SQLAlchemy 2 In Practice - Chapter 7: Asynchronous SQLAlchemy
a day ago
- SQLAlchemy 1.4+ supports asynchronous programming via asyncio for both Core and ORM, requiring async-compatible database drivers (e.g., aiosqlite, aiomysql, asyncpg).
- Asynchronous code must avoid implicit lazy loading; relationships should use eager loaders (like 'joined' or 'selectin') or explicit loaders (like 'write_only') to prevent function coloring issues.
- Sessions and engines are replaced with asynchronous equivalents (create_async_engine, async_sessionmaker), and 'expire_on_commit' is disabled to avoid implicit database refreshes.
- Alembic migrations require an async template ('-t async') and configuration of batch mode for compatibility.
- After a session flush, uninitialized list-style relationships can trigger lazy loading; solutions include initializing relationships explicitly or using the WriteOnlyMapped loader.
- Query execution methods (session.execute, session.scalar, etc.) must be awaited, and stream()/stream_scalars() are recommended for efficient async iteration over large result sets.