Skip to content

Professional Cloud Network Engineer · Domain 5: Managing, monitoring, and troubleshooting network operations · 14% of the exam

Task 5.2: Maintaining and troubleshooting connectivity issues.

Draining and redirecting traffic on Application Load Balancers, and troubleshooting VPNs, Cloud Interconnect and BGP peering with VPC Flow Logs, firewall logs and Packet Mirroring.

Study it

  • Troubleshooting connectivity: draining, VPNs, Cloud Interconnect, BGP sessions, VPC Flow Logs, firewall logs and Packet Mirroring

    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

A global external Application Load Balancer's backend service has two managed instance groups, mig-v1 and mig-v2, and a connection draining timeout of 600 seconds. To retire v1, an engineer removed mig-v1 from the backend service, and large uploads in progress on v1 VMs were cut off at once. How should v1 be retired next time so that in-progress requests get their 600 seconds?

  1. AResize mig-v1 to zero instances, then remove the group from the backend service right away, while its VMs are still draining
  2. BResize mig-v1 to zero instances, wait for draining to finish, then remove the empty group from the backend service
  3. CTurn on connection draining on failover in the backend service, then remove mig-v1
  4. DSet the connection draining timeout on the URL map's path rule for the upload path
Show the answer and why
  • AResize mig-v1 to zero instances, then remove the group from the backend service right away, while its VMs are still draining

    Incorrect

    The resize starts draining, but removing the group from the backend service is the case draining doesn't cover, so the uploads are cut again.

  • BResize mig-v1 to zero instances, wait for draining to finish, then remove the empty group from the backend service

    Correct

    Removing instances from a managed instance group, for example by resizing it, triggers connection draining for those VMs.

  • CTurn on connection draining on failover in the backend service, then remove mig-v1

    Incorrect

    That setting belongs to the failover policy of internal passthrough Network Load Balancers, not to removing a backend group here.

  • DSet the connection draining timeout on the URL map's path rule for the upload path

    Incorrect

    The connection draining timeout is a setting of the backend service, not of a URL map.

Connection draining gives in-progress requests time to complete when a VM is removed from an instance group, an endpoint is removed from a zonal NEG, or a managed instance group removes instances through resize, delete, recreate, abandon, autoscaling or the updater. It doesn't apply when the whole backend group is removed from the backend service, which is what cut the uploads.

Question 2 · choose 1

A Cloud Router BGP session over a VLAN attachment went down after the data center briefly advertised about 6,200 prefixes. Cloud Logging shows "Sent NOTIFICATION 6/1 (CEASE/MAX_PREFIXES_REACHED)". The data center team has since fixed its filter and now advertises 800 prefixes, but the session stays down after an hour. What should the network engineer do?

  1. AWait for the peer's hold timer to expire, after which Cloud Router re-establishes the session by itself
  2. BRequest a higher quota for learned routes in the region, then wait for the session to recover
  3. CEnable BFD on the session so that Cloud Router detects the corrected peer sooner
  4. DDisable the BGP peer on the Cloud Router and then enable it again
Show the answer and why
  • AWait for the peer's hold timer to expire, after which Cloud Router re-establishes the session by itself

    Incorrect

    After this shutdown Cloud Router doesn't try to re-establish the session until the peering is reset manually.

  • BRequest a higher quota for learned routes in the region, then wait for the session to recover

    Incorrect

    The session closed at the per-session limit of 5,000 received prefixes; the advertised count is already below it.

  • CEnable BFD on the session so that Cloud Router detects the corrected peer sooner

    Incorrect

    BFD speeds up failure detection on a working session; it doesn't reset a session that Cloud Router shut down.

  • DDisable the BGP peer on the Cloud Router and then enable it again

    Correct

    The documented recovery is to reduce the prefixes, then disable and re-enable the BGP peer to reset the session.

If an on-premises router advertises more than 5,000 prefixes, Cloud Router closes the BGP session and logs the CEASE/MAX_PREFIXES_REACHED notification. It doesn't attempt to re-establish the session until the peering is reset: reduce the prefixes, disable the BGP peer, then enable it.

Question 3 · choose 1

A partner reports that its TCP connections to port 8443 on a VM are refused by the network. The VPC network's global network firewall policy has logging enabled on every rule, and no rule mentions port 8443, yet no connection record for these attempts appears in Cloud Logging. The team wants future dropped attempts like these to produce connection records. What should it do?

  1. AAdd an ingress deny rule for all sources at the lowest priority of the policy, with logging enabled
  2. BEnable logging on the network's implied deny ingress rule
  3. CAsk the partner to test with ICMP pings instead, so that each dropped packet is logged
  4. DWait 10 minutes, because records are written only after a connection has been idle that long
Show the answer and why
  • AAdd an ingress deny rule for all sources at the lowest priority of the policy, with logging enabled

    Correct

    The attempts now fall through to the implied deny ingress rule, which can't be logged; a rule of their own with logging records them.

