Food Delivery Microservices
is a fictional food delivery microservices, built with .Net Core and different software architecture and technologies like Microservices Architecture, Vertical Slice Architecture , CQRS Pattern, Domain Driven Design (DDD), Event Driven Architecture. For communication between independent services, we use asynchronous messaging with using rabbitmq on top of MassTransit library, and sometimes we use synchronous communication for real-time communications with using REST and gRPC calls.
π‘ This application is not business oriented and my focus is mostly on technical part, I just want to implement a sample with using different technologies, software architecture design, principles and all the thing we need for creating a microservices app.
Warning This project is in progress. I add new features over the time. You can check the Release Notes.
π― This Application ported to modular monolith
approach in food-delivery-modular-monolith repository, we can choose best fit architecture for our projects based on production needs.
Other versions of this project are available in these repositories, We can choose best fit architecture for our projects based on production needs:
- https://github.com/mehdihadeli/food-delivery-modular-monolith
- https://github.com/mehdihadeli/go-food-delivery-microservices
For your simplest .net core projects, you can use my vertical-slice-api-template
project template:
If you like feel free to β this repository, It helps out :)
Thanks a bunch for supporting me!
- Features
- Plan
- Setup
- Technologies - Libraries
- The Domain and Bounded Context - Service Boundary
- Application Architecture
- Application Structure
- Vertical Slice Flow
- Prerequisites
- How to Run
- Contribution
- Project References
- License
- β
Using
Microservices
andVertical Slice Architecture
as a high level architecture - β
Using
Event Driven Architecture
on top of RabbitMQ Message Broker and MassTransit library - β
Using
Domain Driven Design
in most of services like Customers, Catalogs, ... - β
Using
Event Sourcing
andEventStoreDB
inAudit Based
services like Orders, Payment - β
Using
Data Centeric Architecture
based onCRUD
in Identity Service - β
Using
CQRS Pattern
on top ofMediatR
library and splitingread models
andwrite models
- β
Uing
Structured logging
with serilog and exporting logs toElastic Seacrch
andKibana
through serilog-sinks-elasticsearch sink - β
Using
Outbox Pattern
for all microservices for Guaranteed Delivery or At-least-once Delivery - β
Using
Inbox Pattern
for handling Idempotency in reciver side and Exactly-once Delivery - β
Using
UnitTests
andNSubstitute
for mocking dependencies - β
Using
Integration Tests
andEnd To End Tests
on top of testcontainers-dotnet library for cleanup our test enviroment through docker containers - β
Using
Minimal APIs
for handling requests - β
Using
Fluent Validation
and a Validation Pipeline Behaviour on top of MediatR - β
Using
Postgres
for write database as relational DB andMongoDB
andElasric Search
for read database - β
Using docker and
docker-compose
for deployment - β Using Microsoft Tye for deployment
- β Using YARP reverse proxy as API Gateway
- β
Using different type of tests like
Unit Tests
,Integration Tests
,End-To-End Tests
and testcontainers for testing in isolation - π§ Using
Helm
,Kubernetes
andKustomize
for deployment - π§ Using
OpenTelemetry
for collectingMetrics
andDistributed Traces
This project is in progress, new features will be added over time.
