s3rds

Python drivers for s3rds — a truly serverless RDS: a Lambda URL as your database endpoint, SQLite on S3, ≈ $0 when idle.

Zero dependencies (standard library only). Each driver is a submodule named after the standard client it mirrors, so moving to a real database is a one-line import change and nothing else:

pip install s3rds                # s3rds.sqlite: SQLite dialect
pip install "s3rds[mysql]"       # + s3rds.pymysql: write MySQL, translated on the client (sqlglot)

from s3rds import pymysql        # today   -- later: `import pymysql` on a real RDS, rest unchanged

conn = pymysql.connect(
    host="xxxx.lambda-url.us-east-2.on.aws",
    user="admin", password="...", database="blog")

with conn.cursor() as cur:
    cur.execute("CREATE TABLE posts (id INT AUTO_INCREMENT PRIMARY KEY, title VARCHAR(200))")
    cur.execute("INSERT INTO posts (title) VALUES (%s)", ["hello"])
    print(cur.lastrowid)

with conn.cursor(pymysql.cursors.DictCursor) as cur:
    cur.execute("SELECT * FROM posts WHERE id = %s", [1])
    print(cur.fetchall())                         # [{'id': 1, 'title': 'hello'}]

conn.transaction([                                 # all-or-nothing, single S3 write
    ("UPDATE stock SET qty = qty - 1 WHERE id = %s", [7]),
    ("INSERT INTO orders (item) VALUES (%s)", [7]),
])

Semantics

• Transactions: each execute() is its own transaction and is durable in S3 when it returns (like autocommit=True). commit() is a no-op. rollback() cannot undo statements that already ran — it warns if anything was written since the last commit() (and stays silent otherwise, so pools and ORMs may call it freely). For all-or-nothing writes use transaction([...]), which runs several statements as one commit and one S3 write.
• Large results are paged by the server (4 MB per page) and reassembled inside execute(); rows is always complete. If the database changes while paging, the query is re-run rather than mixing versions.
• Writes never duplicate: the driver re-sends only when the request provably never reached the server (stale keep-alive connection). Timeouts raise an error with an explicit "result unknown".
• s3rds.pymysql raises pymysql-style exceptions with MySQL error codes (1062, 1146, 1054, 1064, …); unsupported MySQL features raise NotSupportedError instead of misbehaving silently.

Server

You need an s3rds server in your own AWS account — deploy it once with Terraform or CloudFormation from the main repository. Same as any driver needs its database.