This repository is a real life example of Onion Architecture with use of Node.js / Express
and Typescript
Diagram available here
Diagrams have copyrights, if you want to use it on larger scale, feel free to contact me.
Project is a simple simulator of warehouse and managing storage space.
There are two separate perspectives Administrative
and Client
facing app.
Given that I'm not a Client
And I would like to create account
When I provide required data
Then my account is created
Given I'm not authenticated Client
And I would like to authenticate
When I provide proper authentication details
Then I'm able to get into system
Given I'm authenticated Client
And I would like to delete account
When I perform delete action
Then I'm no longer able to authenticate
Given I'm authenticated Client
When I get into my profile
Then I can see my account data
Given I'm authenticated Client
And I would like to create equipment
When I provide all required data
Then Equipment is created
Given I'm authenticated Client
And I have equipment
When I select Warehouse
Then I receive price preview
Given I'm authenticated Client
And I have multiple equipment items
Then I can preview it
Given I'm authenticated Client
Then I can preview available Warehouses
Given I'm authenticated Client
And I have equipment
When I select Warehouse
And perform store item action
Then I receive Warehouse Item with cost included
Given I'm authenticated Admin
Then I can preview all equipment
And Users related to equipment
Given I'm authenticated Admin
And I would like to create equipment
When I provide required data
Then Equipment is created
Given I'm authenticated Admin
Then I can preview all warehouses
And related Warehouse Items
Given I'm authenticated Admin
Then I can preview specific warehouse
And related Warehouse Items
Given I'm authenticated Admin
And I would like to create new Warehouse
When I provide all required data with State
Then Warehouse is created
Given I'm authenticated Admin
And there is existing Warehouse
When I provide new data with State
Then Warehouse data is updated
Given I'm authenticated Admin
And there is existing WarehouseItem
Then I can preview it
Given I'm authenticated Admin
And there are many existing WarehouseItems
Then I can preview all of them
Given I'm authenticated Admin
And I would like to create new WarehouseItem
When I provide all required data with Warehouse
Then WarehouseItem is created
Given I'm authenticated Admin
And there is existing WarehouseItem
When I provide new data with Warehouse
Then WarehouseItem is updated
Given I'm authenticated Admin
Then I can preview all Users
Given I'm authenticated Admin
Then I can preview all States
And related Rates
Given I'm authenticated Admin
Then I can preview all Rates
Typescript
(v3.7.5
)Inversify.js
TypeOrm
Express.js
Apollo Server
GraphQL
Mocha / Chai
for testing
-
core (
Application Core
)Contains application core related layers like application services, domain and domain services
-
dependency (
Dependency injection layer
)Contains definition for Container and whole project dependencies
-
infrastructure
Contains definition of data sources in case of this boilerplate - database
-
ui
Contains definition of presentation layer like controller, express setup etc
It's important to keep data flow as simple as possible. Generally it's simple to follow - for entry data always specify request object, for output translate data to specific layer. For easier understanding I've prepared a diagram.
Diagram available here
Every layer has its own rules when it comes to access to another layer.
Dependency injection
has access to every layer to provide proper implementations.UI
have access only toCore Layer
Domains
can be grouped intoBounded Context
- Given
Domains
do not see each other, but definesprotocol
of communication likeCLI
,REST
etc.
Infrastructure
have access only toCore Layer
Functionalities
can be grouped intoFunctionality Group
likeMessaging
,Persistence
etc- Given
Functionalities
do not see each other, but have access toData Source
Core Layer
don`t have access to any layer. It means that it's fully independent of implementations and can be extracted from project if neededApplication Service
here have access toDomain Services
andDomain Models
Domain Serivce
have access only to domain models, andDomain Services
don`t see one each otherDomain Model
don`t have access to any upper layer
Visual representation of above restrictions can be seen in a diagram.
Diagram available here
To understand in details how whole architecture works together with applied design patterns etc take a look at detailed diagrams per layer.
Diagram available here
Diagram available here
Diagram available here
It's natural that every project evolve with time. From my perspective process of growth can be divided into specific phases.
- Initial Phase
- Introducing
Onion Architecture
to project bylayering
codebase - separating code intoUI
,CORE
,INFRASTRUCTURE
layers
- Introducing
- Bounded Context Phase
- Grouping domains into
Bounded Contexts
- Grouping domains into
- Bounded Context Growth Phase
- Separating whole project into
Multiple Bounded Contexts
- Separating whole project into
- Modularization Phase
- Dividing Project into
Modules
. Usually you will notice that you can rollback your code toInitial Phase
as modules are separation was based onBounded Contexts
- Dividing Project into
- Modularization Growth Phase
- Every
Module
evolve in its own pace up toBounded Context Growth Phase
. You may have alsoCommon Module
which is shared across other modules.
- Every
- Microservices Phase
- After some time you may notice that having multiple modules within one codebase is too much, or you may se that specific module is a potential candidate to
Microservice
. If you decide to detachModule
intoMicroservice
you may start application lifecycle from the beginning or from any phase you like up toModularization Growth Phase
. You may even start from initial phase and move through all the phases within specificMicroservice
. EveryMicroservice Architecture lifecycle
is independent ( like inModularization Growth Phase
)
- After some time you may notice that having multiple modules within one codebase is too much, or you may se that specific module is a potential candidate to
Visual representation of above process can be seen in a diagram.
Diagram available here
To have better context on general example of architecture, it's also good to understand database structure and how whole project structure is aligned to it.