Feature | Architecture Pattern | Status | CI-CD |
---|
- βοΈ
.NET 8
- .NET Framework and .NET Core, including ASP.NET and ASP.NET Core - βοΈ
MassTransit
- Distributed Application Framework for .NET - βοΈ
StackExchange.Redis
- General purpose redis client - βοΈ
Npgsql Entity Framework Core Provider
- Npgsql has an Entity Framework (EF) Core provider. It behaves like other EF Core providers (e.g. SQL Server), so the general EF Core docs apply here as well - βοΈ
EventStore-Client-Dotnet
- Dotnet Client SDK for the Event Store gRPC Client API written in C# - βοΈ
FluentValidation
- Popular .NET validation library for building strongly-typed validation rules - βοΈ
Swagger & Swagger UI
- Swagger tools for documenting API's built on ASP.NET Core - βοΈ
Serilog
- Simple .NET logging with fully-structured events - βοΈ
Polly
- Polly is a .NET resilience and transient-fault-handling library that allows developers to express policies such as Retry, Circuit Breaker, Timeout, Bulkhead Isolation, and Fallback in a fluent and thread-safe manner - βοΈ
Scrutor
- Assembly scanning and decoration extensions for Microsoft.Extensions.DependencyInjection - βοΈ
Opentelemetry-dotnet
- The OpenTelemetry .NET Client - βοΈ
DuendeSoftware IdentityServer
- The most flexible and standards-compliant OpenID Connect and OAuth 2.x framework for ASP.NET Core - βοΈ
Newtonsoft.Json
- Json.NET is a popular high-performance JSON framework for .NET - βοΈ
Rabbitmq-dotnet-client
- RabbitMQ .NET client for .NET Standard 2.0+ and .NET 4.6.1+ - βοΈ
AspNetCore.Diagnostics.HealthChecks
- Enterprise HealthChecks for ASP.NET Core Diagnostics Package - βοΈ
Microsoft.AspNetCore.Authentication.JwtBearer
- Handling Jwt Authentication and authorization in .Net Core - βοΈ
NSubstitute
- A friendly substitute for .NET mocking libraries. - βοΈ
StyleCopAnalyzers
- An implementation of StyleCop rules using the .NET Compiler Platform - βοΈ
Mapperly
- A .NET source generator for generating object mappings, No runtime reflection. - βοΈ
IdGen
- Twitter Snowflake-alike ID generator for .Net
This application uses Https
for hosting apis, to setup a valid certificate on your machine, you can create a Self-Signed Certificate, see more about enforce certificate here and here.
- Setup on windows and
powershell
:
dotnet dev-certs https --clean
dotnet dev-certs https -ep $env:USERPROFILE\.aspnet\https\aspnetapp.pfx -p <CREDENTIAL_PLACEHOLDER>
dotnet dev-certs https --trust
- Setup in
linux and wsl
:
dotnet dev-certs https --clean
dotnet dev-certs https -ep ${HOME}/.aspnet/https/aspnetapp.pfx -p <CREDENTIAL_PLACEHOLDER>
dotnet dev-certs https --trust
dotnet dev-certs https --trust
is only supported on macOS and Windows. You need to trust certs on Linux in the way that is supported by your distribution. It is likely that you need to trust the certificate in your browser(with this certificate we don't get an exception for https port because of not found certificate but browser shows us this certificate is not trusted).
In this app I use Conventional Commit and for enforcing its rule I use conventional-changelog/commitlint and typicode/husky with a pre-commit hook. For read more about its setup see commitlint docs and this article and this article.
Here I configured a husky hook for conventional commits:
- Install NPM:
npm init
- Install Husky:
npm install husky --save-dev
- Add
prepare
andinstall-dev-cert-bash
commands for installing and activatinghusky hooks
anddotnet tools
in the package.json file:
- Actually prepare is a special
life cycle scripts
that runs automatically onlocal npm install
without any arguments. - The scripts property of your package.json file supports a number of built-in scripts and their preset life cycle events as well as arbitrary scripts. These all can be executed by running
npm run-script <stage>
ornpm run <stage>
for short. - For working
dotnet tools restore
commands to install and update local packages we should have a validnuget.config
file in the root of our project. we can create anuget.config
file with usingdotnet new nugetconfig
command.
npm pkg set scripts.prepare="husky && dotnet tool restore"
npm pkg set scripts.install-dev-cert-bash="curl -sSL https://aka.ms/getvsdbgsh | bash /dev/stdin -v vs2019 -l ~/vsdbg"
{
"scripts": {
"prepare": "husky && dotnet tool restore",
"install-dev-cert-bash": "curl -sSL https://aka.ms/getvsdbgsh | bash /dev/stdin -v vs2019 -l ~/vsdbg"
}
}
- Install CommitLint:
npm install --save-dev @commitlint/config-conventional @commitlint/cli
- Create the
commitlint.config.js
file with this content:
module.exports = { extends: '@commitlint/config-conventional']};
- Create the Husky folder:
mkdir .husky
- Link Husky and CommitLint:
npx husky add .husky/commit-msg 'npx --no -- commitlint --edit ${1}'
- Activate and installing all husky hooks with this command:
npm run prepare
# this command should run in git-bash on the windows or bash in the linux
npm run install-dev-cert-bash
For formatting, I use mix of belav/csharpier and dotnet format.
