrontendConfig that sets securityPolicy and sessionAffinity, then reference it from the Ingress with an annotation
    4. D.Run gcloud compute backend-services update with --security-policy and --session-affinity=CLIENT_IP on the generated backend service
    Show answer & explanation

    Correct answer: B — Create a BackendConfig that sets securityPolicy and sessionAffinity, then reference it from the Service with an annotation

    • A. Incorrect: GCPBackendPolicy is a Gateway API resource consumed by the GKE Gateway controller. The GKE Ingress controller does not read it, so the backend service gets neither the policy nor the affinity.
    • B. Correct: BackendConfig is the GKE Ingress CRD for backend service settings such as Cloud Armor policies and session affinity. The Service references it with the cloud.google.com/backend-config annotation, and the controller keeps applying it during reconciliation.
    • C. Incorrect: FrontendConfig only covers frontend settings of the load balancer, such as SSL policies and HTTP-to-HTTPS redirects. Cloud Armor and session affinity apply to the backend service, so they belong in a BackendConfig.
    • D. Incorrect: the Ingress controller owns the generated backend service and can overwrite manual changes during reconciliation, so settings applied this way are not guaranteed to survive.

    3.1: Configuring load balancing

    14.A company runs a legacy application in an on-premises data centre connected with Dedicated Interconnect. They want a global external Application Load Balancer to front the on-premises servers using a hybrid connectivity NEG, and health checks fail. Which TWO actions fix the problem? Select TWO.(Select 2)

    1. A.Allow the Google health check source ranges on the application port in the on-premises firewall, so probes reach the servers
    2. B.Deploy Cloud NAT in the VPC, so health check probes to the on-premises servers leave with a translated source address
    3. C.Replace the NEG with a GCE_VM_IP_PORT zonal NEG listing the on-premises server IPs, so probes follow VPC subnet routes
    4. D.Add Cloud Router custom advertisements for the health check source ranges, so on-premises routing can return probe replies
    5. E.Set the VPC dynamic routing mode to global, so the Cloud Router exports the health check source ranges on-premises
    Show answer & explanation

    Correct answers: A, D — Allow the Google health check source ranges on the application port in the on-premises firewall, so probes reach the servers; Add Cloud Router custom advertisements for the health check source ranges, so on-premises routing can return probe replies

    • A. Correct: probes arrive from Google's health check ranges (130.211.0.0/22 and 35.191.0.0/16), so the on-premises firewall must permit them to the application port, or the probes are dropped even when routing is in place.
    • B. Incorrect: Cloud NAT translates connections that VMs initiate toward external destinations. Health check probes originate from Google's fixed ranges and are not VM-initiated traffic, so NAT leaves the failure unchanged.
    • C. Incorrect: GCE_VM_IP_PORT endpoints must be VM network interfaces inside the VPC. Servers in an on-premises data centre can only be targeted by a hybrid connectivity NEG (NON_GCP_PRIVATE_IP_PORT), which is the design the company wants.
    • D. Correct: probes come from Google's ranges, and the on-premises network needs a return path to 130.211.0.0/22 and 35.191.0.0/16. Cloud Router custom advertisements are what announce those prefixes over BGP.
    • E. Incorrect: global dynamic routing mode only widens which regions' subnet routes the Cloud Router learns and advertises. The health check ranges are not VPC subnets, so they are announced only through custom advertisements.

    3.2: Configuring Cloud CDN

    15.A media company stores assets in a third-party S3-compatible bucket at assets.storage.example-provider.com and fronts it with a global external Application Load Balancer using an internet NEG and Cloud CDN. Cache fills fail with 403 errors because the provider expects its own bucket hostname. What should be configured?

    1. A.Add the host to the Cloud CDN cache key policy on the backend service, so fills are keyed to the bucket's own name
    2. B.Add a health check on the backend service that sends the bucket FQDN as its Host, so cache fills reach the provider
    3. C.Add a custom request header on the backend service that sets Host to the bucket FQDN, so fills are accepted by the origin
    4. D.Switch the internet NEG to INTERNET_IP_PORT endpoints with the provider's IP, so fills connect straight to the origin
    Show answer & explanation

    Correct answer: C — Add a custom request header on the backend service that sets Host to the bucket FQDN, so fills are accepted by the origin

    • A. The cache key policy only decides how cached objects are identified and stored. It does not change the Host header on requests sent to the origin, so the provider keeps rejecting cache fills with 403.
    • B. Backends that are internet NEGs are not health checked, and probe settings never apply to cache-fill requests. The Host header on those fills stays unchanged, so the 403 responses continue.
    • C. By default the backend service forwards the client's original Host value, which the provider rejects. A custom request header that sets Host to the bucket FQDN makes the cache-fill requests carry the name the provider requires.
    • D. The endpoint type only changes how the origin is addressed, by IP rather than by name. The Host header on cache fills is still the client's value, so the provider still returns 403.

    3.2: Configuring Cloud CDN

    16.An operations team invalidates /promo/* and must confirm the purge has finished before announcing a campaign. Where should they confirm completion?

