go-clean-template

Go Clean template

🇨🇳 中文 🇷🇺 RU

Clean Architecture template for Golang services

Overview

The purpose of the template is to show:

• how to organize a project and prevent it from turning into spaghetti code
• where to store business logic so that it remains independent, clean, and extensible
• how not to lose control when a microservice grows

Using the principles of Robert Martin (aka Uncle Bob).

Go-clean-template is created & supported by Evrone.

This template implements four types of servers:

• AMQP RPC (based on RabbitMQ as transport and Request-Reply pattern)
• MQ RPC (based on NATS as transport and Request-Reply pattern)
• gRPC (gRPC framework based on protobuf)
• REST API (Fiber framework)

The template includes three domains to demonstrate multi-service architecture:

• User Authentication — registration, login, JWT-based authorization
• Task Management — CRUD operations with status transitions (todo, in_progress, done)
• Translation — text translation with history tracking

All domains are available across all four transports (REST, gRPC, AMQP RPC, NATS RPC).

Content

• Domains
• Quick start
• Observability
• Project structure
• Dependency Injection
• Clean Architecture

Domains

The template includes three fully implemented domains, each available across all four transports (REST, gRPC, AMQP RPC, NATS RPC).

User Authentication

Registration, login, and JWT-based authorization.

Operation	REST	gRPC
Register	POST /v1/auth/register	AuthService/Register
Login	POST /v1/auth/login	AuthService/Login
Get profile	GET /v1/user/profile	AuthService/GetProfile

• Passwords hashed with bcrypt
• JWT tokens with configurable expiry
• Auth middleware on all transports

Task Management

CRUD operations with a status state machine.

Operation	REST	gRPC
Create	POST /v1/tasks	TaskService/CreateTask
List	GET /v1/tasks	TaskService/ListTasks
Get	GET /v1/tasks/:id	TaskService/GetTask
Update	PUT /v1/tasks/:id	TaskService/UpdateTask
Transition	PATCH /v1/tasks/:id/status	TaskService/TransitionTask
Delete	DELETE /v1/tasks/:id	TaskService/DeleteTask

• Status transitions: todo → in_progress → done (and in_progress → todo)
• Pagination with limit/offset and optional status filter
• Tasks scoped to the authenticated user

Translation

Text translation via external API with history tracking.

Operation	REST	gRPC
Translate	POST /v1/translation/do-translate	TranslationHistoryService/DoTranslate
History	GET /v1/translation/history	TranslationHistoryService/ShowHistory

Quick start

Local development

# Postgres, RabbitMQ, NATS
make compose-up
# Run app with migrations
make run

Integration tests (can be run in CI)

# DB, app + migrations, integration tests
make compose-up-integration-test

Full docker stack with reverse proxy

make compose-up-all 

Check services:

• AMQP RPC:
  • URL: amqp://guest:[email protected]:5672/
  • Client Exchange: rpc_client
  • Server Exchange: rpc_server
• NATS RPC:
  • URL: nats://guest:[email protected]:4222/
  • Server Exchange: rpc_server
• REST API:
  • http://app.lvh.me/healthz | http://127.0.0.1:8080/healthz
  • http://app.lvh.me/metrics | http://127.0.0.1:8080/metrics
  • http://app.lvh.me/swagger | http://127.0.0.1:8080/swagger
• gRPC:
  • URL: tcp://grpc.lvh.me:8081 | tcp://127.0.0.1:8081
  • v1/auth.proto
  • v1/task.proto
  • v1/translation.history.proto
• PostgreSQL:
  • postgres://user:myAwEsOm3pa55@[email protected]:5432/db
• RabbitMQ:
  • http://rabbitmq.lvh.me | http://127.0.0.1:15672
  • Credentials: guest / guest
• NATS monitoring:
  • http://nats.lvh.me | http://127.0.0.1:8222/
  • Credentials: guest / guest
• Jaeger (traces UI):
  • http://jaeger.lvh.me | http://127.0.0.1:16686

Observability

Distributed tracing is provided by OpenTelemetry. Spans are exported over OTLP/gRPC to a collector — Jaeger in the docker stack.

• Context propagation — W3C traceparent + baggage, so a single trace spans all four transports. REST uses the otelfiber middleware, gRPC uses the otelgrpc stats handler, and AMQP RPC / NATS RPC carry the trace context in message headers via custom carriers (pkg/rabbitmq/rmq_rpc/otel_carrier.go, pkg/nats/nats_rpc/otel_carrier.go).
• Instrumented layers — use cases and repositories are wrapped in tracing decorators (*/tracing.go), so a trace shows the full path: controller → usecase → repo / webapi.
• No-op when disabled — with TRACING_ENABLED=false a no-op tracer provider is installed, so instrumentation runs unconditionally with zero overhead.
• Sampling — TRACING_SAMPLE_RATE is a parent-based ratio. Use 1.0 locally; lower it (0.1 / 0.01) in production.
• Internal routes (/healthz, /metrics, /swagger) are excluded from tracing — only the versioned API groups are instrumented.