- You can integrate csharpier with your prefered IDE.
- You can use your IDE watcher on save for applying
dotnet format
rules. For Rider it is like this:
FileType: C#
Scope: Open Files
Program: Write dotnet
Arguments: format $SolutionPath$ -verbosity diagnostic --include $FileRelativePath$
Advanced (Options)
Disabled all advanced option checkboxes.
All other values were left default
Here I configured a husky hook for formatting:
- Install NPM:
npm init
- Install Husky:
npm install husky --save-dev
- Install manifest file with
dotnet new tool-manifest
because it doesn't exist at first time and then install our required packages as dependency with dotnet tool install, that will add to dotnet-tools.json file in a.config
directory:
dotnet new tool-manifest
dotnet tool install csharpier
- Add
prepare
command for installing and activatinghusky hooks
andrestoring
our dotnet tools in the previous step to the package.json file:
npm pkg set scripts.prepare="husky && dotnet tool restore"
- Create the Husky folder:
mkdir .husky
- Link Husky and formatting tools:
npx husky add .husky/pre-commit "dotnet format --verbosity diagnostic"
npx husky add .husky/pre-commit "dotnet csharpier . && git add -A ."
- Activate and installing all husky hooks with this command:
npm run prepare
For roslyn analizers I use serveral analyzers and config the in .editorconfig
file:
- StyleCop/StyleCop
- JosefPihrt/Roslynator
- meziantou/Meziantou.Analyzer
- Microsoft.VisualStudio.Threading.Analyzers
TODO
The bellow architecture shows that there is one public API (API Gateway) which is accessible for the clients and this is done via HTTP request/response. The API gateway then routes the HTTP request to the corresponding microservice. The HTTP request is received by the microservice that hosts its own REST API. Each microservice is running within its own AppDomain
and has directly access to its own dependencies such as databases, files, local transaction, etc. All these dependencies are only accessible for that microservice and not to the outside world. In fact microservices are decoupled from each other and are autonomous. This also means that the microservice does not rely on other parts in the system and can run independently of other services.
Microservices are event based which means they can publish and/or subscribe to any events occurring in the setup. By using this approach for communicating between services, each microservice does not need to know about the other services or handle errors occurred in other microservices.
In this architecture we use CQRS Pattern for separating read and write model beside of other CQRS Advantages. Here for now I don't use Event Sourcing for simplicity but I will use it in future for syncing read and write side with sending streams and using Projection Feature for some subscribers to syncing their data through sent streams and creating our Custom Read Models in subscribers side.
Here I have a write model that uses a postgres database for handling better Consistency
and ACID Transaction
guaranty. beside o this write side I use a read side model that uses MongoDB for better performance of our read side without any joins with suing some nested document in our document also better scalability with some good scaling features of MongoDB.
For syncing our read side and write side we have 2 options with using Event Driven Architecture (without using events streams in event sourcing):
-
If our
Read Sides
are inSame Service
, during saving data in write side I save a Internal Command record in myCommand Processor
storage (like something we do in outbox pattern) and after commiting write side, ourcommand processor manager
reads unsent commands and sends them to theirCommand Handlers
in same corresponding service and this handlers could save their read models in our MongoDb database as a read side. -
If our
Read Sides
are inAnother Services
we publish an integration event (with saving this message in the outbox) after committing our write side and all of ourSubscribers
could get this event and save it in their read models (MongoDB).
All of this is optional in the application and it is possible to only use what that the service needs. Eg. if the service does not want to Use DDD because of business is very simple and it is mostly CRUD
we can use Data Centric
Architecture or If our application is not Task based
instead of CQRS and separating read side and write side again we can just use a simple CRUD
based application.
Here I used Outbox for Guaranteed Delivery and can be used as a landing zone for integration events before they are published to the message broker .
