Prepare and test your skills

Prepare and test your skills

Free sample question with answers and detailed explanation below. The Professional Cloud Network Engineer practice bank has an additional 2,437 practice questions.Try 10 free realistic questions: instant score, no credit card, and no account required.
You manage a GKE Standard cluster (version 1.33) with heterogeneous node pools (4 to 64 vCPUs per node) running workloads that issue high-QPS DNS queries for internal cluster.local Services and external domains without trailing dots. During traffic bursts and scale-ups, you observe:
CT: Map insertion failed packet drops.kube-dns hits its 200-concurrent-TCP-connection limit.gcloud container clusters update with --update-addons=NodeLocalDNS=ENABLED, node-local-dns Pods do not deploy due to an active maintenance exclusion.Which combination of actions completes the NodeLocal DNSCache rollout, eliminates conntrack exhaustion, and prevents kube-dns-upstream TCP i/o timeout errors?
gcloud container clusters upgrade with the current --cluster-version, scale kube-dns-autoscaler solely using nodesPerReplica without an explicit max, and add 169.254.169.254 to stubDomains for cluster.local in the kube-dns ConfigMap so NodeLocal DNSCache bypasses kube-dns.This option correctly forces node re-creation to deploy node-local-dns, but it scales kube-dns-autoscaler purely by node count without a max cap and attempts to redirect cluster.local resolution to the Compute Engine metadata server (169.254.169.254) using stubDomains.
169.254.169.254) does not host internal Kubernetes cluster.local records when kube-dns is the cluster DNS provider. Routing cluster.local queries away from kube-dns causes internal Service lookups to fail.nodesPerReplica ignores the wide compute disparity (4 to 64 vCPUs per node). High-core nodes hosting dozens of Pods generate disproportionate DNS load that node-count scaling fails to provision for.max value in kube-dns-autoscaler allows kube-dns replicas to scale unchecked during large cluster expansions, overwhelming the GKE control plane VM with Kubernetes API watches.kube-dns: kube-dns is authoritative for cluster.local Services with active Endpoints; 169.254.169.254 is only used by NodeLocal DNSCache for external domains (or by Cloud DNS for GKE).stubDomains from the kube-dns ConfigMap and queries them using both TCP and UDP, but stubDomains are meant for custom external/private domains.ndots:5 amplification: Without trailing-dot FQDNs or ndots:1, external domain queries still trigger five cluster.local and internal search queries first.This would be correct if coresPerReplica and an explicit max were configured in kube-dns-autoscaler, cluster.local remained pointed to kube-dns-upstream over TCP, and ndots:1 or trailing-dot FQDNs were used instead of modifying stubDomains.
In heterogeneous GKE Standard clusters, kube-dns-autoscaler computes max(ceil(cores / coresPerReplica), ceil(nodes / nodesPerReplica)), making coresPerReplica essential for matching replica capacity to dense 64-vCPU nodes.
gcloud container clusters upgrade with the current --cluster-version, tune coresPerReplica, nodesPerReplica, and an explicit max in the kube-dns-autoscaler ConfigMap, and configure application Pods to use FQDNs with trailing dots or ndots:1.NodeLocal DNSCache runs a caching DNS daemon as a DaemonSet on every cluster node, serving Pod lookups locally while forwarding cache misses according to query type: cluster.local queries go to kube-dns-upstream over TCP, custom stub domains use TCP and UDP, and external queries go directly to the local metadata server (169.254.169.254). Pairing this with kube-dns-autoscaler tuning and ndots optimization addresses both node-level bottlenecks and upstream kube-dns saturation.
gcloud container clusters upgrade with the existing --cluster-version forces a managed node upgrade. GKE re-creates the nodes and automatically applies addon.gke.io/node-local-dns-ds-ready=true.node-local-dns cache on the same node bypass Linux kernel and Dataplane V2 conntrack table entries, eliminating CT: Map insertion failed packet drops.kube-dns-upstream TCP timeouts: Lowering coresPerReplica and nodesPerReplica scales kube-dns proportionally across heterogeneous 4-to-64-vCPU nodes, while an explicit max protects the control plane VM. Meanwhile, using trailing-dot FQDNs or ndots:1 stops external lookups from traversing five internal search suffixes (for both A and AAAA records), preventing unnecessary cluster.local TCP queries from exhausting kube-dns's 200-connection limit.NXDOMAIN) resolve on-node without cross-node hops.replicas = max(ceil(cores / coresPerReplica), ceil(nodes / nodesPerReplica)) accounts for high-vCPU nodes.kube-dns completely via 169.254.169.254.max caps concurrent API watches from kube-dns replicas.Unlike manual node labeling or Spot VM workarounds, this solution follows GKE's supported lifecycle mechanics, removes conntrack pressure at the kernel layer, and minimizes TCP fallback traffic to kube-dns-upstream.