Configuration (see .env.example):

Variable	Default	Description
TRACING_ENABLED	false	Enable OpenTelemetry tracing
TRACING_OTLP_ENDPOINT	localhost:4317	OTLP/gRPC collector endpoint
TRACING_OTLP_INSECURE	true	Disable TLS for the OTLP exporter
TRACING_SAMPLE_RATE	0.1	Parent-based sampling ratio

Project structure

cmd/app/main.go

Configuration and logger initialization. Then the main function “continues” in internal/app/app.go.

config

The twelve-factor app stores config in environment variables (often shortened to env vars or env). Env vars are easy to change between deploys without changing any code; unlike config files, there is little chance of them being checked into the code repo accidentally; and unlike custom config files, or other config mechanisms such as Java System Properties, they are a language- and OS-agnostic standard.

Config: config.go

Example: .env.example

docker-compose.yml uses env variables to configure services.

docs

Swagger documentation. Auto-generated by swag library. You don’t need to correct anything by yourself.

docs/proto

Protobuf files. They are used to generate Go code for gRPC services. The proto files are also used to generate documentation for gRPC services. You don’t need to correct anything by yourself.

integration-test

Integration tests. They are launched as a separate container, next to the application container.

internal/app

There is always one Run function in the app.go file, which “continues” the main function.

This is where all the main objects are created. Dependency injection occurs through the “New …” constructors (see Dependency Injection). This technique allows us to layer the application using the Dependency Injection principle. This makes the business logic independent from other layers.

Next, we start the server and wait for signals in select for graceful completion. If app.go starts to grow, you can split it into multiple files.

For a large number of injections, wire can be used.

The migrate.go file is used for database auto migrations. It is included if an argument with the migrate tag is specified. For example:

go run -tags migrate ./cmd/app

internal/controller

Server handler layer (MVC controllers). The template shows 4 servers:

• AMQP RPC (based on RabbitMQ as transport)
• NATS RPC (based on NATS as transport)
• gRPC (gRPC framework based on protobuf)
• REST API (Fiber framework)

Server routers are written in the same style:

• Handlers are grouped by area of application (by a common basis)
• For each group, its own router structure is created, the methods of which process paths
• The structure of the business logic is injected into the router structure, which will be called by the handlers

internal/controller/amqp_rpc

Simple RPC versioning. For v2, we will need to add the amqp_rpc/v2 folder with the same content. And in the file internal/controller/amqp_rpc/router.go add the line:

routes := make(map[string]server.CallHandler)

{
    v1.NewRoutes(routes, t, u, tk, j, l)
}

{
    v2.NewTranslationRoutes(routes, t, l)
}

internal/controller/grpc

Simple gRPC versioning. For v2, we will need to add the grpc/v2 folder with the same content. Also add the v2 folder to the proto files in docs/proto. And in the file internal/controller/grpc/router.go add the line:

{
    v1.NewAuthRoutes(app, u, l)
    v1.NewTaskRoutes(app, tk, l)
    v1.NewTranslationRoutes(app, t, l)
}

{
    v2.NewAuthRoutes(app, u, l)
    v2.NewTaskRoutes(app, tk, l)
    v2.NewTranslationRoutes(app, t, l)
}

reflection.Register(app)

internal/controller/nats_rpc

Simple RPC versioning. For v2, we will need to add the nats_rpc/v2 folder with the same content. And in the file internal/controller/nats_rpc/router.go add the line:

routes := make(map[string]server.CallHandler)

{
    v1.NewRoutes(routes, t, u, tk, j, l)
}

{
    v2.NewTranslationRoutes(routes, t, l)
}

internal/controller/restapi

Simple REST versioning. For v2, we will need to add the restapi/v2 folder with the same content. And in the file internal/controller/restapi/router.go add the line:

apiV1Group := app.Group("/v1")
{
	v1.NewRoutes(apiV1Group, t, u, tk, jwtManager, l)
}
apiV2Group := app.Group("/v2")
{
	v2.NewRoutes(apiV2Group, t, u, tk, jwtManager, l)
}

Instead of Fiber, you can use any other http framework.

In router.go and above the handler methods, there are comments for generating swagger documentation using swag.

internal/entity

Entities of business logic (models) can be used in any layer. There can also be methods, for example, for validation.

internal/usecase

Business logic.

