15.267 14.9.27 Runbooks operativos
| ID | Incidente | Procedimiento central | Salida |
|---|---|---|---|
| RB-01 | Camera disconnected | Check PoE/link/device identity; capture logs; bounded restart; validate first synchronized bundle. | IncidentRecord with timeline, evidence, actions, validation and closure |
| RB-02 | Frame gaps | Inspect sequence, network and writer latency; classify expected vs loss; preserve evidence. | IncidentRecord with timeline, evidence, actions, validation and closure |
| RB-03 | Clock skew | Stop metric cross-stream claims; restart clock epoch; recalibrate synchronization. | IncidentRecord with timeline, evidence, actions, validation and closure |
| RB-04 | Disk pressure | Pause reprocessing, evict scratch/cache, verify backup, stop new capture before unsafe threshold. | IncidentRecord with timeline, evidence, actions, validation and closure |
| RB-05 | Filesystem/read-only | Stop writers, snapshot diagnostics, serve read-only, repair or replace storage. | IncidentRecord with timeline, evidence, actions, validation and closure |
| RB-06 | Metadata database incident | Quiesce, consistent copy, integrity check, restore/reconcile artifacts; never remove WAL blindly. | IncidentRecord with timeline, evidence, actions, validation and closure |
| RB-07 | OOM/crash loop | Collect cgroup/tegrastats, lower concurrency/profile, quarantine deterministic input. | IncidentRecord with timeline, evidence, actions, validation and closure |
| RB-08 | Thermal throttling | Verify cooling/power mode, reduce load, document degraded SLO. | IncidentRecord with timeline, evidence, actions, validation and closure |
| RB-09 | Invalid calibration | Block world metrics; recapture/validate calibration and scene revision. | IncidentRecord with timeline, evidence, actions, validation and closure |
| RB-10 | Model activation failure | Reject/rollback, preserve verification output, compare compatibility matrix. | IncidentRecord with timeline, evidence, actions, validation and closure |
| RB-11 | Sync backlog | Validate destination/auth/bandwidth; reconcile manifests; preserve local retention. | IncidentRecord with timeline, evidence, actions, validation and closure |
| RB-12 | Telemetry overload | Reduce sampling/cardinality, retain critical events, inspect feedback loop. | IncidentRecord with timeline, evidence, actions, validation and closure |
| RB-13 | Power-loss recovery | Reconcile staging, leases, WAL, partial uploads and active deployment slot. | IncidentRecord with timeline, evidence, actions, validation and closure |
| RB-14 | Consent withdrawal | Block access/export, tombstone, lineage deletion, remote propagation and audit. | IncidentRecord with timeline, evidence, actions, validation and closure |
| RB-15 | Backup restore | Provision clean target, restore metadata/artifacts/keys, verify sampled result reproduction. | IncidentRecord with timeline, evidence, actions, validation and closure |
| RB-16 | Security incident | Isolate network, preserve audit/forensic artifacts, rotate credentials, validate trusted release. | IncidentRecord with timeline, evidence, actions, validation and closure |
Los runbooks deben probarse mediante ejercicios. Una instrucción no ejecutada en el hardware real es documentación pendiente de validación.