# Scheduler Storage Mongo — Sprint 16 Handoff

This module now consumes the canonical DTOs defined in `StellaOps.Scheduler.Models`.  
Samples covering REST shapes live under `samples/api/scheduler/` and are referenced from `docs/11_DATA_SCHEMAS.md#3.1`.

## Collections & DTO mapping

| Collection        | DTO                      | Notes                                                                                 |
|-------------------|--------------------------|---------------------------------------------------------------------------------------|
| `schedules`       | `Schedule`               | Persist `Schedule` as-is. `_id` → `Schedule.Id`. Use compound indexes on `{tenantId, enabled}` and `{whenCron}` per doc. |
| `runs`            | `Run`                    | Store `Run.Stats` inside the document; omit `deltas` array when empty.                |
| `impact_snapshots`| `ImpactSet`              | Normalise selector filter fields exactly as emitted by the canonical serializer.      |
| `audit`           | `AuditRecord`            | Lower-case metadata keys are already enforced by the model.                           |

All timestamps are persisted as UTC (`+00:00`). Empty selector filters remain empty arrays (see `impact-set.json` sample).

## Implementation guidance

1. Add a project reference to `StellaOps.Scheduler.Models` and reuse the records directly; avoid duplicate BSON POCOs.
2. When serialising/deserialising to MongoDB, call `CanonicalJsonSerializer` to keep ordering stable for diffable fixtures.
3. Integration tests should load the JSON samples and round-trip through the Mongo persistence layer to guarantee parity.
4. Follow `docs/11_DATA_SCHEMAS.md` for index requirements; update that doc if storage diverges.
5. Register `AddSchedulerMongoStorage` in the host and call `ISchedulerMongoInitializer.EnsureMigrationsAsync` during bootstrap so collections/indexes are created before workers/web APIs start.

With these artefacts in place the dependency on SCHED-MODELS-16-101/102 is cleared—storage work can move to DOING.

## Repositories & Sessions (Sprint 16)

- `AddSchedulerMongoStorage` now registers the scheduler repositories:
- `IScheduleRepository` — CRUD helpers with tenant scoping and soft-delete markers (`deletedAt`, `deletedBy`).
- `IRunRepository` — create/update/list helpers sorted by `createdAt`, honouring the TTL index on `finishedAt`.
- `IImpactSnapshotRepository` — stores canonical `ImpactSet` snapshots with deterministic selector digests.
- `IAuditRepository` — append/list audit log entries with optional category/schedule/run filters.
- `IRunSummaryRepository` — persists the `run_summaries` materialised view keyed by tenant/schedule.
- `IRunSummaryService` — projects run updates into the materialised view (latest run + rolling counters for the last 20 updates).
- `ISchedulerAuditService` — convenience layer that stamps IDs/timestamps and writes `AuditRecord` instances with consistent metadata.
- `ISchedulerMongoSessionFactory` provides causal-consistent Mongo sessions (default `CausalConsistency = true`) that repositories accept for read-after-write flows.
- All repositories persist canonical JSON via `CanonicalJsonSerializer`; helper mappers strip internal fields before materialising DTOs.
- Soft-deleted schedules keep historical documents but are excluded from queries unless `ScheduleQueryOptions.IncludeDeleted` is set; deletions also disable the schedule and bump `updatedAt/updatedBy`.