    1. A.Check Cloud Logging for the two entries written per invalidation and proceed once the second entry appears
    2. B.Check Network Intelligence Center with a Connectivity Test to the backend and proceed once the result shows reachable
    3. C.Check Cloud Monitoring for the Cloud CDN cache hit ratio on /promo/* and proceed once it drops to zero
    4. D.Check the URL map with gcloud describe for the invalidation status and proceed once it changes from pending to done
    Show answer & explanation

    Correct answer: A — Check Cloud Logging for the two entries written per invalidation and proceed once the second entry appears

    • A. Each cache invalidation writes one log entry when the request is accepted and a second when it finishes, so the second entry in Cloud Logging is the confirmation that the purge is done.
    • B. A Connectivity Test analyzes the network path and firewall configuration to the backend. It has no visibility into cache invalidations, so a reachable result says nothing about whether the purge finished.
    • C. The cache hit ratio reflects client traffic and cache fill behavior, and it does not fall to zero because a purge completed. It cannot confirm that an invalidation has finished.
    • D. The URL map is the resource the invalidation is requested against, but its description carries no invalidation status. Completion is reported through the invalidation operation and the log entries, so there is no pending-to-done value to wait for.

    3.3: Configuring Cloud DNS

    17.An operations team publishes api.example.com in a public zone. The primary endpoint is in us-central1, and a standby endpoint is in europe-west4. They want queries to return the primary address while it is healthy and shift to the standby only when the primary is unhealthy. Which configuration should they use?

    1. A.Configure a geolocation routing policy with us-central1 as the matched region and europe-west4 as the default, returning the endpoint nearest each resolver.
    2. B.Configure a failover routing policy with a health-checked primary and a backup target, returning the backup only after the primary fails its health checks.
    3. C.Configure one record set with both A records and a 5-second TTL, letting clients retry the standby address when the primary stops responding.
    4. D.Configure a weighted round robin policy with weight 100 on the primary and 1 on the standby, giving the standby a very small share of answers.
    Show answer & explanation

    Correct answer: B — Configure a failover routing policy with a health-checked primary and a backup target, returning the backup only after the primary fails its health checks.

    • A. Incorrect: geolocation routing picks an endpoint from the location of the querying resolver, not from endpoint health. Resolvers outside the matched region get the standby even while the primary is healthy, so the primary is not preferred.
    • B. Correct: a failover (primary-backup) routing policy returns the primary target while its health checks pass and returns the backup target only after the primary is declared unhealthy. This is the active-passive behaviour the team wants.
    • C. Incorrect: a record set with two A records returns both addresses with no health awareness, and the order is not a priority. Clients may try the standby while the primary is healthy, and DNS never withdraws a failed address, so a short TTL does not help.
    • D. Incorrect: weighted round robin with weights of 100 and 1 still returns the standby address for a small share of queries while the primary is healthy. The standby must receive queries only when the primary is unhealthy, so a skewed weight does not meet this.

    3.3: Configuring Cloud DNS

    18.A VPC uses a Cloud DNS forwarding zone for corp.example.com that targets on-premises DNS servers at 10.20.0.53 over HA VPN. Lookups time out. Cloud Router advertises only the VPC subnet ranges to the on-premises network. Select TWO actions that fix the problem.(Select 2)

    1. A.Replace the forwarding zone with a DNS peering zone that targets the on-premises network containing the server at 10.20.0.53.
    2. B.Permit UDP and TCP port 53 from 35.199.192.0/19 in the on-premises firewall policy that protects the DNS server at 10.20.0.53.
    3. C.Create a VPC firewall rule allowing ingress UDP port 53 from 10.20.0.53 to the VMs that send the lookups, at priority 100.
    4. D.Add 35.199.192.0/19 as a custom advertised range on the Cloud Router session that peers with the on-premises router.
    5. E.Add a static route for 169.254.169.254/32 with the HA VPN tunnel as next hop, in the VPC network that hosts the VMs.
    Show answer & explanation

    Correct answers: B, D — Permit UDP and TCP port 53 from 35.199.192.0/19 in the on-premises firewall policy that protects the DNS server at 10.20.0.53.; Add 35.199.192.0/19 as a custom advertised range on the Cloud Router session that peers with the on-premises router.

    • A. Incorrect: a DNS peering zone can only point to another VPC network, never to an on-premises network or server, so it cannot replace a forwarding zone that targets 10.20.0.53.
    • B. Correct: Cloud DNS sends forwarded queries from 35.199.192.0/19 rather than from the VM addresses, so the on-premises firewall must allow DNS (UDP and TCP 53) from that range to reach the server.
    • C. Incorrect: replies to forwarded queries go back to Cloud DNS, not to the VMs, so a VPC ingress rule for the DNS server's address has no effect on this flow.
    • D. Correct: because Cloud Router currently advertises only the subnet ranges, the on-premises network has no route back to 35.199.192.0/19, the range forwarded queries come from. A custom advertised range supplies that route.
    • E. Incorrect: 169.254.169.254 is the metadata-server address VMs use to reach the internal resolver. Routing it into the tunnel does not carry forwarded queries, which originate from 35.199.192.0/19.

