Spot Capacity Orchestrator · View 32 of 32 · 7 · Assurance
Reading the matrix
- Containment is designed in before the failure: caps before correlated loss, periodic checkpoints before no-notice loss, a pre-provisioned floor before a fallback storm. Recovery is what the platform does afterwards, and none of it requires the failure to have been foreseen in detail.
The ones that depend on assumptions
- No-notice termination assumes 2% of reclamations. Hall loss assumes committed capacity exists in the other halls to rebuild the floor. Quota exhaustion assumes the private cloud team raises quotas within a working day. Each is monitored, and each has a card on its own view.
Exercised
- The daily 1% drill covers single-pool reclaim, drain, overrun detection and no-notice handling. A quarterly game day takes a hall's reclaimable capacity and the control plane down together, because those two are most likely to fail together.