This tutorial will guide you through the process of creating a TODO REST API backend application using Kotlin & Klite, including Postgres database.
As you will see, Klite is very small, but still provides turn-the-key features for modern web applications, more so than many other frameworks.
To get more information about any class or function, navigate freely inside to see how it works and what optional parameters does it provide.
Create a Kotlin Gradle project in IntelliJ IDEA, then add the Klite dependency to the build.gradle.kts file:
repositories {
mavenCentral()
maven { url = uri("https://jitpack.io") }
}
dependencies {
val kliteVersion = "paste latest version here"
implementation("com.github.codeborne.klite:klite-server:$kliteVersion")
}Fortunately, you won't get anything extra besides Klite itself, as it has no other dependencies, not even 3rd-party http server.
Good frameworks allow you to write your own main function, so that you are in control of the application lifecycle. In Klite, you need to create an instance of Server:
import klite.Server
fun main() {
Server().start()
}This will start a server on the default port 8080.
You can change the listening IP and port by passing it as an argument to the Server constructor.
Most applications need some configuration. 12-factor applications use environment variables for configuration.
Klite provides a simple way to access environment variables using the Config object, and the Server already uses it to read the port such as Config["PORT"], among other things.
In development, it makes sense to use an .env file to set environment variables, and also commit this file, so that fellow developers will get things running out of the box:
fun main() {
Config.useEnvFile() // loads .env file if it exists, and uses values from file only if they are not set in the environment
Server().start()
}Klite also supports defining environment name with the ENV environment variable. Default values are dev, test, prod.
You may want to disable loading of .env file in production, and use environment variables only:
if (!Config.isProd) Config.useEnvFile()Even if you don't add this if, Config still has precedence for environment variables over .env file.
It's time to add our routes. Klite routes must be defined in a context (URL prefix with common meaning).
Server().apply {
context("/api") {
get("/hello") { "Hello, world!" }
}
start()
}You can now get the plain text response from http://localhost:8080/api/hello.
This will also handle request logging for you.
As is typical, we want our API to return JSON.
Klite uses registered BodyRenderers to render the response. By default, it uses TextRenderer to render plain text.
You can register more renderers to handle different content types.
Klite uses the Accept header to determine which content type the client wants.
Add klite-json dependency to your build.gradle.kts:
implementation("com.github.codeborne.klite:klite-json:$kliteVersion")Note: klite-jackson and klite-serialization are alternatives, if you prefer more heavy-weight and harder to configure libraries.
Now you can register the JsonBody renderer and parser for the whole application:
use<JsonBody>()
context("/api") {
useOnly<JsonBody>() // this removes other renderers in /api context
get("/todos") { listOf(Todo("Buy groceries")) }
}
data class Todo(val item: String, val completedAt: Instant? = null)Klite JsonMapper will omit nulls by default to make responses smaller.
You can now get the json response using http://localhost:8080/api/todos.
In real applications, it makes sense to divide routes into separate classes, which can also be unit-tested just like normal classes, independent of the framework.
Klite provides klite.annotations package with annotations to make it possible.
Let's create a TodoRoutes:
import klite.annotations.*
class TodoRoutes {
@GET("/todos") fun todos() = listOf(Todo("Buy groceries"))
}Then, you can register this class in the /api context:
context("/api") {
useOnly<JsonBody>()
annotated<TodoRoutes>("/todos")
}This will add all annotated methods from TodoRoutes to the context as route handlers.
In some frameworks, this can is called "Controller" or "Resource". You are free to use any name in Klite.
Why not use annotated(TodoRoutes())? Because you may want Klite to create the instance and inject any dependencies into it that it declares as constructor parameters.
Let's create a basic in-memory repository for storing of our todos:
class TodoRepository {
private val todos = mutableListOf<Todo>()
fun list() = todos.toList()
fun save(todo: Todo) = todos.add(todo)
}Now, we can inject this repository into our TodoRoutes:
class TodoRoutes(private val repo: TodoRepository) {
@GET fun todos() = repo.list()
@POST fun save(todo: Todo) = repo.save(todo)
}JsonBody will deserialize Todo instance from json request automatically.
Klite uses Server.registry for dependency injection.
Registry will create singleton classes recursively by default.
If you need to register instances of interfaces, you can use the register() function.
Let's add a route to get a single todo.
First, our Todo class should have an id field:
typealias Id<T> = TSID<T>
data class Todo(val item: String, val completedAt: Instant? = null, val id: Id<Todo> = Id())We use TSID to generate unique and type-safe IDs for our todos, let's have it auto-generate when a new Todo is posted. Alternatively, you can use UUID or any type that you like.
Let's add a corresponding method to the TodoRepository:
fun get(id: Id<Todo>) = todos.first { it.id == id }Now, we can add a route to get a single todo by its id:
@GET("/todos/:id") fun todoById(@PathParam id: Id<Todo>) = repo.get(id)In real applications, you would use a database to store your data.
Let's spin-up a Postgres database using Docker.