addon.gke.io/node-local-dns-ds-ready=true label to all existing nodes to schedule node-local-dns immediately, set max to 500 in kube-dns-autoscaler to pre-provision 500 kube-dns replicas, and configure node-local-dns to forward cluster.local queries to kube-dns-upstream over UDP.This approach attempts to bypass GKE's managed node re-creation workflow by manually labeling nodes, misinterprets how the max parameter controls kube-dns-autoscaler, and alters the upstream transport protocol used between NodeLocal DNSCache and kube-dns.
addon.gke.io/node-local-dns-ds-ready=true label automatically only after re-creating nodes to configure host networking and Pod DNS settings. Manually applying this label schedules node-local-dns onto unconfigured nodes without updating Pod /etc/resolv.conf or host bindings.max: 500 in the kube-dns-autoscaler ConfigMap does not create 500 replicas; it only sets an upper limit. Without tuning nodesPerReplica or coresPerReplica, replica counts remain too low during traffic spikes.kube-dns Pods via kube-dns-upstream over TCP, not UDP.node-local-dns without node re-creation leaves Pods routing UDP queries to the central kube-dns ClusterIP, sustaining CT: Map insertion failed drops.kube-dns replicas watching the Kubernetes API can overwhelm the GKE control plane VM.kube-dns, TCP/UDP for custom stubDomains and upstreamNameservers, and the local metadata server for external URLs.This option is never valid as written, but it would become correct if nodes were re-created via gcloud container clusters upgrade with the current version, coresPerReplica and nodesPerReplica were lowered with a conservative max, and kube-dns-upstream remained on TCP.
GKE strictly prohibits manually adding addon.gke.io/node-local-dns-ds-ready=true because node-level kubelet and CNI state must be rebuilt during the managed upgrade cycle.
kube-dns Deployment on a Spot VM node pool to absorb the query load without re-creating existing nodes, configure dnsPolicy: None with nameservers: ["169.254.20.10"] in application Pod specs, and create a NetworkPolicy ipBlock for the kube-dns ClusterIP.This option skips NodeLocal DNSCache node re-creation, shifts kube-dns Pods onto preemptible Spot VMs, hardcodes the NodeLocal DNSCache IP (169.254.20.10) in Pod dnsConfig, and adds a NetworkPolicy ipBlock rule.
node-local-dns Pods are not running on the nodes and nothing is listening on 169.254.20.10, causing immediate DNS failures for any Pod configured with dnsPolicy: None pointing to that address.node-local-dns daemon, Pod DNS queries continue to traverse the CNI dataplane to remote kube-dns Pods, creating conntrack entries and triggering CT: Map insertion failed drops.kube-dns availability: Spot VMs can be preempted at any time; running critical DNS infrastructure on Spot VMs causes sudden resolution outages and EAI_AGAIN errors.kube-dns Deployment without modifying or disabling kube-dns-autoscaler causes the autoscaler to overwrite the manual replica count.NetworkPolicy rule: With GKE Dataplane V2 enabled, an ipBlock rule for kube-dns ClusterIP is not required for NodeLocal DNSCache operation.kube-dns replicas, high-density nodes still exhaust local conntrack maps under high UDP DNS query rates.This approach would still be flawed on Spot VMs; to fix the cluster properly, you must complete the node re-creation for NodeLocal DNSCache, schedule kube-dns on standard on-demand nodes, and tune kube-dns-autoscaler.
The dnsPolicy: None workaround pointing to kube-dns-upstream is only used for hostNetwork: true Pods experiencing ClusterFirstWithHostNet timeouts on Dataplane V2, not for general workloads targeting 169.254.20.10.
Answer them yourself and get your score the moment you finish. No account needed.
Full access includes
Keep the momentum going with these hand-picked practice scenarios
Last updated Validated