    4: Configuring and implementing hybrid and multicloud network interconnectivity

    4.1: Configuring Cloud Interconnect

    19.A payments company needs a 99.99% availability SLA for Dedicated Interconnect between its data centers and one VPC network that hosts workloads in us-east4 and us-west2. Which topology satisfies the SLA requirements?

    1. A.Four connections over two metros, two per metro in separate edge availability domains, with attachments on Cloud Routers in two regions and global dynamic routing.
    2. B.Two connections in separate metros, one per metro, with both VLAN attachments on one Cloud Router in us-east4 and the VPC network set to global dynamic routing mode.
    3. C.Four connections in a single metro, two in each edge availability domain, with every VLAN attachment on one Cloud Router in us-east4 and the VPC set to regional dynamic routing.
    4. D.Two 100 Gbps connections in one metro placed in different edge availability domains, with VLAN attachments on Cloud Routers in both regions plus a backup HA VPN tunnel.
    Show answer & explanation

    Correct answer: A — Four connections over two metros, two per metro in separate edge availability domains, with attachments on Cloud Routers in two regions and global dynamic routing.

    • A. Correct: the 99.99% topology uses two metros with two edge availability domains each, so four connections, with Cloud Routers in two regions and global dynamic routing to allow cross-region failover.
    • B. Incorrect: two connections give only one per metro, so losing a connection in either metro removes that metro's redundancy. The topology needs two attachments per metro and Cloud Routers in two regions.
    • C. Incorrect: a single metro cannot meet 99.99% because one metro-wide event removes every connection. It also leaves just one region and regional routing.
    • D. Incorrect: bigger circuits do not raise the SLA and a single metro remains a shared failure domain. An HA VPN backup does not extend the Interconnect SLA.

    4.2: Configuring a site-to-site IPSec VPN

    20.A team is setting up a Classic VPN route-based tunnel to a branch router. Select TWO configuration elements required for it.(Select 2)

    1. A.Configure the tunnel's local and remote traffic selectors as 0.0.0.0/0, so the VPC routing table decides which packets enter the tunnel.
    2. B.Add a static route in the VPC for the branch CIDR with the Classic VPN tunnel as next hop, so matching packets are forwarded through the tunnel.
    3. C.Create a Cloud Router with a BGP session for the tunnel, so the branch prefixes are learned dynamically and installed as VPC routes.
    4. D.Set the local selector to 10.128.0.0/20 and the remote selector to 192.168.10.0/24 on both ends, so the tunnel carries traffic for those subnets.
    5. E.Set the tunnel's IKE version to IKEv2 rather than IKEv1, so the branch device can negotiate a route-based tunnel with the Classic VPN gateway.
    Show answer & explanation

    Correct answers: A, B — Configure the tunnel's local and remote traffic selectors as 0.0.0.0/0, so the VPC routing table decides which packets enter the tunnel.; Add a static route in the VPC for the branch CIDR with the Classic VPN tunnel as next hop, so matching packets are forwarded through the tunnel.

    • A. Correct. A route-based Classic VPN tunnel is created with traffic selectors of 0.0.0.0/0 on both sides. VPC routes, not the selectors, then decide which packets enter the tunnel.
    • B. Correct. A route-based tunnel carries nothing until a static route to the branch range names it as the next hop. That route is what steers matching packets into the tunnel.
    • C. Incorrect. Cloud Router with BGP is used with HA VPN, and dynamic routing on Classic VPN is deprecated. A route-based Classic tunnel gets its branch prefixes from static routes.
    • D. Incorrect. Narrow selectors that must be mirrored on the peer define a policy-based tunnel. A route-based tunnel uses 0.0.0.0/0 selectors and leaves scoping to routes.
    • E. Incorrect. Classic VPN supports both IKEv1 and IKEv2 with either routing type, so a route-based tunnel does not need IKEv2. The selectors and the static route are what make the tunnel route-based.

    4.2: Configuring a site-to-site IPSec VPN

    21.A company wants Cloud Router to advertise only a summary range of 10.20.0.0/16 to on-premises over its HA VPN, instead of each individual VPC subnet. What should be configured?

    1. A.Set the Cloud Router BGP sessions to custom advertisement mode and add 10.20.0.0/16 as a custom range, with subnet route advertisement switched off.
    2. B.Create a static route for 10.20.0.0/16 with the HA VPN tunnel as next hop, and let the Cloud Router redistribute it to on-premises in place of subnet routes.
    3. C.Change the VPC dynamic routing mode to global, so the Cloud Router advertises one combined range for the subnets instead of a route per subnet.
    4. D.Add an import filter on the on-premises router that accepts only 10.20.0.0/16, with the Cloud Router left in its default advertisement mode.
    Show answer & explanation

    Correct answer: A — Set the Cloud Router BGP sessions to custom advertisement mode and add 10.20.0.0/16 as a custom range, with subnet route advertisement switched off.

    • A. Correct. In custom advertisement mode the Cloud Router sends the prefixes you list instead of the subnet routes. Adding 10.20.0.0/16 as a custom range and not advertising subnets leaves only the summary visible to on-premises.
    • B. Incorrect. The Cloud Router does not redistribute static routes into BGP. It advertises only subnet routes or custom ranges set on the router, so this route would not reach on-premises and the individual subnets would still be advertised.
    • C. Incorrect. Global dynamic routing only widens the advertisement to subnets in every region of the VPC. The Cloud Router does not merge subnets into an aggregate, so each subnet is still advertised as its own prefix.
    • D. Incorrect. In default mode the Cloud Router still advertises each individual subnet and never sends 10.20.0.0/16. A filter on the on-premises side cannot create that summary, and it does not change what the Cloud Router advertises.

