Rollout, disruption and recovery evidence
This required 22–32-hour integrated package strengthens the existing guided labs. Allocation: lessons/predictions 5–7 hours, cluster experiments 11–17, assessment/repair 6–8. Author estimates require learner timing pilots and are included in the semester workload contract. Reuse the same service and evidence; avoid creating a second portfolio.
Prerequisite bridge
Bring the preceding module's implementation, risk record and failure observations. Separate assumptions, model results and actual runtime evidence. Review the specific prerequisite named in each lesson before its exercise. Missing runtime tools leave execution evidence pending; a written prediction is not an executed experiment.
Lesson sequence
- Give each probe one operational purpose.
- Distinguish voluntary eviction from rollout availability.
- Finish or safely retry work during termination.
Integrated experiment
Use the original local API/worker/database cluster. Freeze versions, image digests, network plugin and storage assumptions. Run normal traffic first, then slow startup, downstream loss, voluntary eviction, bad-image rollout and interrupted worker acknowledgement one at a time. Keep prediction, exact injection, events/status, client result, recovery and limitation for each. A cluster without a network-policy-enforcing plugin cannot prove a denied-traffic policy merely by accepting its manifest. Include an allowed control request alongside every denied case. Clean up only the dedicated lab environment. Kubernetes/container tools are needed for actual execution; a tabletop alone cannot satisfy the cluster gates.
Complete the assessment. Original guided labs, security controls, deployment/testing and peer requirements remain mandatory. Use synthetic data and a disposable local environment where supported; no cloud purchase is required by this extension.