Simple Microservice Configuration
Well-written microservices are small and single-purpose; any non-trivial ecosystem will have a fleet of such services, each performing a different function. Inevitably, these services will use common code and structure; this library provides a simple mechanism for constructing these shared components and wiring them together into services.
microserviceis a small software application. It is composed of several smaller pieces of software, many of which are reusable.
componentis one of these (possibly reusable) pieces of software.
factoryis a function used to create a component; it may be an object's constructor.
config dictis a nested dictionary with string-valued keys. It contains data used by factories to create components.
object graphis a collection of components that may reference each other (acyclically).
bindingis a string-valued key. It is used to identify a component within an object graph and the subsection of the config dict reserved for a component's factory.
Define factory functions for
components, attach them to a
binding, and provide (optional) configuration
from microcosm.api import defaults, binding @binding("foo") @defaults(baz="value") def create_foo(graph): return dict( # factories can reference other components bar=graph.bar, # factories can reference configuration baz=graph.config.foo.baz, ) @binding("bar") def create_bar(graph): return dict()
Factory functions have access to the
object graphand, through it, the
config dict. Default configuration values, if provided, are pre-populated within the provided binding; these may be overridden from data loaded from an external source.
Wire together the microservice by creating a new object graph along with service metadata:
from microcosm.api import create_object_graph graph = create_object_graph( name="myservice", debug=False, testing=False, )
Factories may access the service metadata via
graph.metadata. This allows for several best practices:
- Components can implement ecosystem-wide conventions (e.g. for logging or persistence), using the service name as a discriminator.
- Components can customize their behavior during development (
debug=True) and unit testing (
object graphto access the corresponding
Components are initialized lazily. In this example, the first time
graph.foois accessed, the bound factory (
create_foo()) is automatically invoked. Since this factory in turn accesses
graph.bar, the next factory in the chain (
create_bar()) would also be called if it had not been called yet.
Graph cycles are not allowed, although dependent components may cache the graph instance to access depending components after initialization completes.
Optionally, initialize the microservice's components explicitly:
graph.use( "foo", "bar", )
While the same effect could be achieved by accessing
graph.bar, this construction has the advantage of initializes the listed components up front and triggering any configuration errors as early as possible.
It is also possible to then disable any subsequent lazy initialization, preventing any unintended initialization during subsequent operations:
This library was influenced by the pinject project, but makes a few assumption that allow for a great deal of simplication:
Microservices are small enough that simple string bindings suffice. Or, put another way, conflicts between identically bound components are a non-concern and there is no need for explicit scopes.
Microservices use processes, not threads to scale. As such, thread synchronization is a non-goal.
Mocking (and patching) of the object graph is important and needs to be easy. Unit tests expect to use `unittest.mock library; it should be trivial to temporarily replace a component.
Some components will be functions that modify other components rather than objects that need to be instantiated.