rapsqlite

True async SQLite — no fake async, no GIL stalls.

Overview

rapsqlite provides true async SQLite for Python, backed by Rust, Tokio, and sqlx. Database operations run outside the Python GIL, so the event loop never stalls. Use it as a drop-in replacement for aiosqlite with better concurrency and no thread pools.

📚 Full Documentation · Quickstart · API Reference · ROADMAP

Why rap*?

Packages prefixed with rap stand for Real Async Python. Unlike many libraries that merely wrap blocking I/O in async syntax, rap* packages guarantee that all I/O work is executed outside the Python GIL using native runtimes (primarily Rust). This means event loops are never stalled by hidden thread pools, blocking syscalls, or cooperative yielding tricks. If a rap* API is async, it is structurally non-blocking by design, not by convention. The rap prefix is a contract: measurable concurrency, real parallelism, and verifiable async behavior under load.

See the rap-manifesto for philosophy and guarantees.

Top Features

• ⚡ True async — All SQLite I/O runs outside the Python GIL (Rust + Tokio + sqlx)
• 🚫 No fake async — Zero thread pools; event-loop-safe concurrency
• 🔄 aiosqlite-compatible — ~95% API parity, drop-in replacement
• 🏊 Connection pooling — Configurable size and timeouts
• 🚀 Prepared statement caching — Automatic (2–5x faster repeated queries)
• 🐍 SQLAlchemy 2.0+ — sqlite+rapsqlite dialect for async Core and ORM
• 📦 Alembic — Full support for async migrations (alembic init -t async)

See the documentation for the full feature list (transactions, cursors, row factories, backup, callbacks, type adapters, and more).

Requirements

• Python 3.10+ (including Python 3.13 and 3.14)
• Rust 1.70+ (for building from source)
• Python development headers (included with most Python installations)

Installation

pip install rapsqlite

To verify: run the installation example in the docs (it prints [[1]]). For building from source, see Installation.

Documentation

📖 rapsqlite.readthedocs.io – Quickstart, API reference, migration from aiosqlite, performance, and advanced usage. Code examples are tested and show real output.

Quick Start

import asyncio
from rapsqlite import connect


async def main():
    async with connect("example.db") as conn:
        await conn.execute("CREATE TABLE users (id INTEGER PRIMARY KEY, name TEXT)")
        await conn.execute("INSERT INTO users (name) VALUES ('Alice')")
        rows = await conn.fetch_all("SELECT * FROM users")
        print(rows)


asyncio.run(main())

Output: [[1, 'Alice']]

SQLAlchemy & Alembic (0.4.0)

Use the sqlite+rapsqlite dialect with SQLAlchemy 2.0+ for true async ORM and Core. Alembic migrations are fully supported with the async template (alembic init -t async).

import asyncio
from sqlalchemy import text
from sqlalchemy.ext.asyncio import create_async_engine


async def main():
    engine = create_async_engine("sqlite+rapsqlite:///app.db")
    async with engine.connect() as conn:
        result = await conn.execute(text("SELECT 1"))
        print(result.scalar())  # 1
    await engine.dispose()


asyncio.run(main())

Install with pip install rapsqlite[sqlalchemy] (or pip install rapsqlite sqlalchemy). For Alembic, use pip install rapsqlite[sqlalchemy] alembic. See the Compatibility Guide for AsyncSession, ORM, and step-by-step Alembic setup.

For more (transactions, cursors, row factories), see the Quickstart Guide and API Reference. Code examples in the docs are tested and show real output.

API Reference

Complete API documentation is at rapsqlite.readthedocs.io:

• Connection API
• Cursor API
• Exception Classes
• Row Factory

Backup Support

The Connection.backup() method supports backing up to both rapsqlite.Connection and Python's standard sqlite3.Connection targets. For sqlite3.Connection targets, the backup uses Python's sqlite3 backup API on the on-disk database file (file-backed databases only; :memory: and non-file URIs are not supported).

For more details, see the Backup documentation in the API reference.

