Feature: Seed data for DbContainer

What are you trying to do?

In order to do meaningful tests using my fresh database, I wish to load arbitrary SQL scripts from my DbContainer-s.

I can do (and have done) this per-project, in test folders, but believe there's value in allowing an arbitrary list of SQL scripts to be run before the testcontainer is yielded, and making this a pytest fixture to reuse.

I do this usually by first passing a path to a folder with scripts, volume-mount it to /seeds path in the database container, and also pass a list of script files in that folder to be executed. Then before yielding the ready db-container, I loop over each script filename and do container.exec_run(["mysql", "-e" f"source /seeds/{script}").

The list of scripts means we can feed both schema first, then arbitrary sample data files.

I suggest a new DbContainer._seed() method, defaulting to raise NotImplementedError, to be overriden per database, allowing a new optional parameter seed.

Each database then just has to define their _seed() function to the specific way to exec_run the scripts (psql for Postgres, mysql for MySQL CLI...)

Example from #542 implementation draft:

>>> import sqlalchemy
>>> from testcontainers.mysql import MySqlContainer
>>> seed_data = ("db/", ["schema.sql", "data.sql"])
>>> with MySqlContainer(seed=seed_data) as mysql:
...     engine = sqlalchemy.create_engine(mysql.get_connection_url())
...     with engine.begin() as connection:
...         query = "select * from stuff"  # Can now rely on schema/data
...         result = connection.execute(sqlalchemy.text(query))
...         first_stuff, = result.fetchone()

Why should it be done this way?

I believe most users of testcontainers, beyond just testing connectivity with a database, want a simple-to-reproduce test environment with realistic datastructure and mock data. This is currently annoying to do locally, and can be upstreamed into a simple-looking seed parameter, which could enable the feature for most databases, by pushing it into DbContainer itself.

See #542 for a draft implementation using MySQL as target (generalizeable to others of course)

Start with the #542 draft implementation and the DbContainer and MySqlContainer entry points described here. Trace how the proposed seed tuple, database-specific _seed() method, and reusable pytest fixture would fit together. Done means the example can load schema.sql and data.sql before the container is yielded, with the approach generalizable beyond MySQL.