- Multiple environment setup
- DB Agnostic setup, supports multiple datasource
- Infrastructure -> Domain Mapping -> UI Mapping
- Migrations, Fixtures, Seeds
- Multiple API versions support ( REST implementation )
- Global Error Handling
- Test Parallelization
- Multiple protocols within one codebase ( GraphQL / REST )
Inspired by following articles:
https://www.slideshare.net/matthidinger/onion-architecture
-
Yarn
-
NVM
( tested onv10.13.0+
)wget -qO- https://raw.githubusercontent.com/nvm-sh/nvm/v0.34.0/install.sh | bash`
-
PostgreSQL
( tested onv11+
)
- Database
- Look at the
ormconfig.sample.js
file. It's a sample setup of database connection, you can provide your own data for database if needed. From app perspective you have to manually create a database for development ( in sample with nameonion_dev
) and for testingonion_test
. - Migrations will autorun on application start
- Look at the
- Env Variables
-
.env.example
contains example env config - for local / dev use you can use same values as provided in sample -
for production use generate token with following command
node -e "console.log(require('crypto').randomBytes(256).toString('base64'));"
-
- Follow
SETUP
section first and installPREREQUISITIES
Yarn install
- installing dependenciesYarn dev
- run app with watch and rebuild
- To prepare a database with the latest migrations run
yarn db:reload
, it also removes all data from db and recreates it. Useful when playing with seed data. - To seed database run
yarn db:seed
- To generate migration based on changes in entity object run
yarn db:generate <my_migration_name>
When there is a swagger host provided in .env
file then you can navigate to http://localhost:3000/api-docs/
Update swagger.json
file located at ui > config
every time you apply changes to api.
Navigate to http://localhost:3000/graphql
to make queries through GraphQL
playground
- Prepare tests database first (
SETUP SECTION
) - Run
yarn test
- should run mocha tests in parallel
- Read guide here to setup global dependencies
- Run
yarn test:mutate
command
There are some universal concepts in programming ( designed patterns ) which are common for general engineering, but it's not always obvious how to use environment specific concepts. In this boilerplate I'm going to show how to handle that.
Request object defines parameters / input to specific module input ( domain / infrastructure ), and holds required data which cannot be changed on the fly.
Unit Of Work is simply speaking - a wrapper. It wraps repositories and performs required operations usually in transaction. It solves issue related to circular repository dependency in ioc, or nested repository dependency on each other. With unit of work specific repository methods don't have any reference to external repositories which makes them more atomic.
Interactor
is a single, independent action to execute in any place of the system. It contains logic related to specific problem ( usually also to specific domain )
and can be shared across multiple domains. It has clear input definition and output. The idea of interactor
is to avoid need of nesting
services
in each other. If you need to nest services
then you also combine domains, and you will face cross domain issues. As interactor
is
independent operation to reuse, it CANNOT have service
as dependency ( repository
is allowed ). It can be injected into service
though.
UseCase
you can see it as a wrapper, if you have business use case which depends on multiple interactor
results, and you need to apply some logic on those results,
then UseCase
is a way to go. UseCase
may have multiple interactors
as dependencies + repositories
.
Scenario
is a wrapper around UseCases
. Same way of thinking as in UseCase
- if you have a complex, reusable across multiple domains / modules operation, which depends
on multiple UseCases
, Interactors
, repositories
results, and you need to apply specific logic on those results, then scenario
is a way to go.
All of those concepts should be presented as single action to execute - there shouldn't be multiple methods / functions applied to their interfaces, just execute
.
The Simple concept, where one module data structure is translated to another module
This mapper is prepared for mapping data source format data into domain format. The Simplest example would be that, in
a database we store first_name
and last_name
in separate columns, but in a domain we need to have field name
which
is combined value of previously mentioned columns. In that case we define domain model with required fields and new name
field.
In a Mapper, we can perform merging of those 2 values. Thanks to that we can have separation
between definition of Entity and Domain, and also we have just plain values in domain object instance without
any overhead related to persistence data etc, which for sure would be stored by Entity object instance. We can also calculate
simple values in mappers etc.
This mapper is for preparing Domain data format into specific ui data format. Sometimes we may need to perform
some logic in domain services on domain object format, but we would like to make a response in totally different format.
For example, we may fetch data as array from a database, perform operations in services on an array but on UI
, we would like to
group array elements into map structure in a different format. In a repository, we mapped User
domain object into User
ui object
where UI
object do not contain password field and contains only required fields for authentication purposes.
Used for managing database changes. In a repository, we generate migrations based on entity changes. So we can
add a new column on entity and then just use one command to generate required migration. It's recommended to split database related
changes into multiple migrations instead creating one migration for all related feature changes. For example, it's better to
have separate migration for creating x
table and separate migration for adding / updating table columns definition to
table y
.
Used for local development or testing - it's just data for specific use cases which can be also used for dev environment where QA's can test specific endpoints or screens. It's also useful as start data for local development especially when you are working as a full stack.
Every test runs on it's separate database, and we can spawn multiple tests at the same time, and run in transaction specific test cases, thanks to which we don't have to clear db after running every test.
Mutational testing is integrated with Mocha
test runner and shows how many mutations are still available in
system, and where we should apply additional test coverage.
We are testing whole layers data flow - from UI
layer up to Infrastructure
. We are testing
not only responses but also saved data in database and authentication context.
- Prepare FP version of architecture - separate repo
- Introduce the
Docker
into project - Prepare testing infrastructure for
GraphQL
endpoints, add more tests, update testing and fix issues - Provide example of project modularization (
lerna
+yarn workspaces
) - Resolve
TODO's
comments
-
To authenticate provide token this way as swagger 2.0 do not support bearer strategy
https://github.com/OAI/OpenAPI-Specification/issues/583#issuecomment-267554000
-
http context is empty in controllers, looks like http context is incorrectly injected into controller, everything is fine though in middlewares - applied workaround take a look at
getCurrentUser
helper