Outbox pattern ensures that a message was sent (e.g. to a queue) successfully at least once. With this pattern, instead of directly publishing a message to the queue, we put it in the temporary storage (e.g. database table) for preventing missing any message and some retry mechanism in any failure (At-least-once Delivery). For example When we save data as part of one transaction in our service, we also save messages (Integration Events) that we later want to process in another microservices as part of the same transaction. The list of messages to be processed is called a StoreMessage with Message Delivery Type Outbox
that are part of our MessagePersistence service. This infrastructure also supports Inbox
Message Delivery Type and Internal
Message Delivery Type (Internal Processing).
Also we have a background service MessagePersistenceBackgroundService that periodically checks the our StoreMessages in the database and try to send the messages to the broker with using our MessagePersistenceService service. After it gets confirmation of publishing (e.g. ACK from the broker) it marks the message as processed to avoid resending
.
However, it is possible that we will not be able to mark the message as processed due to communication error, for example broker
is unavailable
. In this case our MessagePersistenceBackgroundService try to resend the messages that not processed and it is actually At-Least-Once delivery. We can be sure that message will be sent once
, but can be sent multiple times
too! Thatβs why another name for this approach is Once-Or-More delivery. We should remember this and try to design receivers of our messages as Idempotents, which means:
In Messaging this concepts translates into a message that has the same effect whether it is received once or multiple times. This means that a message can safely be resent without causing any problems even if the receiver receives duplicates of the same message.
For handling Idempotency and Exactly-once Delivery in receiver side, we could use Inbox Pattern.
This pattern is similar to Outbox Pattern. Itβs used to handle incoming messages (e.g. from a queue) for unique processing
of a single message
only once
(even with executing multiple time). Accordingly, we have a table in which weβre storing incoming messages. Contrary to outbox pattern, we first save the messages in the database, then weβre returning ACK to queue. If save succeeded, but we didnβt return ACK to queue, then delivery will be retried. Thatβs why we have at-least-once delivery again. After that, an inbox background process
runs and will process the inbox messages that not processed yet. also we can prevent executing a message with specific MessgaeId
multiple times. after executing our inbox message for example with calling our subscribed event handlers we send a ACK to the queue when they succeeded. (Inbox part of the system is in progress, I will cover this part soon as possible)
Also here I used RabbitMQ
as my Message Broker
for my async communication between the microservices with using eventually consistency mechanism, for now I used MassTransit tools for doing broker communications. beside of this eventually consistency we have a synchronous call with using REST
(in future I will use gRpc) for our immediate consistency needs.
We use a Api Gateway
and here I used YARP that is microsoft reverse proxy (we could use envoy, traefik, Ocelot, ...), in front of our services, we could also have multiple Api Gateway for reaching BFF pattern. for example one Gateway for mobile apps, One Gateway for web apps and etc.
With using api Gateway our internal microservices are transparent and user can not access them directly and all requests will serve through this Gateway.
Also we could use gateway for load balancing, authentication and authorization, caching ,...
In this project I used vertical slice architecture or Restructuring to a Vertical Slice Architecture also I used feature folder structure in this project.
- We treat each request as a distinct use case or slice, encapsulating and grouping all concerns from front-end to back.
- When We adding or changing a feature in an application in n-tire architecture, we are typically touching many different "layers" in an application. we are changing the user interface, adding fields to models, modifying validation, and so on. Instead of coupling across a layer, we couple vertically along a slice and each change affects only one slice.
- We
Minimize coupling
between slices
, andmaximize coupling
in a slice
. - With this approach, each of our vertical slices can decide for itself how to best fulfill the request. New features only add code, we're not changing shared code and worrying about side effects. For implementing vertical slice architecture using cqrs pattern is a good match.
Also here I used CQRS for decompose my features to very small parts that makes our application:
- maximize performance, scalability and simplicity.
- adding new feature to this mechanism is very easy without any breaking change in other part of our codes. New features only add code, we're not changing shared code and worrying about side effects.
- easy to maintain and any changes only affect on one command or query (or a slice) and avoid any breaking changes on other parts
- it gives us better separation of concerns and cross cutting concern (with help of MediatR behavior pipelines) in our code instead of a big service class for doing a lot of things.
With using CQRS, our code will be more aligned with SOLID principles, especially with:
- Single Responsibility rule - because logic responsible for a given operation is enclosed in its own type.
- Open-Closed rule - because to add new operation you donβt need to edit any of the existing types, instead you need to add a new file with a new type representing that operation.
