Skip to content


Switch branches/tags

Latest commit


Git stats


Failed to load latest commit information.
Latest commit message
Commit time


Opinionated data loading with SQLite.

While most distributed application runtimes will use a networked data store for mutable state, the usage patterns of data that is read-only at runtime are great fit for SQLite.

In particular, microcosm-sqlite assumes that applications will

  • Build data sets in advance and ship them as static artifacts (e.g. in source control)
  • Load data immutable sets at runtime without loading entire data sets into memory

Writing Models

Persistent data is expected to use SQLAlchemy's declarative base classes. Because different data sets may be shipped in different SQLite databases, each declarative base class needs to have a unique name and a separate engine configuration, which is achieved by adding DataSet as the base of the declarative base class:

Base = DataSet.create("some_name")

class SomeModel(Base):
    __tablename__ = "sometable"

    id = Column(Integer, primary_key=True)

Using Stores

Basic persistence operations are abstracted through a store:

class SomeStore(Store):

    def model_class(self):
        return SomeModel

 store = SomeStore()
 results =

Configuring SQLite

Each DataSet defaults to using :memory: storage, but can be customized in two ways:

  1. The SQLiteBindFactory can be configured with custom paths:

    loader = load_from_dict(
                "some_name": "/path/to/database",
    graph = create_object_graph("example", loader=loader)
  2. The microcosm.sqlite entrypoint can contain a mapping from a data set name to a function that returns a path.