Observability Lab Part 4 Quiz
Last Updated: September 13, 2026
- How do k6 VUs relate to RPS?
- A) One VU always equals one RPS.
- B) VUs are concurrent execution contexts; RPS also depends on requests, latency and sleep.
- C) VUs equal the node count.
- D) RPS is independent of response time.
Show answer
Answer: B) VUs are concurrent execution contexts; RPS also depends on requests, latency and sleep.
Equal VUs do not imply equal throughput across different workloads and waits.
- How should failed k6 checks fail CI?
- A) Calling check always exits with code 1.
- B) Set a checks/failure-rate threshold and inspect the exit code.
- C) A summary JSON file proves success.
- D) Count only HTTP 200 responses.
Show answer
Answer: B) Set a checks/failure-rate threshold and inspect the exit code.
HTTP success and business success differ; assert JSON, IDs and payment state too.
- Which order IDs should the load test read?
- A) Random IDs from 1 to 1000.
- B) IDs actually created by the test.
- C) Customer IDs.
- D) HTTP status codes.
Show answer
Answer: B) IDs actually created by the test.
Use creation-response IDs so random 404s do not contaminate the workload.
- What should scale directly with SQS backlog?
- A) Always only the API producer.
- B) The consumer processing that queue.
- C) Alertmanager replicas.
- D) The queue name.
Show answer
Answer: B) The consumer processing that queue.
KEDA reads queue attributes and manages scaling with HPA; it does not consume messages.
- When does KEDA cooldownPeriod apply?
- A) Every 10-to-9 replica change.
- B) When scaling to zero after the last active trigger.
- C) EC2 boot delay.
- D) Log retention.
Show answer
Answer: B) When scaling to zero after the last active trigger.
Inspect HPA behavior and stabilization windows for scaling within 1..N replicas.
- What does an HPA scale-down stabilization window do?
- A) Freezes all nodes.
- B) Considers the highest replica recommendation within the window.
- C) Uses only instantaneous CPU.
- D) Deletes a Pod every configured interval.
Show answer
Answer: B) Considers the highest replica recommendation within the window.
It is not simply an unconditional fixed delay; policies and recommendation history matter.
- Which problem can Karpenter normally address?
- A) A misspelled image name.
- B) An unschedulable Pod needing capacity compatible with its NodePool.
- C) An application syntax error.
- D) An incorrect database password.
Show answer
Answer: B) An unschedulable Pod needing capacity compatible with its NodePool.
More nodes do not fix image pulls or application bugs; inspect scheduling reasons first.
- What does kube_deployment_status_replicas measure?
- A) Always ready replicas.
- B) Total Deployment replicas, distinct from ready replicas.
- C) All controller replicas including Rollouts.
- D) Node count.
Show answer
Answer: B) Total Deployment replicas, distinct from ready replicas.
Use kube_deployment_status_replicas_ready for ready replicas; Rollouts need their own state/exporter.
- How should Running Pods be counted from phase metrics?
- A) Count all Running series without filtering values.
- B) Sum the Running phase 0/1 gauges directly.
- C) Sum Pod-name lengths.
- D) Always return three.
Show answer
Answer: B) Sum the Running phase 0/1 gauges directly.
Running phase series can exist with value zero, so an unfiltered count overcounts. Summing the gauges returns zero for all-Pending Pods and remains absent when telemetry is missing.
- What is evidence of a successful load experiment?
- A) A table copied from expected documentation numbers.
- B) Measured requests, errors, latency, queue, replicas, nodes and exit status.
- C) Successful Job creation.
- D) Only a lower node count.
Show answer
Answer: B) Measured requests, errors, latency, queue, replicas, nodes and exit status.
Do not present unmeasured throughput, availability or cost savings as results.