Files
rehearshalhub/ARCHITECTURE.md
Steffen Schuhmann 5e169342db docs: add ARCHITECTURE.md for agent/developer onboarding
Covers: service topology, directory layout, data model, full API surface,
scan/import pipeline, audio analysis flow, auth model, and key conventions.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
2026-03-29 16:08:23 +02:00

250 lines
12 KiB
Markdown

# RehearsalHub — Architecture
POC for a band rehearsal recording manager. Audio files live in Nextcloud; this app indexes, annotates, and plays them back.
---
## Services (Docker Compose)
```
┌─────────────┐ HTTP/80 ┌─────────────┐ REST /api/v1 ┌───────────────┐
│ Browser │ ──────────► │ web │ ──────────────► │ api │
└─────────────┘ │ (nginx + │ │ (FastAPI / │
│ React PWA) │ │ uvicorn) │
└─────────────┘ └──────┬────────┘
┌───────────────────────────────────────────┤
│ │ │ │
┌────▼────┐ ┌──────▼──────┐ ┌────▼────┐ ┌──▼──────────┐
│ db │ │ redis │ │Nextcloud│ │audio-worker │
│(Postgres│ │ (job queue │ │(WebDAV) │ │ (Essentia │
│ 16) │ │ + pub/sub) │ │ │ │ analysis) │
└─────────┘ └─────────────┘ └────┬────┘ └─────────────┘
┌─────▼──────┐
│ nc-watcher │
│(polls NC │
│ activity) │
└────────────┘
```
| Service | Image | Role |
|---|---|---|
| `web` | `rehearsalhub/web` | React 18 PWA (Vite + React Router + TanStack Query), served by nginx |
| `api` | `rehearsalhub/api` | FastAPI async REST API + SSE endpoints |
| `audio-worker` | `rehearsalhub/audio-worker` | Background job processor: downloads audio from NC, runs Essentia analysis, writes results to DB |
| `nc-watcher` | `rehearsalhub/nc-watcher` | Polls Nextcloud Activity API every 30s, pushes new audio uploads to `api` internal endpoint |
| `db` | `postgres:16-alpine` | Primary datastore |
| `redis` | `redis:7-alpine` | Job queue (audio analysis jobs) |
All services communicate on the `rh_net` bridge network. Only `web:80` is exposed to the host.
---
## Directory Layout
```
rehearsalhub-poc/
├── api/ # FastAPI backend
│ ├── alembic/ # DB migrations (Alembic)
│ └── src/rehearsalhub/
│ ├── db/
│ │ ├── models.py # SQLAlchemy ORM models
│ │ └── engine.py # Async engine + session factory
│ ├── repositories/ # DB access layer (one file per model)
│ ├── routers/ # FastAPI route handlers
│ ├── schemas/ # Pydantic request/response models
│ ├── services/ # Business logic
│ │ ├── nc_scan.py # Core scan logic (recursive, yields SSE events)
│ │ ├── song.py
│ │ ├── session.py # Date parsing helpers
│ │ └── band.py
│ ├── storage/
│ │ └── nextcloud.py # WebDAV client (PROPFIND / download)
│ └── queue/
│ └── redis_queue.py # Enqueue audio analysis jobs
├── worker/ # Audio analysis worker
│ └── src/worker/
│ ├── main.py # Redis job consumer loop
│ ├── pipeline/ # Download → analyse → persist pipeline
│ └── analyzers/ # Essentia-based BPM / key / waveform analysers
├── watcher/ # Nextcloud file watcher
│ └── src/watcher/
│ ├── event_loop.py # Poll NC activity, filter audio uploads
│ └── nc_client.py # NC Activity API + etag fetch
├── web/ # React frontend
│ └── src/
│ ├── pages/ # Route-level components
│ ├── api/ # Typed fetch wrappers
│ └── hooks/ # useWaveform, etc.
├── docker-compose.yml
└── Makefile
```
---
## Data Model
```
Member ──< BandMember >── Band ──< RehearsalSession
│ │
└──< Song >────┘
└──< AudioVersion
└──< SongComment
└──< Annotation
└──< RangeAnalysis
└──< Reaction
└──< Job
```
**Key tables:**
| Table | Purpose |
|---|---|
| `members` | User accounts. Store per-user Nextcloud credentials (`nc_username`, `nc_url`, `nc_password`) |
| `bands` | A band. Has a `slug`, optional `nc_folder_path` (defaults to `bands/{slug}/`), and `genre_tags[]` |
| `band_members` | M2M: member ↔ band with `role` (admin / member) |
| `band_invites` | Time-limited invite tokens (72h) |
| `rehearsal_sessions` | One row per dated rehearsal. `date` parsed from a `YYMMDD` or `YYYYMMDD` folder segment in the NC path. Unique on `(band_id, date)` |
| `songs` | A recording / song. `nc_folder_path` is the canonical grouping key (all versions of one song live in this folder). `session_id` links to a rehearsal session if the path contained a date segment |
| `audio_versions` | One row per audio file. Identified by `nc_file_etag` (used for idempotent re-scans). Stores format, size, version number |
| `annotations` | Time-stamped text annotations on a version (like comments at a waveform position) |
| `range_analyses` | Essentia analysis results for a time range within a version (BPM, key, loudness, waveform) |
| `jobs` | Redis-backed job records tracking audio analysis pipeline state |
---
## API
Base path: `/api/v1`
### Auth
| Method | Path | Description |
|---|---|---|
| `POST` | `/auth/register` | Create account |
| `POST` | `/auth/login` | Returns JWT |
JWT is sent as `Authorization: Bearer <token>`. Endpoints that need to work without auth headers (WaveSurfer, SSE EventSource) also accept `?token=<jwt>`.
