COPY and UNLOAD will work out of the box, when you’re testing code which accepts a sqlalchemy engine or session object, because we can preconfigure it to work properly. In these scenarios, you should simply be able to send in the fixture provided into your test and be on your merry way.

This should also work seamlessly if you’re testing code which creates its own connection directly. Consider the following module that creates a redshift engine and then uses said engine to run a COPY command:

# src/
from sqlalchemy import create_engine

def main(**connect_args):
    engine = get_redshift_engine(connect_args)

    return run_major_thing(engine)

def get_redshift_engine(config):
    return create_engine(**config)

def run_major_thing(engine):
        COPY x.y FROM 's3://bucket/file.csv' credentials....

The redshift fixture should automatically instrument direct calls to psycopg2.connect() (or sqlalchemy.create_engine().