    4.1: Configuring Cloud Interconnect

    22.Which circuit sizes can be used when ordering a Dedicated Interconnect connection?

    1. A.1 Gbps circuits, up to eight per connection, or 10 Gbps circuits, up to four per connection, bundled with LACP.
    2. B.10 Gbps circuits, up to eight per connection, or 100 Gbps circuits, up to two per connection, bundled with LACP.
    3. C.10 Gbps circuits, up to four per connection, or 40 Gbps circuits, up to two per connection, bundled with LACP.
    4. D.50 Mbps to 10 Gbps circuits in 50 Mbps steps, up to sixteen circuits per connection, bundled with LACP.
    Show answer & explanation

    Correct answer: B — 10 Gbps circuits, up to eight per connection, or 100 Gbps circuits, up to two per connection, bundled with LACP.

    • A. Incorrect: 1 Gbps circuits are not offered for Dedicated Interconnect, since sub-10 Gbps needs go through Partner Interconnect.
    • B. Correct: Dedicated Interconnect circuits are 10 Gbps or 100 Gbps, and LACP bundles them into up to 80 Gbps or 200 Gbps per connection.
    • C. Incorrect: 40 Gbps circuits are not offered. The two sizes are 10 Gbps and 100 Gbps.
    • D. Incorrect: 50 Mbps increments describe Partner Interconnect attachment capacity, not circuits.

    4.3: Configuring Cloud Router

    23.A company builds two HA VPN tunnels to its on-premises firewall and must define the BGP session interfaces on the Cloud Router. Select TWO configuration choices that are valid.(Select 2)

    1. A.Allocate a distinct /30 from 169.254.0.0/16 to each tunnel, skipping the address blocks that Google Cloud reserves.
    2. B.Assign one usable address of each /30 to the Cloud Router interface and the other usable address to the on-premises peer.
    3. C.Reserve one /29 from a VPC subnet range and let both tunnels' BGP sessions share a single Cloud Router interface address.
    4. D.Configure each tunnel's BGP session on the HA VPN gateway interface IPs, with the firewall's public IP as the peer address.
    5. E.Choose a private RFC 1918 /30 for each tunnel's BGP session, picking ranges that no VPC subnet or peer network uses.
    Show answer & explanation

    Correct answers: A, B — Allocate a distinct /30 from 169.254.0.0/16 to each tunnel, skipping the address blocks that Google Cloud reserves.; Assign one usable address of each /30 to the Cloud Router interface and the other usable address to the on-premises peer.

    • A. Each BGP session needs its own /30 taken from the link-local range 169.254.0.0/16, leaving out the blocks Google Cloud reserves. Two tunnels therefore need two separate /30 ranges.
    • B. A /30 has exactly two usable addresses. One goes on the Cloud Router interface and the other on the on-premises peer, and both ends must be configured to match.
    • C. BGP session addressing does not come from VPC subnet space. Each tunnel's session uses its own link-local /30 with separate Cloud Router and peer addresses, so one shared /29 cannot serve both.
    • D. The gateway interface IPs and the firewall's public IP carry the IPsec transport between the two sites. BGP runs inside the tunnel on link-local addresses, so public IPs cannot be the session endpoints.
    • E. HA VPN BGP sessions require a /30 from the link-local range 169.254.0.0/16. A private RFC 1918 range is not accepted for session addressing, even if it is unused elsewhere.

    4.3: Configuring Cloud Router

    24.A trading firm's Dedicated Interconnect BGP session takes about a minute to withdraw routes after a physical path failure, and the network team wants failover within a few seconds. Which change should they make?

    1. A.Enable BFD on the BGP peer in ACTIVE initialization mode, with matching BFD timers configured on the on-premises router.
    2. B.Set the BGP keepalive interval to one second on the peer, with the same keepalive value configured on the on-premises router.
    3. C.Enable graceful restart on the BGP peer, with matching graceful restart settings configured on the on-premises router.
    4. D.Set advertised route priorities on both attachments, with the lower MED on the primary path and a higher MED on the standby.
    Show answer & explanation

    Correct answer: A — Enable BFD on the BGP peer in ACTIVE initialization mode, with matching BFD timers configured on the on-premises router.

    • A. BFD detects a forwarding-path failure independently of BGP timers, so the session drops in a few seconds instead of waiting out the hold timer. It only works if the on-premises router also runs BFD with compatible timers.
    • B. Cloud Router only accepts keepalive intervals between 20 and 60 seconds, so a one-second value cannot be set. The hold timer is derived from the keepalive interval, so failure detection stays far slower than a few seconds.
    • C. Graceful restart keeps forwarding with stale routes while a peer restarts, which is the opposite of fast withdrawal. Enabling it would make failover after a path failure slower, not faster.
    • D. Route priority (MED) only ranks paths that are already advertised. It does not detect a failure, so the failed path stays installed until BGP itself times out, and failover remains about a minute.