• Methods are grouped by area of application (on a common basis)
• Each group has its own structure
• One file - one structure

Repositories, webapi, rpc, and other business logic structures are injected into business logic structures (see Dependency Injection).

internal/repo/persistent

A repository is an abstract storage (database) that business logic works with.

internal/repo/webapi

It is an abstract web API that business logic works with. For example, it could be another microservice that business logic accesses via the REST API. The package name changes depending on the purpose.

pkg/rabbitmq

RabbitMQ RPC pattern:

• There is no routing inside RabbitMQ
• Exchange fanout is used, to which 1 exclusive queue is bound, this is the most productive config
• Reconnect on the loss of connection

pkg/tracing

OpenTelemetry setup: configures a global TracerProvider with an OTLP/gRPC exporter and W3C trace-context + baggage propagators. When tracing is disabled, a no-op provider is installed so instrumentation code runs unconditionally. See Observability.

Dependency Injection

In order to remove the dependence of business logic on external packages, dependency injection is used.

For example, through the New constructor, we inject the dependency into the structure of the business logic. This makes the business logic independent (and portable). We can override the implementation of the interface without making changes to the usecase package.

package usecase

import (
// Nothing!
)

type Repository interface {
	Get()
}

type UseCase struct {
	repo Repository
}

func New(r Repository) *UseCase {
	return &UseCase{
		repo: r,
	}
}

func (uc *UseCase) Do() {
	uc.repo.Get()
}

It will also allow us to do auto-generation of mocks (for example with go.uber.org/mock) and easily write unit tests.

We are not tied to specific implementations in order to always be able to change one component to another. If the new component implements the interface, nothing needs to be changed in the business logic.

Clean Architecture

Key idea

Programmers realize the optimal architecture for an application after most of the code has been written.

A good architecture allows decisions to be delayed to as late as possible.

The main principle

Dependency Inversion (the same one from SOLID) is the principle of dependency injection. The direction of dependencies goes from the outer layer to the inner layer. Due to this, business logic and entities remain independent from other parts of the system.

So, the application is divided into 2 layers, internal and external:

1. Business logic (Go standard library).
2. Tools (databases, servers, message brokers, any other packages and frameworks).

The inner layer with business logic should be clean. It should:

• Not have package imports from the outer layer.
• Use only the capabilities of the standard library.
• Make calls to the outer layer through the interface (!).

The business logic doesn’t know anything about Postgres or a specific web API. Business logic has an interface for working with an abstract database or abstract web API.

The outer layer has other limitations:

• All components of this layer are unaware of each other’s existence. How to call another from one tool? Not directly, only through the inner layer of business logic.
• All calls to the inner layer are made through the interface (!).
• Data is transferred in a format that is convenient for business logic (internal/entity).

For example, you need to access the database from HTTP (controller). Both HTTP and database are in the outer layer, which means they know nothing about each other. The communication between them is carried out through usecase (business logic):

    HTTP > usecase
           usecase > repository (Postgres)
           usecase < repository (Postgres)
    HTTP < usecase

The symbols > and < show the intersection of layer boundaries through Interfaces. The same is shown in the picture:

Or more complex business logic:

    HTTP > usecase
           usecase > repository
           usecase < repository
           usecase > webapi
           usecase < webapi
           usecase > RPC
           usecase < RPC
           usecase > repository
           usecase < repository
    HTTP < usecase

Layers

Clean Architecture Terminology

• Entities are structures that business logic operates on. They are located in the internal/entity folder. In MVC terms, entities are models.
• Use Cases is business logic located in internal/usecase.

The layer with which business logic directly interacts is usually called the infrastructure layer. These can be repositories internal/repo/persistent, external webapi internal/repo/webapi, any pkg, and other microservices. In the template, the infrastructure packages are located inside internal/repo.

You can choose how to call the entry points as you wish. The options are:

• controller (in our case)
• delivery
• transport
• gateways
• entrypoints
• primary
• input

Additional layers

The classic version of Clean Architecture was designed for building large monolithic applications and has 4 layers.

In the original version, the outer layer is divided into two more, which also have an inversion of dependencies to each other (directed inward) and communicate through interfaces.

The inner layer is also divided into two (with separation of interfaces), in the case of complex logic.

---

Complex tools can be divided into additional layers. However, you should add layers only if really necessary.

Alternative approaches

In addition to Clean architecture, Onion architecture and Hexagonal (Ports and adapters) are similar to it. Both are based on the principle of Dependency Inversion. Ports and adapters are very close to Clean Architecture, the differences are mainly in terminology.

Similar projects

• https://github.com/bxcodec/go-clean-arch
• https://github.com/zhashkevych/courses-backend

Useful links

• The Clean Architecture article
• Twelve factors