Full FastAPI backend for the AI music continuation editor:
Services
- chain.py: chain + node CRUD, commit/discard, recursive CTE spine query
- musicgen.py: MusicGenClient with cf-orch allocation + mock mode (CF_MUSICGEN_MOCK=1)
- stems.py: Demucs 4-stem separation subprocess wrapper + mock mode
- export.py: ffmpeg concat demuxer to stitch committed spine into WAV/MP3
API endpoints
- chains: CRUD, multipart audio upload (WAV/MP3/FLAC/OGG/M4A/AIFF)
- nodes: branch creation (202 + BackgroundTasks), commit, discard, audio stream
- gpu: cf-orch capacity status; session allocation stubbed pending cf-orch#43
- stems: Paid-tier stem separation (Demucs, gated via tiers.py)
- export: POST /{chain_id}/export → FileResponse download
- events: SSE stream (node-status events) per chain via asyncio Queue pub/sub
Infrastructure
- lifespan: reads SPARROW_DB_PATH/DATA_DIR at startup (not import time)
- events_store: subscribe/unsubscribe/broadcast pattern for SSE
- CORS: open in dev, SPARROW_CORS_ORIGINS in production
- Background generation opens its own DB connection (WAL-safe)
Tests: 30/30 passing across service units and API integration
27 lines
522 B
Python
27 lines
522 B
Python
# app/models/schemas/chain.py — Chain Pydantic models
|
|
from __future__ import annotations
|
|
|
|
from pydantic import BaseModel
|
|
|
|
|
|
class ChainCreate(BaseModel):
|
|
name: str
|
|
|
|
|
|
class ChainRow(BaseModel):
|
|
id: str
|
|
name: str
|
|
created_at: float
|
|
node_count: int = 0
|
|
|
|
|
|
class ChainTree(BaseModel):
|
|
id: str
|
|
name: str
|
|
created_at: float
|
|
nodes: list["NodeRow"] = []
|
|
|
|
|
|
# Avoid circular import — NodeRow imported at runtime
|
|
from app.models.schemas.node import NodeRow # noqa: E402
|
|
ChainTree.model_rebuild()
|