duckdb-driver 0.16.0

SQLAlchemy driver for duckdb

duckdb_driver

This project is a fork of duckdb-engine.

Basic SQLAlchemy driver for DuckDB

• duckdb_driver
  • Installation
  • Usage
  • Usage in IPython/Jupyter
  • Configuration
  • How to register a pandas DataFrame
  • Things to keep in mind
    • Auto-incrementing ID columns
    • Pandas read_sql() chunksize
    • Unsigned integer support
  • Alembic Integration
  • Preloading extensions (experimental)
  • Registering Filesystems
  • The name

Installation

$ pip install duckdb-driver

Usage

Once you've installed this package, you should be able to just use it, as SQLAlchemy does a python path search

from sqlalchemy import Column, Integer, Sequence, String, create_engine
from sqlalchemy.ext.declarative import declarative_base
from sqlalchemy.orm.session import Session

Base = declarative_base()


class FakeModel(Base):  # type: ignore
    __tablename__ = "fake"

    id = Column(Integer, Sequence("fakemodel_id_sequence"), primary_key=True)
    name = Column(String)


eng = create_engine("duckdb:///:memory:")
Base.metadata.create_all(eng)
session = Session(bind=eng)

session.add(FakeModel(name="Frank"))
session.commit()

frank = session.query(FakeModel).one()

assert frank.name == "Frank"

Usage in IPython/Jupyter

With IPython-SQL and DuckDB-Driver you can query DuckDB natively in your notebook! Check out DuckDB's documentation or Alex Monahan's great demo of this on his blog.

Configuration

You can configure DuckDB by passing connect_args to the create_engine function

create_engine(
    'duckdb:///:memory:',
    connect_args={
        'read_only': False,
        'config': {
            'memory_limit': '500mb'
        }
    }
)

The supported configuration parameters are listed in the DuckDB docs

How to register a pandas DataFrame

conn = create_engine("duckdb:///:memory:").connect()

# with SQLAlchemy 1.3
conn.execute("register", ("dataframe_name", pd.DataFrame(...)))

# with SQLAlchemy 1.4+
conn.execute(text("register(:name, :df)"), {"name": "test_df", "df": df})

conn.execute("select * from dataframe_name")

Things to keep in mind

Duckdb's SQL parser is based on the PostgreSQL parser, but not all features in PostgreSQL are supported in duckdb. Because the duckdb_driver dialect is derived from the postgresql dialect, SQLAlchemy may try to use PostgreSQL-only features. Below are some caveats to look out for.

Auto-incrementing ID columns

When defining an Integer column as a primary key, SQLAlchemy uses the SERIAL datatype for PostgreSQL. Duckdb does not yet support this datatype because it's a non-standard PostgreSQL legacy type, so a workaround is to use the SQLAlchemy.Sequence() object to auto-increment the key. For more information on sequences, you can find the SQLAlchemy Sequence documentation here.

The following example demonstrates how to create an auto-incrementing ID column for a simple table:

>>> import sqlalchemy
>>> engine = sqlalchemy.create_engine('duckdb:////path/to/duck.db')
>>> metadata = sqlalchemy.MetaData(engine)
>>> user_id_seq = sqlalchemy.Sequence('user_id_seq')
>>> users_table = sqlalchemy.Table(
...     'users',
...     metadata,
...     sqlalchemy.Column(
...         'id',
...         sqlalchemy.Integer,
...         user_id_seq,
...         server_default=user_id_seq.next_value(),
...         primary_key=True,
...     ),
... )
>>> metadata.create_all(bind=engine)

Pandas read_sql() chunksize

NOTE: this is no longer an issue in versions >=0.5.0 of duckdb

The pandas.read_sql() method can read tables from duckdb_engine into DataFrames, but the sqlalchemy.engine.result.ResultProxy trips up when fetchmany() is called. Therefore, for now chunksize=None (default) is necessary when reading duckdb tables into DataFrames. For example:

>>> import pandas as pd
>>> import sqlalchemy
>>> engine = sqlalchemy.create_engine('duckdb:////path/to/duck.db')
>>> df = pd.read_sql('users', engine)                ### Works as expected
>>> df = pd.read_sql('users', engine, chunksize=25)  ### Throws an exception

Unsigned integer support

Unsigned integers are supported by DuckDB, and are available in duckdb_engine.datatypes.

Alembic Integration

SQLAlchemy's companion library alembic can optionally be used to manage database migrations.

This support can be enabling by adding an Alembic implementation class for the duckdb dialect.

from alembic.ddl.impl import DefaultImpl

class AlembicDuckDBImpl(DefaultImpl):
    """Alembic implementation for DuckDB."""

    __dialect__ = "duckdb"

After loading this class with your program, Alembic will no longer raise an error when generating or applying migrations.

Preloading extensions (experimental)

DuckDB 0.9.0+ includes builtin support for autoinstalling and autoloading of extensions, see the extension documentation for more information.

Until the DuckDB python client allows you to natively preload extensions, I've added experimental support via a connect_args parameter

from sqlalchemy import create_engine

create_engine(
    'duckdb:///:memory:',
    connect_args={
        'preload_extensions': ['https'],
        'config': {
            's3_region': 'ap-southeast-1'
        }
    }
)

Registering Filesystems

DuckDB allows registering filesystems from fsspec, see documentation for more information.

Support is provided under connect_args parameter

from sqlalchemy import create_engine
from fsspec import filesystem

create_engine(
    'duckdb:///:memory:',
    connect_args={
        'register_filesystems': [filesystem('gcs')],
    }
)

The name

Yes, i thought forking and looking to maintain this was a good moment to also rename it to duckdb-driver.