    4.4: Configuring Network Connectivity Center

    25.A company connects Site A and Site B to one VPC with separate HA VPN tunnels, each ending on its own Cloud Router, with no NCC hub. Each site reaches the VPC, but Site A cannot reach Site B. What is the cause, and the fix?

    1. A.The two Cloud Routers advertise different base route priorities (MED) to the VPC; set one matching base priority on the BGP sessions of both sites.
    2. B.Cloud Router does not re-advertise routes learned from one BGP peer to another site; create hybrid spokes for both sites on an NCC hub with site-to-site data transfer.
    3. C.The VPC is in regional dynamic routing mode, so Cloud Router cannot relay Site A's prefixes toward Site B; switch the VPC to global dynamic routing mode.
    4. D.Each site ends on its own Cloud Router, so the sessions share no routes; terminate both HA VPN tunnel pairs on one Cloud Router to relay between sites.
    Show answer & explanation

    Correct answer: B — Cloud Router does not re-advertise routes learned from one BGP peer to another site; create hybrid spokes for both sites on an NCC hub with site-to-site data transfer.

    • A. Incorrect. Base route priority (MED) only ranks competing routes for the same prefix. Both sites already reach the VPC, so their sessions work, and equal priorities would not make Cloud Router pass one site's prefixes to the other.
    • B. Correct. Cloud Router does not pass routes learned from one BGP peer on to another site, so without NCC the sites stay isolated. Hybrid spokes on a hub with site-to-site data transfer enabled provide the exchange.
    • C. Incorrect. Global dynamic routing widens which regions' subnets and learned routes the VPC uses, but Cloud Router still does not re-advertise routes learned from one BGP peer to another. Both sites would remain unable to reach each other.
    • D. Incorrect. The limit is that Cloud Router does not re-advertise routes learned from one BGP peer to another, whether the peers sit on one router or two. Consolidating the tunnels onto one Cloud Router would not make it relay Site A's prefixes to Site B.

    4.4: Configuring Network Connectivity Center

    26.After moving two data centers to NCC hybrid spokes, routes from Site A never appear at Site B. Select TWO things to verify.(Select 2)

    1. A.Confirm that the BGP sessions for both hybrid spokes use an identical base advertised route priority on their Cloud Routers.
    2. B.Confirm that site-to-site data transfer is turned on individually for each hybrid spoke that connects a data center to the hub.
    3. C.Confirm that the VLAN attachments for each hybrid spoke are located in regions that support site-to-site data transfer.
    4. D.Confirm that the Interconnect links behind the VLAN attachments have MACsec encryption turned on at both data center sites.
    5. E.Confirm that the hub also has a VPC spoke attached for the VPC network that hosts the Cloud Routers of both hybrid spokes.
    Show answer & explanation

    Correct answers: B, C — Confirm that site-to-site data transfer is turned on individually for each hybrid spoke that connects a data center to the hub.; Confirm that the VLAN attachments for each hybrid spoke are located in regions that support site-to-site data transfer.

    • A. Incorrect. Advertised route priority only affects which path is preferred. Matching or differing values do not decide whether routes from one site are passed to another.
    • B. Correct. Site-to-site data transfer is a per-spoke setting, not a hub-wide one. If either spoke lacks it, that site's routes are not exchanged with the other.
    • C. Correct. Site-to-site data transfer is available only in supported locations, so an attachment placed outside them cannot take part in route exchange.
    • D. Incorrect. MACsec protects traffic on the Interconnect link itself. NCC does not require link encryption before it exchanges routes between hybrid spokes.
    • E. Incorrect. Hybrid spokes exchange routes with each other through the hub. A VPC spoke for the network hosting the Cloud Routers is not a prerequisite for that exchange.

    5: Managing, monitoring, and troubleshooting network operations

    5.1: Logging and monitoring with Google Cloud Observability

    27.Workloads in a private subnet behind a Cloud NAT gateway intermittently fail to open outbound connections to a single partner API during peak hours. The gateway uses auto-allocated IP addresses with static port allocation of 64 minimum ports per VM. Select TWO actions that confirm and then address port exhaustion.(Select 2)

    1. A.Chart the nat/dropped_sent_packets_count metric by reason OUT_OF_RESOURCES for the gateway, and alert when the count rises above zero
    2. B.Raise the TCP established idle timeout on the gateway to its maximum, so long-lived sessions keep their NAT mappings between bursts
    3. C.Switch the gateway to dynamic port allocation and raise its maximum ports per VM, so busy VMs can draw more ports toward one destination
    4. D.Move the gateway to manual IP allocation and attach extra reserved external addresses, so more NAT source IPs are available
    5. E.Enable Private Google Access on the subnet, so workloads reach the partner API directly, without consuming NAT source ports
    Show answer & explanation

    Correct answers: A, C — Chart the nat/dropped_sent_packets_count metric by reason OUT_OF_RESOURCES for the gateway, and alert when the count rises above zero; Switch the gateway to dynamic port allocation and raise its maximum ports per VM, so busy VMs can draw more ports toward one destination