Here instead of some Technical Splitting for example a folder or layer for our services
, controllers
and data models
which increase dependencies between our technical splitting and also jump between layers or folders, We cut each business functionality into some vertical slices, and inner each of these slices we have Technical Folders Structure specific to that feature (command, handlers, infrastructure, repository, controllers, data models, ...).
Usually, when we work on a given functionality we need some technical things for example:
- API endpoint (Controller)
- Request Input (Dto)
- Request Output (Dto)
- Some class to handle Request, For example Command and Command Handler or Query and Query Handler
- Data Model
Now we could all of these things beside each other and it decrease jumping and dependencies between some layers or folders.
Keeping such a split works great with CQRS. It segregates our operations and slices the application code vertically instead of horizontally. In Our CQRS pattern each command/query handler is a separate slice. This is where you can reduce coupling between layers. Each handler can be a separated code unit, even copy/pasted. Thanks to that, we can tune down the specific method to not follow general conventions (e.g. use custom SQL query or even different storage). In a traditional layered architecture, when we change the core generic mechanism in one layer, it can impact all methods.
TODO
TODO
- This application uses
Https
for hosting apis, to setup a valid certificate on your machine, you can create a Self-Signed Certificate, see more about enforce certificate here. - Install git - https://git-scm.com/downloads.
- Install .NET Core 7.0 - https://dotnet.microsoft.com/download/dotnet/7.0.
- Install Visual Studio, Rider or VSCode.
- Install docker - https://docs.docker.com/docker-for-windows/install/.
- Make sure that you have ~10GB disk space.
- Clone Project https://github.com/mehdihadeli/food-delivery-microservices-sample, make sure that's compiling
- Run the docker-compose.infrastructure.yaml file, for running prerequisites infrastructures with
docker-compose -f ./deployments/docker-compose/docker-compose.infrastructure.yaml up -d
command. - Open food-delivery-microservices.sln solution.
For Running this application we could run our microservices one by one in our Dev Environment, for me, it's Rider, Or we could run it with using Docker-Compose or we could use Kubernetes.
For testing apis I used REST Client plugin of VSCode its related file scenarios are available in _httpclients folder. also after running api you have access to swagger open api
for all microservices in /swagger
route path.
In this application I use a fake email sender
with name of ethereal as a SMTP provider for sending email. after sending email by the application you can see the list of sent emails in ethereal messages panel. My temp username and password is available inner the all of appsettings file.
For ruining all microservices and control on their running mode we could use PM2 tools. for installing pm2
on our system globally we should use this command:
npm install pm2 -g
After installing pm2 on our machine, we could run all of our microservices with running bellow command in root of the application with using pm2.yaml file.
pm2 start pm2.yaml
Some PM2 useful commands:
pm2 -h
pm2 list
pm2 logs
pm2 monit
pm2 info pm2.yaml
pm2 stop pm2.yaml
pm2 restart pm2.yaml
pm2 delete pm2.yaml
- First we should create a dev-certificate for our docker-compose file with this commands, see more about enforce certificate here:
dotnet dev-certs https --clean
dotnet dev-certs https -ep ${HOME}/.aspnet/https/aspnetapp.pfx -p $CREDENTIAL_PLACEHOLDER$
dotnet dev-certs https --trust
This local certificate will mapped to our containers in docker-compose file with setting ~/.aspnet/https:/https:ro
volume mount
- Our docker-compose files are based on linux
- Run the docker-compose.infrastructure.yaml file, for running prerequisites infrastructures with
docker-compose -f ./deployments/docker-compose/docker-compose.infrastructure.yaml up -d
command. - Run the docker-compose.services.yaml with
docker-compose -f ./deployments/docker-compose/docker-compose.services.yaml
for production mode that uses pushed docker images for services or for development mode you can use docker-compose.services.dev.yaml override docker-compose file withdocker-compose -f ./deployments/docker-compose/docker-compose.services.yaml -f ${workspaceFolder}/deployments/docker-compose/docker-compose.services.dev.yaml up
command for buildingdockerfiles
instead of using images in docker registry. Also fordebugging
purpose of docker-containers in vscode you can use ./deployments/docker-compose/docker-compose.services.debug.yaml override docker-compose file with runningdocker-compose -f ./deployments/docker-compose/docker-compose.services.yaml -f ${workspaceFolder}/deployments/docker-compose/docker-compose.services.debug.yaml up -d
, I defined some tasks for vscode for executing this command easier. For debugging in vscode we should use launch.json. - Wait until all dockers got are downloaded and running.
