Kink

Dependency injection container made for python

Features

Easy to use interface
Extensible with custom dependency resolvers
Automatic dependency injection (Autowiring)
Lightweight
Support for async with asyncio

Installation

Pip

pip install kink

Poetry

If you don't know poetry, I highly recommend visiting their webpage

poetry add kink

Why using dependency injection in python?

Short story

Because python is a multi paradigm language and this should encourage you to use best OOP practices improving your workflow and your code and have more time for your hobbies and families instead monkey-patching entire world.

Long story

Dependency happens when one component (component might be a class, or a function) A uses other component B. We say than that A depends on B.

Instead hardcoding dependency inside your components and making your code tightly coupled you are loosing it by providing(injecting) required behaviour either by subclassing or plugging additional code. This is called Inversion of Control which keeps your code oriented around behaviour rather than control. There are many benefits coming out of it:

increased modularity
better extensibility and flexibility
it helps you understand higher concepts like event driven programming

This is where dependency injection comes in place. Dependency injection is a specific style of inversion of control, which generally says instead hardcoding dependency pass dependant object as a parameter to a method rather than having method creating it itself. ( who would thought it is so easy :)? ). It can go even further than that; when you pass a dependency don't rely on a particular implementation rely on an abstraction (Dependency Inversion Principle).

So you might ask why do I need it? Here is couple reasons:

Relying on the global state is evil

Coding is hard enough ( business requirements are changing all the time, deadlines are shortening, clients wants more, there are so many unknowns you have to figure out), relying on unpredictable state makes it even harder:

it might introduce potential bugs
makes code harder to maintain
concurrency becomes harder to achieve
balancing mokey-patching well is a hard task

Great, but now I have additional work I have to manage now all my dependencies write more code and deadlines are coming even closer!

True, that is why you should pick up Dependency Injection Container to do all this work for you. Kink gives you one decorator and simple dict-like object to bootstrap and manipulate your container. No need for manual work and manual dependency management. Give it a try and you will love it!

Usage

To fully utilise the potential of kink it is recommended to bootstrap your initial dependencies (config values, or instances of classes that are standalone, requires no other dependencies than themselves). Some people prefer to keep it in __init__.py in the top module of your application, other create separate bootstrap.py file for this purpose. Once all is setup the only step left is to decorate your classes/functions with @inject decorator.

Bootstrapping/Adding services manually

Adding service to di container

Dependency container is a dict-like object, adding new service to dependency container is as simple as the following example:

from kink import di
from os import getenv

di["db_name"] = getenv("DB_NAME")
di["db_password"] = getenv("DB_PASSWORD")

Adding on-demand service to dependency injection container

Kink also supports on-demand service creation. In order to define such a service, lambda function should be used:

from kink import di
from sqlite3 import connect

di["db_connection"] = lambda di: connect(di["db_name"])

In this scenario connection to database will not be established until service is requested.

Adding factorised services to dependency injection

Factorised services are services that are instantiated every time they are requested.

from kink import di
from sqlite3 import connect

di.factories["db_connection"] = lambda di: connect(di["db_name"])

connection_1 = di["db_connection"]
connection_2 = di["db_connection"]

connection_1 != connection_2

In the above example we defined factorised service db_connection, and below by accessing the service from di we created two separate connection to database.

Requesting services from dependency injection container

To access given service just reference it inside di like you would do this with a normal dictionary, full example below:

from kink import di
from sqlite3 import connect

# Bootstrapping
di["db_name"] = "test_db.db"
di["db_connection"] = lambda di: connect(di["db_name"])


# Getting a service
connection = di["db_connection"] # will return instance of sqlite3.Connection
assert connection == di["db_connection"] # True

Autowiring dependencies

Autowiring is the ability of the container to automatically create and inject dependencies. It detects dependencies of the component tries to search for references in the container and if all references are present an instance of requested service is returned.

Autowiring system in kink works in two ways:

matching argument's names
matching argument's type annotation

How dependencies are prioritised by autowiring mechanism

Autowiring mechanism priorities dependencies automatically, so when multiple matches are found for the service this is how it works; Firstly passed arguments are prioritied - if you pass arguments manually to the service they will take precendence over anything else. Next argument's names are taken into consideration and last but not least argument's type annotations.