    • A. Correct: Cloud NAT counts packets it drops with reason OUT_OF_RESOURCES when a VM has no free NAT source ports left. Charting and alerting on that metric confirms exhaustion as the cause of the intermittent failures.
    • B. Incorrect: a longer TCP established idle timeout keeps idle connections' mappings, and the ports they use, reserved for longer. That leaves fewer free ports for new connections to the partner API at peak, so it aggravates exhaustion instead of relieving it.
    • C. Correct: with dynamic port allocation, each VM's port allotment grows from a minimum up to a configured maximum as demand rises. Busy VMs can then open more simultaneous connections to the same destination, which static allocation at 64 ports per VM prevents.
    • D. Incorrect: the gateway already uses auto-allocated addresses, so it adds NAT IPs on its own as needed. The bottleneck is the fixed number of ports each VM gets under static allocation, and extra reserved addresses do not raise that per-VM figure.
    • E. Incorrect: Private Google Access only gives private VMs a path to Google APIs and services. A third-party partner API is not covered, so those connections still go through Cloud NAT and still use its source ports.

    5.3: Using Network Intelligence Center to monitor and troubleshoot common networking issues

    28.A network lead wants proactive detection of misconfigurations without running tests. Select TWO issues that Network Analyzer reports automatically from the project's configuration.(Select 2)

    1. A.A static route whose next-hop VM instance was deleted, causing traffic that matches the route to be dropped
    2. B.A load balancer backend with no firewall rule allowing the Google health check ranges, causing it to fail probes
    3. C.A backend service returning HTTP 5xx errors under load, causing failed client requests during traffic peaks
    4. D.A VM whose egress traffic hits the per-VM bandwidth limit, causing packets to be dropped and sessions to slow
    5. E.An SSL certificate on a target proxy that expires within 30 days, causing client handshake failures
    Show answer & explanation

    Correct answers: A, B — A static route whose next-hop VM instance was deleted, causing traffic that matches the route to be dropped; A load balancer backend with no firewall rule allowing the Google health check ranges, causing it to fail probes

    • A. Network Analyzer inspects routes and flags a static route whose next hop is a deleted instance. The problem is visible in the route definition, so it is found without sending any traffic.
    • B. Network Analyzer checks whether firewall rules permit the Google health check ranges to reach load balancer backends and reports those that cannot be probed. This is a static rule check, not a traffic measurement.
    • C. A rise in 5xx errors under load is runtime behaviour, seen in load balancer logs and Cloud Monitoring. Network Analyzer reads resource settings and does not observe application responses.
    • D. Hitting a per-VM bandwidth limit is a utilization problem, visible in monitoring metrics or Network Topology. Network Analyzer does not measure per-VM throughput, so it cannot report this.
    • E. Certificate expiry is a lifecycle matter handled through certificate management and monitoring tools. It is not among the network configuration insights that Network Analyzer produces.

    5.3: Using Network Intelligence Center to monitor and troubleshoot common networking issues

    29.An operations team enabled VPC Flow Logs on a subnet last week and opened Flow Analyzer in the console, but it cannot query the flow data. The logs are routed to a Cloud Logging bucket they have not changed. What is required?

    1. A.Create a log sink that routes the flow logs to a BigQuery dataset, so Flow Analyzer can read the exported tables
    2. B.Set the flow log aggregation interval to 15 minutes, so Flow Analyzer can work from larger summarized records
    3. C.Grant the Compute Network User role on the host project, so Flow Analyzer can read the records for each subnet
    4. D.Upgrade the log bucket holding the flow logs to Log Analytics, so Flow Analyzer can run SQL queries on the records
    Show answer & explanation

    Correct answer: D — Upgrade the log bucket holding the flow logs to Log Analytics, so Flow Analyzer can run SQL queries on the records

    • A. Flow Analyzer does not use a BigQuery export. It analyzes flow logs where they are stored in the Cloud Logging bucket, through Log Analytics. A sink to BigQuery leaves that bucket unchanged, so the problem stays.
    • B. The aggregation interval only sets how finely flows are summarized when they are recorded. It does not change whether the bucket can be analyzed, and a 15-minute interval would only reduce detail. Flow Analyzer still cannot query the data.
    • C. Compute Network User lets principals attach resources to subnets in a shared VPC. It does not govern analysis of log data. The obstacle is how the logs are stored, not a permission on the subnets.
    • D. Flow Analyzer runs SQL-based analysis through Log Analytics, which requires the log bucket to be upgraded. After the upgrade, the flow logs already in that bucket can be queried.

    5.2: Maintaining and troubleshooting connectivity issues

    30.An e-commerce site serves traffic through a global external Application Load Balancer with backends in us-central1 and europe-west1. The europe-west1 MIG must be taken out of service for emergency maintenance within minutes, without deleting the backend or changing the URL map, and with minimal disruption. Select TWO actions that achieve this.(Select 2)

    1. A.Set the capacity scaler on the europe-west1 backend to 0, so the load balancer stops routing new requests to that group's instances.
    2. B.Point the URL map's default service at a backend service containing only us-central1, so europe-west1 stops receiving new requests.
    3. C.Keep connection draining enabled on the backend service, so in-flight requests on the europe-west1 instances get time to complete.
    4. D.Switch the europe-west1 backend's balancing mode from UTILIZATION to RATE with a low max rate, so the load balancer sends it fewer new requests.
    5. E.Add a firewall rule denying the Google health check ranges to the europe-west1 instances, so the backend is marked unhealthy and removed from rotation.
    Show answer & explanation

    Correct answers: A, C — Set the capacity scaler on the europe-west1 backend to 0, so the load balancer stops routing new requests to that group's instances.; Keep connection draining enabled on the backend service, so in-flight requests on the europe-west1 instances get time to complete.

