Tutorial1 min readDraft
Restricted cluster operations workshop
Design, deploy, update, and recover a restricted AI workload using an approved lab profile.
Last content review 2026-08-03Included in SAI-270
Scenario
Plan a restricted cluster deployment for one model service and private RAG workload. The cluster has internal registries, limited accelerators, no direct runtime internet, and explicit recovery targets.
Tasks
- Draw cluster, node, namespace, network, storage, identity, registry, model, telemetry, and administration boundaries.
- Declare all internal dependencies and the tested compatibility set.
- Create admission, security-context, identity, secret, network, storage, quota, and break-glass baselines.
- Define accelerator scheduling, isolation, health, model loading, and failure response.
- Plan a disconnected platform and runtime update with rollback.
- Design backups and execute or tabletop a restore and rebuild.
- Define operational indicators and collect change and recovery evidence.
Deliverables
Submit the dependency map, platform baseline, workload manifests or platform-neutral specification, GPU operations plan, update runbook, backup catalog, recovery record, and evidence package.
Do not modify a production cluster for this public exercise.