Matching argument's names

If you don't like type annotations or would like to take advantage of autowiring's precedence mechanism use this style.

This is a very simple mechanism we have already seen in previous examples. Autowiring system checks function argument's names and tries to search for services with the same names inside the container.

Matching argument's type annotations

If you are like me and like type annotations and use static analysis tools this is a preferred way working with DI container.

In this scenario names are ignored instead argument's type annotations are inspected and looked up inside di container. This requires aliases when bootstrapping your services in DI container or simply adding them to container in the way that its type is the key by which service is accessed. Please consider the following example:

from kink import di, inject
from sqlite3 import connect, Connection


di["db_name"] = "test_db.db"
di[Connection] = lambda di: connect(di["db_name"])  # sqlite connection can be accessed by its type

@inject # Constructor injection will happen here
class UserRepository:
  def __init__(self, db: Connection): # `db` argument will be resolved because `Connection` instance is present in the container. 
    self.db = db

repo = di[UserRepository]
assert repo.db == di[Connection] # True

Constructor injection

from kink import inject, di
import MySQLdb

# Set dependencies
di["db_host"] = "localhost"
di["db_name"] = "test"
di["db_user"] = "user"
di["db_password"] = "password"
di["db_connection"] = lambda di: MySQLdb.connect(host=di["db_host"], user=di["db_user"], passwd=di["db_password"], db=di["db_name"])

@inject
class AbstractRepository:
    def __init__(self, db_connection):
        self.connection = db_connection


class UserRepository(AbstractRepository):
    ...


repository = di[UserRepository] # will retrieve instance of UserRepository from di container
repository.connection # mysql db connection is resolved and available to use.

When class is annotated by inject annotation it will be automatically added to the container for future use (eg autowiring).

Services aliasing

When you register a service with @inject decorator you can attach your own alias name, please consider the following example:

from kink import inject
from typing import Protocol

class IUserRepository(Protocol):
    ...

@inject(alias=IUserRepository)
class UserRepository:
    ...


assert di[IUserRepository] == di[UserRepository] # returns true

For more examples check tests directory

Retrieving all instances with the same alias

Aliases in kink do not have to be unique, but by default when autowiring mechnism is called the service that was registered first within given alias will be returned. If for some reason you would like to retrieve all services that alias to the same name (eg implementing strategy pattern), kink provides a useful functionality for doing so. Please consider the following example:

from kink import inject
from typing import Protocol, List

class IUserRepository(Protocol):
    ...

@inject(alias=IUserRepository)
class MongoUserRepository:
    ...

@inject(alias=IUserRepository)
class MySQLUserRepository:
    ...

@inject()
class UserRepository:
    def __init__(self, repos: List[IUserRepository]) -> None: # all services that alias to IUserRepository will be passed here
        self._repos = repos
        
    def store_to_mysql(self, user: ...):
        self._repos[1].store(user)
    
    def store_to_mongo(self, user: ...):
        self._repos[0].store(user)

Clearing di cache

Sometimes it might come handy to clear cached services in di container. Simple way of doing this is calling di.clear_cache() method like in the following example.

from kink import inject, di

... # set and accesss your services

di.clear_cache() # this will clear cache of all services inside di container that are not factorised services

Integration with FastAPI

from fastapi import APIRouter, Depends, status
from fastapi.responses import JSONResponse, Response
from kink import di

router = APIRouter()

# register service in the DI container
di[ClientService] = ClientService()

@router.post(
    "/clients",
    response_model=ClientDTO,
    responses={400: {"model": APIErrorMessage}, 500: {"model": APIErrorMessage}},
    tags=["clients"],
)
async def create_client(
    request: CreateClientDTO, service: ClientService = Depends(lambda: di[ClientService])
) -> JSONResponse:
    result = service.create(request)
    return JSONResponse(content=result.dict(), status_code=status.HTTP_201_CREATED)

A complete example, together with tests you can find it here.

Articles on Kink

https://www.netguru.com/codestories/dependency-injection-with-python-make-it-easy

stnley/kink