M2.6 module quiz
Ten questions. 80% to pass (8 of 10). Open book, unlimited retakes.
Answers are collapsed under each question. Answer first, then check.
Q1
Kubernetes is structurally hard to cost-optimize because:
A. Cloud providers do not expose any per-cluster cost figures at all
B. Cost lives at the node level while decisions live at the workload level
C. Pods cannot be scheduled onto specific nodes
D. Cluster-level metrics are simply unavailable
Show answer
Correct: B. You pay for instance-hours and you reason in pod-hours. That mismatch, plus the fact that stopping the cluster loses everything, is why workload-level scheduling exists.
Q2
Namespace-level Stop on an EKS, GKE or AKS cluster:
A. Deletes the whole namespace along with absolutely everything that sits inside it as well
B. Cordons the nodes and waits for pods to drain
C. Scales Deployments and StatefulSets to 0 and suspends CronJobs, saving replicas, HPA and PDB
D. Is available on EKS only, not AKS or GKE
Show answer
Correct: C. Start restores all of it. One shared implementation drives all three clouds with per-cloud auth and UID parsing injected.
Q3
System namespaces are:
A. Rejected at the executor before any kube call
B. Scheduled anyway, with a lower priority class
C. Scaled to one replica rather than to zero
D. Handled by the provider’s own controller
Show answer
Correct: A. kube-*, zopnight-*, istio-*, helm-*, default and kube-node-lease. Rejecting at the executor rather than only in the UI means an API caller cannot bypass it.
Q4
Scaling a Deployment to zero interacts with an HPA by:
A. Ignoring the HPA and scaling underneath it anyway
B. Failing outright if an HPA exists on it
C. Removing the HPA on Stop and restoring it on Start
D. Setting the HPA’s minReplicas to zero
Show answer
Correct: C. An HPA left in place would immediately scale the Deployment back up, so the HPA is part of what gets saved and restored.
Q5
CronJobs are handled by:
A. Deleting and recreating them
B. Scaling their parent Deployment
C. Deleting their Jobs
D. Suspending them
Show answer
Correct: D. Suspend skips executions without losing the definition or its history. Deleting would lose both, and recreating would reset the schedule’s relationship to the clock.
Q6
Scheduling a StatefulSet is dangerous when:
A. It participates in quorum, where scaling below the quorum threshold breaks the cluster
B. It has more than one replica currently configured across the whole StatefulSet
C. It has a PersistentVolumeClaim attached
D. It uses an ordered update strategy
Show answer
Correct: A. PVCs are preserved through a scheduling cycle, so storage is not the risk. Quorum is, which is why scale-to-one on etcd, ZooKeeper or Kafka is the classic incident.
Q7
The cluster autoscaler interacts with scale-to-zero by:
A. Preventing the scale-down from happening
B. Removing now-empty nodes, which is where the actual saving comes from
C. Immediately adding the removed nodes straight back into the pool
D. Ignoring workloads that were scaled down
Show answer
Correct: B. Scaling pods to zero saves nothing by itself: the nodes still bill. The saving arrives when the cluster autoscaler reclaims the emptied nodes.
Q8
Cross-cluster grouping lets you:
A. Apply one schedule across workloads in several clusters
B. Merge two clusters into one
C. Share a single control plane
D. Move workloads freely between two separate clusters
Show answer
Correct: A. The three patterns are global, per-region and workload-based grouping, and the right one depends on what the schedule is actually expressing.
Q9
ZopNight does NOT discover, for a K8s workload:
A. Which container images are currently running
B. The replica count per workload
C. Resource requests and limits set
D. In-cluster network policy enforcement decisions
Show answer
Correct: D. Discovery reads the workload objects. What the CNI decides at runtime is not a workload property and does not appear.
Q10
The service-dependency problem in scale-to-zero is:
A. Services themselves cannot be scaled to zero
B. Service endpoints are simply not updated when the workload scales to zero
C. A dependent workload calling a scaled-to-zero service fails rather than waiting
D. DNS caching holds the old endpoint address
Show answer
Correct: C. Which is why scale-to-zero suits leaf workloads and environments stopped as a unit, and suits a shared dependency far less well.
What’s next
Back to K8s workload scheduling.