    • A. A capacity scaler of 0 stops new requests from reaching the europe-west1 backend while it stays configured on the backend service, so restoring it later is a one-field change.
    • B. Repointing the URL map is exactly the change the scenario rules out. It also means editing the routing configuration twice, once now and once to revert it after maintenance.
    • C. Connection draining gives requests already in progress time to finish as traffic leaves the instances, which keeps the maintenance from cutting users off mid-request.
    • D. A low max rate only caps how much traffic europe-west1 receives. Requests below that cap still reach the group, so the MIG is not taken out of service and maintenance would still hit live users.
    • E. Blocking the probes marks the backend unhealthy only after the check interval and unhealthy threshold have passed, so requests keep landing on europe-west1 until then. It also removes traffic indirectly by breaking the health signal instead of controlling the backend's share.

    6: Configuring, implementing and managing a cloud network security solution

    6.1: Configuring Google Cloud Armor policies

    31.A retailer's Cloud Armor policy uses the preconfigured expression evaluatePreconfiguredWaf('sqli-v33-stable', {'sensitivity': 1}). A legitimate checkout form is blocked because a single signature fires on a free-text field. Security wants to keep SQL injection protection for all other traffic. What should the engineer do?

    1. A.Add the offending signature's ID to the opt_out_rule_ids parameter of the expression and keep the remaining sqli-v33-stable signatures at sensitivity 1.
    2. B.Raise the sensitivity level in the expression from 1 to 4 and retest the checkout form against the sqli-v33-stable signature set at that level.
    3. C.Switch the SQL injection rule to preview mode and review the matches in Cloud Logging while the sqli-v33-stable expression stays at sensitivity 1.
    4. D.Change the rule action from deny(403) to throttle and set a per-IP request threshold on the sqli-v33-stable expression at sensitivity 1.
    Show answer & explanation

    Correct answer: A — Add the offending signature's ID to the opt_out_rule_ids parameter of the expression and keep the remaining sqli-v33-stable signatures at sensitivity 1.

    • A. Correct. The opt_out_rule_ids parameter of evaluatePreconfiguredWaf excludes only the signature that matches the free-text field. The other sqli-v33-stable signatures are still evaluated and enforced.
    • B. Incorrect. Higher sensitivity levels enable more signatures, not fewer, so moving from 1 to 4 adds detections and more false positives. The signature that fires on the checkout field is still evaluated.
    • C. Incorrect. Preview mode logs matches without enforcing the rule, so the checkout form passes, but so does injection traffic aimed at the rest of the site. Security needs those requests to keep being blocked.
    • D. Incorrect. The throttle action only limits the request rate of matching clients. Requests below the threshold, including low-volume injection attempts, still reach the backend, so protection for other traffic is lost.

    6.2: Configuring and managing NGFW policies and VPC Firewall rules

    32.An auditor asks for evidence of every denied inbound connection to a VPC network's VMs. Logging is enabled on all explicit firewall rules, yet scanner probes to closed ports never appear in Cloud Logging. Which change fixes this?

    1. A.Enable VPC Flow Logs on the subnets with a 5-second aggregation interval, so connections rejected by the firewall are written to Cloud Logging.
    2. B.Create an ingress deny rule for 0.0.0.0/0 at priority 65535 with logging enabled, so connections hitting no other rule are written to Cloud Logging.
    3. C.Set the existing rules to include metadata in their log entries, so connections that match no explicit rule are written to Cloud Logging as well.
    4. D.Enable Data Access audit logs for the Compute Engine API, so connections dropped by the implied deny ingress rule are written to Cloud Logging.
    Show answer & explanation

    Correct answer: B — Create an ingress deny rule for 0.0.0.0/0 at priority 65535 with logging enabled, so connections hitting no other rule are written to Cloud Logging.

    • A. VPC Flow Logs sample traffic that is permitted to flow and do not record packets the firewall rejects. A shorter aggregation interval only changes reporting granularity, so the denied scanner probes stay invisible.
    • B. Correct. The implied deny-ingress rule cannot have logging turned on. An explicit deny rule at the lowest priority with logging enabled matches the unmatched probes and makes them appear in Cloud Logging.
    • C. Metadata inclusion only adds fields to entries that a rule already writes for connections it matches. Probes that match none of the existing rules are still not logged, so the auditor's gap remains.
    • D. Data Access audit logs record API calls made against Compute Engine, such as reads of rule configuration, not packets on the network. Drops by the implied deny rule never produce audit entries.

    6.2: Configuring and managing NGFW policies and VPC Firewall rules

    33.A web VM has an ingress rule allowing `tcp:443` from `0.0.0.0/0` at priority 1000. An engineer adds an egress deny rule for all protocols to `0.0.0.0/0` at priority 100 and no other egress rule. What happens next?