### Bands
| Method | Path | Description |
|---|---|---|
| `GET` | `/bands` | List bands for current member |
| `POST` | `/bands` | Create band (validates NC folder exists if path given) |
| `GET` | `/bands/{id}` | Band detail |
| `PATCH` | `/bands/{id}` | Update band (nc_folder_path, etc.) |
| `GET` | `/bands/{id}/members` | List members |
| `DELETE` | `/bands/{id}/members/{mid}` | Remove member |
| `POST` | `/bands/{id}/invites` | Generate invite link |
| `POST` | `/invites/{token}/accept` | Join band via invite |
### Sessions
| Method | Path | Description |
|---|---|---|
| `GET` | `/bands/{id}/sessions` | List rehearsal sessions with recording counts |
| `GET` | `/bands/{id}/sessions/{sid}` | Session detail with flat song list |
| `PATCH` | `/bands/{id}/sessions/{sid}` | Update label/notes (admin only) |
### Songs
| Method | Path | Description |
|---|---|---|
| `GET` | `/bands/{id}/songs` | All songs for band |
| `GET` | `/bands/{id}/songs/search` | Filter by `q`, `tags[]`, `key`, `bpm_min/max`, `session_id`, `unattributed` |
| `POST` | `/bands/{id}/songs` | Create song manually |
| `PATCH` | `/songs/{id}` | Update title, status, tags, key, BPM, notes |
### Scan
| Method | Path | Description |
|---|---|---|
| `GET` | `/bands/{id}/nc-scan/stream` | **SSE / ndjson stream** — scan NC folder incrementally; yields `progress`, `song`, `session`, `skipped`, `done` events |
| `POST` | `/bands/{id}/nc-scan` | Blocking scan (waits for completion, returns summary) |
### Versions & Playback
| Method | Path | Description |
|---|---|---|
| `GET` | `/songs/{id}/versions` | List audio versions |
| `GET` | `/versions/{id}/stream` | Proxy-stream the audio file from Nextcloud (accepts `?token=`) |
| `POST` | `/versions/{id}/annotate` | Add waveform annotation |
### Internal (watcher → api)
| Method | Path | Description |
|---|---|---|
| `POST` | `/internal/nc-upload` | Called by nc-watcher when a new audio file is detected. No auth — internal network only |
---
## Scan & Import Pipeline
### Manual scan (SSE)
```
Browser → GET /nc-scan/stream?token=
scan_band_folder() [nc_scan.py]
│ recursive PROPFIND via collect_audio_files()
│ depth ≤ 3
For each audio file:
1. PROPFIND for etag + size
2. Skip if etag already in audio_versions
3. Parse YYMMDD/YYYYMMDD from path → get_or_create RehearsalSession
4. Determine nc_folder_path:
- File directly in session folder → unique per-file folder (bands/slug/231015/stem/)
- File in subfolder → subfolder path (bands/slug/231015/groove/)
5. get_or_create Song
6. Register AudioVersion
7. Yield ndjson event → browser invalidates TanStack Query caches incrementally
```
### Watcher-driven import
```
Nextcloud → Activity API (polled every 30s by nc-watcher)
event_loop.poll_once()
filter: audio extension only
normalize path (strip WebDAV prefix)
filter: upload event type
POST /internal/nc-upload
band lookup: slug-based OR nc_folder_path prefix match
same folder/session/song logic as manual scan
enqueue audio analysis job → Redis
```
---
## Audio Analysis
When a new `AudioVersion` is created the API enqueues a `Job` to Redis. The `audio-worker` picks it up and runs:
1. Download file from Nextcloud to `/tmp/audio/`
2. Run Essentia analysers: BPM, key, loudness, waveform peak data
3. Write `RangeAnalysis` rows to DB
4. Update `Song.global_bpm` / `Song.global_key` if not yet set
5. Clean up temp file
---
## Auth & Nextcloud Credentials
- JWT signed with `SECRET_KEY` (HS256), `sub` = member UUID
- Per-member Nextcloud credentials stored on the `members` row (`nc_url`, `nc_username`, `nc_password`). The API creates a `NextcloudClient` scoped to the acting member for all WebDAV operations.
- The watcher uses a single shared NC account configured via env vars (`NEXTCLOUD_USER` / `NEXTCLOUD_PASS`).
---
## Key Conventions
- **Repository pattern**: one `*Repository` class per model in `repositories/`. All DB access goes through repos; routers never touch the session directly except for passing it to repos/services.
- **Pydantic v2**: `model_validate(obj).model_copy(update={...})``model_validate` does not accept an `update` kwarg.
- **Async SQLAlchemy**: sessions are opened per-request via `get_session()` FastAPI dependency. SSE endpoints create their own session via `get_session_factory()()` because the dependency session closes when the handler returns.
- **Idempotent scans**: deduplication is by `nc_file_etag`. Re-scanning is always safe.
- **nc_folder_path grouping**: files in the same subfolder (e.g. `bands/slug/groove/`) are treated as multiple versions of one song. Files directly in a dated session folder get a unique virtual folder (`bands/slug/231015/stem/`) so each becomes its own song.
- **Migrations**: Alembic in `api/alembic/`. After rebuilding the DB run `docker compose exec api uv run alembic upgrade head`.