  • BEnable logging on the network's implied deny ingress rule

    Incorrect

    Firewall policy rules logging isn't supported for the implied deny ingress and implied allow egress rules.

  • CAsk the partner to test with ICMP pings instead, so that each dropped packet is logged

    Incorrect

    Firewall policy rules logging records only TCP and UDP connections, so pings would produce no records at all.

  • DWait 10 minutes, because records are written only after a connection has been idle that long

    Incorrect

    The 10-minute idle rule concerns connections already logged; these attempts match no logged rule at all.

Connection records are created only for rules that have logging enabled and only for TCP and UDP. Traffic that matches no rule meets the implied deny ingress rule, for which logging isn't supported, so it leaves no trace. A catch-all deny rule with logging, placed below every other rule, makes those drops visible.

Question 4 · choose 1

During a migration, requests to /legacy/* on a global external Application Load Balancer must make clients request the same paths from a new host, and no backend may receive these requests. How should the URL map handle them?

  1. AA route rule for /legacy/* with a urlRedirect action
  2. BA route rule for /legacy/* with request mirroring to the new host
  3. CA route rule for /legacy/* with a URL rewrite of the host
  4. DA connection draining timeout on the backend serving /legacy/*
Show the answer and why
  • AA route rule for /legacy/* with a urlRedirect action

    Correct

    The redirect action responds to matching requests with an HTTP redirect instead of forwarding them to a backend.

  • BA route rule for /legacy/* with request mirroring to the new host

    Incorrect

    Mirroring doesn't answer the client with a redirect, so clients wouldn't move to the new host.

  • CA route rule for /legacy/* with a URL rewrite of the host

    Incorrect

    A rewrite doesn't answer the client with a redirect either, so the request still goes to a backend.

  • DA connection draining timeout on the backend serving /legacy/*

    Incorrect

    Connection draining lets in-progress requests finish; it doesn't send redirects to clients.

The traffic management features of the global external Application Load Balancer include routing, rewrites, mirroring and redirects. A route rule with urlRedirect answers matching requests with an HTTP redirect, so they never reach a backend service.

Question 5 · choose 1

A new HA VPN tunnel to a peer device doesn't come up, and the Cloud VPN logs show a no-proposal-chosen error. Both sides use IKEv2, and the shared secret was checked. What should the network engineer check next?

  1. AThat UDP ports 500 and 4500 are forwarded by a NAT device in front of the peer
  2. BThat the Cloud Router BGP session uses the right peer ASN
  3. CThat the peer proposes at least one cipher that Cloud VPN supports
  4. DThat the peer's ciphers match Cloud VPN's list exactly, entry by entry
Show the answer and why
  • AThat UDP ports 500 and 4500 are forwarded by a NAT device in front of the peer

    Incorrect

    That applies to peers behind one-to-one NAT; no-proposal-chosen points to the cipher negotiation.

  • BThat the Cloud Router BGP session uses the right peer ASN

    Incorrect

    BGP runs only after the tunnel is up; this error happens during IKE.

  • CThat the peer proposes at least one cipher that Cloud VPN supports

    Correct

    No-proposal-chosen means the gateways couldn't agree on ciphers; for IKEv2 each side needs at least one common cipher.

  • DThat the peer's ciphers match Cloud VPN's list exactly, entry by entry

    Incorrect

    An exact match is the rule for IKEv1; IKEv2 needs only one common cipher.

A no-proposal-chosen error in the VPN logs means Cloud VPN and the peer couldn't agree on a set of ciphers. With IKEv1 the cipher sets must match exactly; with IKEv2 each gateway must propose at least one common cipher from the supported list.

Question 6 · choose 1

A VM mirrored by VPC Packet Mirroring for both directions receives 1 Gbps and sends 1 Gbps of its own traffic. How much egress traffic does the VM carry in total?

  1. A1 Gbps, because mirrored copies don't count against the VM
  2. B3 Gbps: its own 1 Gbps plus mirrored copies of both directions
  3. C2 Gbps, because only the received traffic is copied
  4. D4 Gbps, because each copy is sent twice for redundancy
Show the answer and why
  • A1 Gbps, because mirrored copies don't count against the VM

    Incorrect

    Mirroring traffic consumes bandwidth on the mirrored instance.

  • B3 Gbps: its own 1 Gbps plus mirrored copies of both directions

    Correct

    The example in the documentation gives 1 Gbps ingress and 3 Gbps egress for this case.

  • C2 Gbps, because only the received traffic is copied

    Incorrect

    Both directions were configured, so both are copied.

  • D4 Gbps, because each copy is sent twice for redundancy

    Incorrect

    Each mirrored packet is copied once to the collector.

Packet Mirroring copies traffic from the mirrored VMs to collectors, and those copies use the mirrored VM's bandwidth. Mirroring 1 Gbps in and 1 Gbps out adds 2 Gbps of egress, for 3 Gbps in total.

Google Cloud documentation

Practice domain 5 →Practice all domains →