Performance

This package passes the Fake Async Detector. For detailed performance benchmarks and optimization tips, see the Performance Guide.

Key advantages:

• True async: All operations execute outside the Python GIL
• Prepared statement caching: Automatic query optimization via sqlx (2-5x faster for repeated queries)
• Better throughput: Superior performance under concurrent load due to GIL independence
• Connection pooling: Efficient connection reuse with configurable pool size

For process-local cache lookups, 0.5 adds fetch_scalar()/fetch_blob(), an opt-in raw_fetch_scalar() path, reusable conn.prepare(...) operations, and opt-in session_affinity=True. These APIs are measured separately from the general row API; use the Phase 0.5 benchmark and do not treat unmatched raw SQLite or published Redis figures as a direct speed claim.

Migration from aiosqlite

rapsqlite is designed to be a drop-in replacement for aiosqlite. The simplest migration is a one-line change:

# Before
import aiosqlite

# After
import rapsqlite as aiosqlite

For most applications, this is all you need! All core aiosqlite APIs are supported, including:

• Connection and cursor APIs
• async with db.execute(...) pattern
• Async iteration on cursors (async for row in cursor)
• Parameterized queries (named and positional)
• Transactions and context managers
• Row factories (including rapsqlite.Row class)
• Connection properties (total_changes, in_transaction, text_factory)
• executescript() and load_extension() methods
• Exception types

Practical compatibility notes:

• total_changes / in_transaction: both aiosqlite and rapsqlite expose these as properties (same API):

  # aiosqlite and rapsqlite
  changes = db.total_changes
  in_tx = db.in_transaction
  

• iterdump(): rapsqlite supports both async iteration (aiosqlite-style) and await-to-list:

  # aiosqlite and rapsqlite (async iterator)
  lines = [line async for line in db.iterdump()]
  
  # rapsqlite
  lines = await db.iterdump()
  dump_sql = "\n".join(lines)
  

• backup() targets: rapsqlite supports backups to both rapsqlite.Connection and sqlite3.Connection targets. For sqlite3.Connection targets, only file-backed databases are supported (not :memory: or non-file URIs).

See the Migration Guide for a complete migration guide with:

• Step-by-step migration instructions
• Code examples for common patterns
• API differences and limitations
• Troubleshooting guide
• Performance considerations

Compatibility Analysis: See the Compatibility Guide for detailed analysis based on running the aiosqlite test suite. Overall compatibility: ~95% for core use cases (updated 2026-01-26). All high-priority compatibility features implemented including total_changes(), in_transaction(), executescript(), load_extension(), text_factory, Row class, and async iteration on cursors.

Roadmap

See docs/ROADMAP.md for full details.

• ✅ 0.1–0.3 – Async core, aiosqlite compatibility, callbacks, pooling, True Async DBAPI, SQLAlchemy/Alembic integration, and advanced SQLite features
• ✅ 0.4 – Post-v0.3.3 compatibility, security, CI, SQLAlchemy 2.1 support, and release stabilization (v0.4.0)
• ✅ 0.5 – Released as v0.5.0: measured low-latency execution, hot-path reductions, scalar/BLOB and prepared-query paths, opt-in session affinity, and an opt-in raw path
• 📋 0.6 – In progress: cache get/set with TTL shipped in v0.5.0; bulk operations and multiplexed concurrent reads remain
• 📋 0.7–0.9 – Pooling, observability, reliability, ecosystem tooling, and stabilization toward 1.0

Related Projects

• rap-manifesto - Philosophy and guarantees
• rap-bench - Fake Async Detector CLI
• rapfiles - True async filesystem I/O
• rapcsv - Streaming async CSV

Changelog

See CHANGELOG.md for detailed release notes and version history.

Limitations

• Async only – Not designed for synchronous use; use sqlite3 for sync code.
• Backup to sqlite3.Connection – Supported for file-backed databases only (not :memory: or non-file URIs). See Backup Support above.

Release history and full feature list: CHANGELOG.md.

Contributing

Contributions are welcome! Please see our contributing guidelines.

License

MIT