Files
bellsystems-cp/docs/melody-binary-serving.md
bonamin 1ed5012a95 fix: key melody binary storage on pid instead of uid, add migration script
melody.uid is never actually populated anywhere in MelodyForm.jsx, so keying
local .bsm storage on it (as the previous commit did) would silently break
for every existing melody. pid is the correct key anyway: it identifies the
underlying archetype binary, and multiple melodies legitimately share one
pid (each remaps the same note sequence to different bells/speed/duration
via its own settings). Deletion is now share-aware — a melody's binary is
only removed from disk once no other melody still references its pid.

Also adds backend/scripts/migrate_melody_binaries_to_local.py to backfill
existing melodies from their old Firebase URLs to local storage, with
--dry-run support and a warning list for pids whose melodies point at
different source files (a pre-existing data issue, flagged for manual
review via each melody's playback button rather than silently resolved).

Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
2026-07-06 10:53:25 +03:00

120 lines
6.6 KiB
Markdown

# Melody Binary Serving (Plain HTTP)
## Why this exists
ESP32 devices download `.bsm` melody files directly, using a URL supplied through the
Android app. The download previously pointed at a Firebase Storage public URL (HTTPS
only). ESP32's TLS client needs 40KB+ of RAM to open an HTTPS connection, which these
devices can't reliably spare. To fix this, melody `.bsm` binaries are now stored and
served by our own backend over plain HTTP, instead of Firebase Storage.
This does **not** affect the audio preview file (`information.previewURL`) — that's
only ever fetched by the browser admin UI, which is already HTTPS, so it stays on
Firebase Storage unchanged.
## PID vs UID — why storage is keyed on pid, not uid
- **`pid`** ("Playback ID") identifies the underlying **archetype binary** — the raw
note sequence (A, B, C...) baked into a `.bsm` file. Multiple melodies can
legitimately share one `pid`: each melody remaps those notes to actual bells via its
own `noteAssignments`, speed, and duration settings. One binary can back many
different "remix" melodies. **Duplicate PIDs across melodies are correct and
expected, not a bug.**
- **`uid`** is meant to be each melody's own unique database identity. It is *not*
currently populated anywhere in `MelodyForm.jsx` — this is a pre-existing, separate
bug, deferred as its own task. Because of this, local binary storage is keyed on
`pid`, not `uid`.
## What changed
- **Storage**: `.bsm` binaries are written to `./data/melody_binaries/{pid}.bsm` on
the host (mounted into the backend container at `/app/storage/melody_binaries`,
same pattern as `./data/firmware` and `./data/built_melodies`). Because storage is
keyed on `pid`, multiple melodies sharing a `pid` share one file — saving from any
of them overwrites the same file, which is expected.
- **Upload path**: `SelectArchetypeModal` / `BuildOnTheFlyModal` still call
`POST /api/melodies/{melodyId}/upload/binary` exactly as before. The backend
(`backend/melodies/service.py::save_binary_for_melody`) now writes the bytes to
local disk instead of uploading to Firebase Storage, and returns a URL like
`http://melodies.bellsystems.net/download/{pid}`. The melody must have a `pid` set
before uploading a binary (the endpoint returns 400 if not).
- **Stored URL**: the melody's `url` field (Firestore + SQLite) now holds that
plain-HTTP URL instead of a Firebase public URL. No schema change — `url` was
already a plain string field.
- **Download route**: `GET /api/melodies/download/{pid}` (`backend/melodies/router.py`)
is unauthenticated (devices have no login token) and resolves `pid` directly to
`{pid}.bsm` on disk, same pattern as the existing firmware download route
(`backend/firmware/router.py::download_firmware`).
- **Deletion is share-aware**: deleting a melody or its binary
(`service.delete_file` / `service._delete_storage_files`) only removes the local
`.bsm` file if no *other* melody still references the same `pid`
(`service._pid_used_by_other_melody`). This prevents one melody's deletion from
breaking playback for other melodies sharing its archetype binary.
## Migrating existing melodies
Melodies created before this change still have a Firebase Storage URL in their `url`
field — deploying this code does not touch existing data. They keep working exactly
as before (still HTTPS to Firebase) until migrated.
Two ways to move a melody to the new plain-HTTP URL:
1. **Per-melody, via the UI**: open the melody and re-run "Select Archetype" or
"Build on the Fly" — this naturally re-uploads through the same endpoint, which now
writes to local disk and updates `url`.
2. **Bulk, via script**: `backend/scripts/migrate_melody_binaries_to_local.py`
downloads each melody's current Firebase binary and writes it to local disk under
its `pid`, then updates `url` (SQLite + Firestore if published). Run with
`--dry-run` first:
```
docker exec -it bellsystems-backend python scripts/migrate_melody_binaries_to_local.py --dry-run
docker exec -it bellsystems-backend python scripts/migrate_melody_binaries_to_local.py
```
**Known caveat**: some existing melodies share a `pid` despite their Firebase URLs
pointing at *different* source files (observed on real data — e.g. voice-count
variant filenames like `1N_`/`2N_`/`3N_`/`4N_` all filed under one `pid`). The
script downloads only the first melody's file for each `pid` and reuses it for
every other melody sharing that `pid` — it does not attempt to detect which file is
"correct". It prints a warning list of every `pid` where this happened; use each
melody's playback button afterward to verify the right archetype plays, and
manually re-assign the correct one via "Select Archetype" if it's wrong.
## Infrastructure (outside this repo)
The plain-HTTP requirement is handled by keeping `melodies.bellsystems.net` on a
**separate** hostname from `console.bellsystems.net`, so the console's NPM proxy host
can stay HTTPS-only with no per-path exceptions.
Required, one-time setup outside this repository:
1. **DNS**: add an A/CNAME record for `melodies.bellsystems.net` pointing at the same
host as `console.bellsystems.net`.
2. **NPM (Nginx Proxy Manager) proxy host**: create a new proxy host for
`melodies.bellsystems.net` forwarding to the `nginx` container's exposed port
(`90` per `docker-compose.yml`). **Do not force SSL / do not enable the "Force
SSL" redirect** on this proxy host — it must remain reachable over plain HTTP.
The in-repo `nginx/nginx.conf` already has a dedicated `server_name
melodies.bellsystems.net` block that maps `/download/{pid}` to the backend's
`/api/melodies/download/{pid}` route, so no further nginx changes are needed once
the NPM proxy host exists.
## Verification
1. Build/select an archetype for a melody with a `pid` set → confirm a file appears
at `./data/melody_binaries/{pid}.bsm` and the melody's `url` becomes
`http://melodies.bellsystems.net/download/{pid}`.
2. `curl http://localhost:8000/api/melodies/download/{pid}` (direct to backend,
bypassing nginx) → returns the `.bsm` bytes, no auth header needed.
3. `curl http://localhost:90/api/melodies/download/{pid}` (through nginx on the
console server block) → same result.
4. Once NPM is configured, `curl http://melodies.bellsystems.net/download/{pid}`
→ same result, over plain HTTP.
5. Delete a melody whose `pid` is *not* shared with any other melody → confirm the
local `.bsm` file is removed. Delete one of two melodies sharing a `pid` → confirm
the file survives until the last one is deleted.
6. Upload a preview audio file → confirm it still lands in Firebase Storage and
`previewURL` still populates (regression check).