- You should automatically get:
- Postgres running
- RabbitMQ running
- MongoDB running
- Microservies running and accessible
Some useful docker commands:
// start dockers
docker-compose -f .\docker-compose.yaml up
// build without caching
docker-compose -f .\docker-compose.yaml build --no-cache
// to stop running dockers
docker-compose kill
// to clean stopped dockers
docker-compose down -v
// showing running dockers
docker ps
// to show all dockers (also stopped)
docker ps -a
We could run our microservices with new microsoft tools with name of Project Tye.
Project Tye is an experimental developer tool that makes developing, testing, and deploying microservices and distributed applications easier.
For installing Tye
globally on our machine we should use this command:
dotnet tool install -g Microsoft.Tye --version "0.11.0-alpha.22111.1"
OR if you already have Tye installed and want to update:
dotnet tool update -g Microsoft.Tye
After installing tye, we could run our microservices with following command in the root of our project:
tye run
One of key feature from tye run is a dashboard to view the state of your application. Navigate to http://localhost:8000 to see the dashboard running.
Also We could run some docker images with Tye and Tye makes the process of deploying your application to Kubernetes very simple with minimal knowledge or configuration required.
For using kubernetes we can use multiple approach with different tools like plain kubernetes
, kustomize
and helm
and here I will use show use of all of them.
Here I uses plain kubernetes command and resources for applying kubernetes manifest to the cluster. If you're a Docker/Compose user new to Kubernetes, you might have noticed that you canβt substitutes
and replace
the environment variables
in your kubernetes manifest files (exception is substitutes
and replace
environment variables in env
attribute, with using environment dependent variable
). we often have some personal .env
files for projects that we use to store credentials and configurations.
For substitutes
and replace
environment variables in our kubernetes manifest
or resource
files we can use envsubst tools, and we can even pipe it into other commands like Kubernetes kubectl
, read more in this and this articles.
So here we should first install this tools on our OS with this guid.
Now for running manifest files, firstly we should load
our environment variables
inner .env
file in our current shell session
by using source .env
command (after closing our shell environments will destroy).
Make sure to use
export
, otherwise our variables are considered shell variables and might not be accessible toenvsubst
Here is a example of .env
file:
export ASPNETCORE_ENVIRONMENT=docker
export REGISTRY=ghcr.io
After loading environment variables
to the our shell session
we can run manifests with using envsubst
and pipe envsubst output to kubectl
command as a input like this example:
# pipe a deployment "deploy.yml" into kubectl apply
envsubst < deploy.yml | kubectl apply -f -
Also it is also possible to write our substitution
to a new file
(envsubst < input.tmpl > output.text):
envsubst < deploy.yml > compiled_deploy.yaml
I've create a shell script kubectl, we can call this script
just like we would kubectl
. This script sources any .env
file in the manifests directory , If we're running ./kubectl apply
or ./kubectl delete
with kubectl script, it calls envsubst
to swap out
your environment variables, then it passes the code on to the real kubectl
.
#!/bin/bash
ENV_FILE=.env
source $ENV_FILE
if [[ "$1" == "apply" ]] || [[ "$1" == "delete" ]]; then
envsubst < $3 | kubectl $1 $2 -
else
kubectl "$@"
fi
- For installing
Infrastructure
manifests for kubernetes cluster we run bellow command withkubectl script
:
./kubectl apply -f ./deployments/k8s/kubernetes/infrastructure.yaml
The application is in development status. You are feel free to submit pull request or create the issue.
- https://github.com/oskardudycz/EventSourcing.NetCore
- https://github.com/dotnet-architecture/eShopOnContainers
- https://github.com/jbogard/ContosoUniversityDotNetCore-Pages
- https://github.com/kgrzybek/modular-monolith-with-ddd
- https://github.com/thangchung/clean-architecture-dotnet
The project is under MIT license.