    1. A.Clients on port 443 and the VM's outbound traffic both keep working, since the priority 1000 allow rule overrides the priority 100 deny rule.
    2. B.Clients on port 443 keep working, since return traffic is tracked, while new outbound connections from the VM are dropped by the deny rule.
    3. C.Clients on port 443 time out and new outbound connections are dropped, since replies need an egress allow with a priority number below 100.
    4. D.Clients on port 443 keep working, and the VM can still make outbound connections on port 443, since the ingress allow rule also covers that port.
    Show answer & explanation

    Correct answer: B — Clients on port 443 keep working, since return traffic is tracked, while new outbound connections from the VM are dropped by the deny rule.

    • A. The allow rule is an ingress rule and the deny is an egress rule, so they are evaluated separately and never compete on priority. The deny matches new outbound connections to any destination, so the VM loses outbound access.
    • B. Correct. Firewall rules are stateful: once the ingress rule admits a connection, its replies are permitted by connection tracking regardless of egress rules. New outbound connections from the VM match the egress deny and are dropped.
    • C. Firewall rules in Google Cloud are stateful, so replies to a connection admitted by the ingress rule are allowed without an egress rule. Clients therefore do not time out, and no extra allow is needed for return traffic.
    • D. An ingress allow rule only governs inbound connections and grants nothing for new outbound ones. The egress deny covers all protocols to 0.0.0.0/0, so outbound connections on port 443 are blocked like any other.

    6.3: Configuring and securing internet egress traffic using Public Cloud NAT and Secure Web Proxy

    34.A network engineer is deploying a Secure Web Proxy gateway in a VPC that has never used Envoy-based load balancers. The gateway deployment fails in us-east4. Select TWO prerequisites that must be put in place in that region.(Select 2)

    1. A.Create a Cloud Router in us-east4 and configure a BGP session on it, to advertise internet-bound routes to the proxies that serve the gateway's traffic.
    2. B.Reserve a static external IPv4 address in the Premium network tier and attach it to the gateway, so that clients can reach the proxy endpoint over the internet.
    3. C.Deploy a Cloud NAT gateway with manually allocated IP addresses in us-east4, to give the proxies a fixed source address when they forward client requests.
    4. D.Create a `proxy-only` subnet in us-east4 with purpose `REGIONAL_MANAGED_PROXY` in the gateway's VPC, to provide the address range the gateway's proxies draw from.
    5. E.Issue or import a regional TLS certificate in Certificate Manager in us-east4 and reference it from the gateway, to secure the connection to the proxy endpoint.
    Show answer & explanation

    Correct answers: D, E — Create a `proxy-only` subnet in us-east4 with purpose `REGIONAL_MANAGED_PROXY` in the gateway's VPC, to provide the address range the gateway's proxies draw from.; Issue or import a regional TLS certificate in Certificate Manager in us-east4 and reference it from the gateway, to secure the connection to the proxy endpoint.

    • A. Incorrect: deploying the gateway does not involve a Cloud Router or BGP session. The proxies are managed by Google and clients reach the gateway directly at its internal address and port, so no route advertisement is needed.
    • B. Incorrect: a Secure Web Proxy gateway is a regional resource with an internal address in the client subnet. It is not fronted by an external IP address, so reserving one does not address the failed deployment.
    • C. Incorrect: Cloud NAT is not a prerequisite for deployment. It is only an optional way to give egress traffic fixed source addresses, and the gateway accepts explicit-proxy requests without it.
    • D. Correct: Secure Web Proxy runs on managed Envoy proxies. A VPC that has never used Envoy-based load balancers has no proxy-only subnet, so the region needs one with purpose REGIONAL_MANAGED_PROXY.
    • E. Correct: the gateway terminates client TLS with a regional certificate stored in Certificate Manager. The certificate must exist in the same region and be referenced when the gateway is created.

    6.4: Configuring self-managed network virtual appliance and Packet Mirroring

    35.Which type of load balancer must front a Packet Mirroring collector instance group?

    1. A.Internal passthrough Network Load Balancer
    2. B.Regional internal proxy Network Load Balancer
    3. C.External passthrough Network Load Balancer
    4. D.Global external proxy Network Load Balancer
    Show answer & explanation

    Correct answer: A — Internal passthrough Network Load Balancer

    • A. Correct. The collector destination is the forwarding rule of an internal passthrough NLB enabled for mirroring, which delivers packets unchanged to the collectors.
    • B. Incorrect. Proxy Network Load Balancers terminate TCP and re-originate it, so the mirrored packets would not be delivered as they are.
    • C. Incorrect. External passthrough NLBs use external addresses and are not valid collector destinations, which must be internal.
    • D. Incorrect. A global external proxy Network Load Balancer is externally addressed and terminates TCP connections at a proxy. It cannot be a mirroring collector destination, which must be an internal passthrough load balancer that delivers the mirrored packets unchanged.

    Want the full experience?

    These are just samples. Practice the full Google Cloud Professional Cloud Network Engineer question bank in quiz mode — free, no signup, with domain practice and exam simulation.