Skip to main content

Project structure

The example counter project demonstrates a way to structure your applications.

The application is divided up into multiple packages, each with its own purpose.

  • cdk - Infrastructure setup for deploying the application.
  • components - templ components.
  • db - Database access code used to increment and get counts.
  • handlers - HTTP handlers.
  • lambda - The AWS Lambda entry point.
  • services - Services used by the handlers.
  • session - Middleware for implementing HTTP session IDs.
  • main.go - Used to run the application locally.

Application architecture

The architecture follows a typical "onion model" where each layer doesn't know about the layer above it, and each layer is responsible for a specific thing.

  • HTTP Handler
    • Processes HTTP requests
    • Does not contain application logic itself
    • Uses services that carry out application logic
    • Takes the responses from services and uses components to render HTML
    • Creates HTTP responses
  • Services
    • Carries out application logic such as orchestrating API calls, or making database calls
    • Does not do anything related to HTML or HTTP
    • Is not aware of the specifics of database calls
  • Database access code
    • Handles database activity such as inserting and querying records
    • Ensures that the database representation (records) doesn't leak to the service layer

A more complex application may have a models package containing plain structs that represent common data structures in the application, such as User.


As with most things, taking the layering approach to an extreme level can have a negative effect. Ask yourself whether what you're doing is really helping to make the code understandable, or is just spreading application logic across lots of files, and making it hard to see the overall structure.

Dependency injection

Layering an application in this way can simplify code structure, since the responsibility of each type is clear.

To ensure that each part of the application is initialized with its dependencies, each struct defines a constructor (the New function in this example).

As per the HTTP handler defines the interface that it's expecting, rather than the service defining its own interface.

type CountService interface {
Increment(ctx context.Context, it services.IncrementType, sessionID string) (counts services.Counts, err error)
Get(ctx context.Context, sessionID string) (counts services.Counts, err error)

func New(log *slog.Logger, cs CountService) *DefaultHandler {
return &DefaultHandler{
Log: log,
CountService: cs,

type DefaultHandler struct {
Log *slog.Logger
CountService CountService

Changing the signature of New to add a new dependency will result in a compilation error that shows you all the affected code in your application.


Dependency injection frameworks are not typically used in Go. If you're coming from a language like C# or Java, this may seem unusual to you, but go with it, you don't need one.

HTTP layer

This HTTP handler reads HTTP requests, uses the CountService to Get or Increment the counters, and renders the templ Components.


Note that the View method uses the templ Components from the components directory to render the page.

func (h *DefaultHandler) ServeHTTP(w http.ResponseWriter, r *http.Request) {
if r.Method == http.MethodPost {
h.Post(w, r)
h.Get(w, r)

func (h *DefaultHandler) Get(w http.ResponseWriter, r *http.Request) {
var props ViewProps
var err error
props.Counts, err = h.CountService.Get(r.Context(), session.ID(r))
if err != nil {
h.Log.Error("failed to get counts", slog.Any("error", err))
http.Error(w, "failed to get counts", http.StatusInternalServerError)
h.View(w, r, props)

func (h *DefaultHandler) Post(w http.ResponseWriter, r *http.Request) {

// Decide the action to take based on the button that was pressed.
var it services.IncrementType
if r.Form.Has("global") {
it = services.IncrementTypeGlobal
if r.Form.Has("session") {
it = services.IncrementTypeSession

counts, err := h.CountService.Increment(r.Context(), it, session.ID(r))
if err != nil {
h.Log.Error("failed to increment", slog.Any("error", err))
http.Error(w, "failed to increment", http.StatusInternalServerError)

// Display the view.
h.View(w, r, ViewProps{
Counts: counts,

type ViewProps struct {
Counts services.Counts

func (h *DefaultHandler) View(w http.ResponseWriter, r *http.Request, props ViewProps) {
components.Page(props.Counts.Global, props.Counts.Session).Render(r.Context(), w)

Service layer

The service layer coordinates API and database activity to carry out application logic.

type Counts struct {
Global int
Session int

func (cs Count) Get(ctx context.Context, sessionID string) (counts Counts, err error) {
globalAndSessionCounts, err := cs.CountStore.BatchGet(ctx, "global", sessionID)
if err != nil {
err = fmt.Errorf("countservice: failed to get counts: %w", err)
if len(globalAndSessionCounts) != 2 {
err = fmt.Errorf("countservice: unexpected counts returned, expected 2, got %d", len(globalAndSessionCounts))
counts.Global = globalAndSessionCounts[0]
counts.Session = globalAndSessionCounts[1]

This allows us to use Go's parallelism features to run operations more efficiently without adding complexity to the HTTP or database code.

func (cs Count) Increment(ctx context.Context, it IncrementType, sessionID string) (counts Counts, err error) {
// Work out which operations to do.
var global, session func(ctx context.Context, id string) (count int, err error)
switch it {
case IncrementTypeGlobal:
global = cs.CountStore.Increment
session = cs.CountStore.Get
case IncrementTypeSession:
global = cs.CountStore.Get
session = cs.CountStore.Increment
return counts, ErrUnknownIncrementType

// Run the operations in parallel.
var wg sync.WaitGroup
errs := make([]error, 2)
go func() {
defer wg.Done()
counts.Global, errs[0] = global(ctx, "global")
go func() {
defer wg.Done()
counts.Session, errs[1] = session(ctx, sessionID)

return counts, errors.Join(errs...)


To wire all of the dependencies together and start up your web server or serverless function handler, your application will require an entrypoint.

In this example, the code for configuring the HTTP server and HTTP routes is also in the main.go because it's a very simple application. In more complex applications, this might be migrated into another package.

package main

import (


func main() {
log := slog.New(slog.NewJSONHandler(os.Stderr))
s, err := db.NewCountStore(os.Getenv("TABLE_NAME"), os.Getenv("AWS_REGION"))
if err != nil {
log.Error("failed to create store", slog.Any("error", err))
cs := services.NewCount(log, s)
h := handlers.New(log, cs)

var secureFlag = true
if os.Getenv("SECURE_FLAG") == "false" {
secureFlag = false

// Add session middleware.
sh := session.NewMiddleware(h, session.WithSecure(secureFlag))

server := &http.Server{
Addr: "localhost:9000",
Handler: sh,
ReadTimeout: time.Second * 10,
WriteTimeout: time.Second * 10,

fmt.Printf("Listening on %v\n", server.Addr)