15.260 14.9.20 Modos de fallo y recuperación
| ID | Fallo | Dominio | Respuesta | Debe preservar |
|---|---|---|---|---|
| FAR-01 | OAK disconnect or PoE interruption | Capture | Reconnect with bounded backoff; mark gap; never fabricate frames. | raw evidence state transition history error provenance |
| FAR-02 | DepthAI pipeline crash | Capture | Collect device state/logs, restart pipeline, preserve committed media. | raw evidence state transition history error provenance |
| FAR-03 | Timestamp discontinuity | Synchronization | Start new clock epoch; invalidate cross-epoch derivatives unless reconciled. | raw evidence state transition history error provenance |
| FAR-04 | RGB present, depth absent | Sensor | Continue 2D profile; mark metric geometry unavailable. | raw evidence state transition history error provenance |
| FAR-05 | Corrupt media chunk | Storage | Quarantine object, retry source when possible, block dependent outputs. | raw evidence state transition history error provenance |
| FAR-06 | Disk full | Storage | Stop new capture cleanly; preserve committed attempts; enter Critical. | raw evidence state transition history error provenance |
| FAR-07 | SQLite/WAL inconsistency | Metadata | Stop writers, snapshot diagnostics, execute documented recovery; never delete WAL ad hoc. | raw evidence state transition history error provenance |
| FAR-08 | Worker OOM | Compute | Restart worker, lower profile/batch, quarantine repeated input, preserve job lease semantics. | raw evidence state transition history error provenance |
| FAR-09 | Thermal throttling | Compute | Reduce concurrency/profile, extend latency expectations, notify operator. | raw evidence state transition history error provenance |
| FAR-10 | TensorRT engine incompatible | ModelRuntime | Reject engine; rebuild from trusted source model for exact runtime profile or use approved fallback. | raw evidence state transition history error provenance |
| FAR-11 | Model signature/digest invalid | SupplyChain | Quarantine and refuse activation. | raw evidence state transition history error provenance |
| FAR-12 | Calibration missing or stale | Geometry | Allow non-metric outputs only; block force/3D claims. | raw evidence state transition history error provenance |
| FAR-13 | Scene revision mismatch | Geometry | Request/construct correct revision; do not reuse stale hold IDs silently. | raw evidence state transition history error provenance |
| FAR-14 | Queue lease lost | Orchestration | Another worker may resume; output commit uses idempotency/digest check. | raw evidence state transition history error provenance |
| FAR-15 | Service crash loop | Operations | Exhaust restart budget then enter safe mode with incident bundle. | raw evidence state transition history error provenance |
| FAR-16 | WAN unavailable | Network | No effect on capture/analysis; sync remains pending. | raw evidence state transition history error provenance |
| FAR-17 | LAN client unavailable | Serving | Analysis continues; results remain local. | raw evidence state transition history error provenance |
| FAR-18 | Telemetry backend unavailable | Observability | Buffer within quota, sample/drop noncritical data, preserve critical domain events. | raw evidence state transition history error provenance |
| FAR-19 | Clock/timezone configuration changed | Operations | Use monotonic/device timestamps for durations; record wall-clock epoch change. | raw evidence state transition history error provenance |
| FAR-20 | Power loss | Platform | On reboot reconcile staging, leases, WAL, manifests and partial uploads. | raw evidence state transition history error provenance |
| FAR-21 | Backup target unavailable | Recovery | Keep local retention; alert; never delete source based on attempted backup. | raw evidence state transition history error provenance |
| FAR-22 | Consent withdrawn | Governance | Block export/sync, create tombstone, execute lineage-aware deletion workflow. | raw evidence state transition history error provenance |
| FAR-23 | Schema version unsupported | Compatibility | Reject or migrate explicitly; do not parse as nearest version. | raw evidence state transition history error provenance |
| FAR-24 | Unknown repeated input causes deterministic crash | Safety | Quarantine input, cap retries, generate reproducible diagnostic package. | raw evidence state transition history error provenance |
15.260.1 Recuperación tras reinicio
- Identificar release activo y verificar su receipt.
- Abrir metadata store mediante procedimiento normal y ejecutar checks configurados.
- Reconciliar staging y objetos finales.
- Expirar leases cuyo owner ya no existe.
- Reanudar jobs
Ready/RetryWaitingy comprobar checkpoints. - Reconstruir outbox de eventos no confirmados.
- Reanudar uploads desde receipts.
- Validar cámara y crear un nuevo clock epoch si reinició.
- Publicar un RecoveryRecord antes de aceptar captura.