Create the following docker-compose.yml file:
service:
db:
image: postgres:alpine
environment:
POSTGRES_USER: todo
POSTGRES_PASSWORD: todo
ports:
- "5432:5432"Then add klite-jdbc dependency to your build.gradle.kts:
implementation("com.github.codeborne.klite:klite-jdbc:$kliteVersion")Now we can auto-start the DB and connect to it in our Launcher:
Server().apply {
if (Config.isDev) startDevDB()
use<DBModule>()
...
}DBModule uses PooledDataSource, wrapping ConfigDataSource by default.
For the connection to succeed, you need to add the following to your .env file:
DB_URL=jdbc:postgresql://localhost:5432/todo
DB_USER=todo
DB_PASS=todoNow, we can reimplement TodoRepository to use the todo table in the DB:
import klite.jdbc.*
class TodoRepository(private val db: DataSource) {
fun list() = db.select<Todo>("todos")
fun save(todo: Todo) = db.upsert("todos", todo.toValues())
fun get(id: Id<Todo>) = db.select<Todo>(Todo::id to id).first()
}See klite-jdbc docs for more examples of how to use it.
If you ran the previous code, you probably have realized that the todo table does not exist in the database yet.
Let's use DBMigrator to create the table for us:
use<DBModule>()
use<DBMigrator>()DBMigrator will look for db.sql file in the root of the classpath (resources) and run it against the database, let's create it:
--changeset todos
create table todos (
id bigint primary key,
item text not null,
completed_at timestamptz
);
--changeset todos:initial-data context:dev
insert into todos (id, item) values (123, 'Buy groceries');Later, you can add more changesets to the same file, and DBMigrator will run only new ones, or even extract them to separate files and use --include file.sql. See ChangeSet for more attributes.
To make it even easier to implement repositories, klite-jdbc includes BaseCrudRepository:
data class Todo(...): BaseEntity<Id<Todo>>
class TodoRepository(db: DataSource): BaseCrudRepository<Todo, Id<Todo>>(db, "todos")And then you will have the common list(), get(), save() methods already implemented for you.
By default, JDBC Connections work in autoCommit mode, which means that every statement is a separate transaction.
In real life it makes sense to use transaction per request model, which will issue a rollback automatically in case anything fails with an exception.
Add use<RequestTransactionHandler>(), which will do it for you.
Then, @NoTransaction can be used to disable transaction for a specific route.
Klite makes it easy to handle errors without extra code by default.
You can register custom exception types to produce specific HTTP error responses:
errors.on<IllegalAccessException>(StatusCode.Forbidden)In fact, many Klite modules add their own error handlers, e.g. require() and error() will produce BadRequest 400 responses.
Data validation is also already done by default, e.g. Kotlin nullability is respected and reported to the user automatically.
The most convenient for additional validation is to use correct data types (e.g. Email, Phone, URL, not just String).
With Klite, it's easy to add custom value types, that would work across request parameters, json and database columns. In fact, Klite already provides Email, Phone, Password types.
Any type with String constructor or String static factory method will be supported by default.
class EstonianPersonalCode(value: String): StringValue(value) {
init {
require(value.length == 11) { "EE personal code should be 11 characters long" }
}
}Custom type creation can be registered with Converter:
Converter.use { Locale.forLanguageTag(it.replace('_', '-')) }If value type are not enough, then add require calls to entity constructor:
data class Todo(...) {
init {
require(item[0].isUpperCase()) { "Item should start with upper case" }
}
}Then no extra validation code is needed in route handlers, that is easy to forget.
JsonMapper.trimToNull is enabled by default, so you get more clean data into your objects, and won't get empty strings into required (non-null) fields.
To enable sessions support, pass a SessionStore implementation to Server's constructor:
Server(sessionStore = CookieSessionStore())CookieSessionStore will store session data in a cookie, which survives server restarts and works with multiple server instances. Session data is encrypted and cannot be tamprered with, but can be subject to replay attacks.
Store as few data as possible in the session, and use it mostly for user identification and authorization. Any state-related data in session will be problematic with the browser's back button.
Then you can use session object to store and retrieve session data, e.g:
@POST fun login(credentials: Credentials) {
val user = userRepository.by(credentials) ?: throw UnauthorizedException()
session["userId"] = user.id.toString()
}and then have a before handler to load current user:
context("/api") {
...
before {
val userId = it.session["userId"] ?: throw ForbiddenException()
it.attr("user", userRepository.get(userId))
}
}attr() values can be accessed in annotated routes with @AttrParam annotation.
before, after, decorate handlers can be used to add common logic to all routes in a context.
They affect only the routes that follow them.
If you want to render HTML server-side, you can implement your own BodyRenderer using your favorite template engine or just use Kotlin template strings for HTML generation.
context("/html") {
get("/:name") {
send(OK, """
<h1>Hello, ${+path("name")}!</h1>
""", MimeTypes.html)
}
}You can now get the json response using http://localhost:8080/html/klite.
Note the + before path("name"). This is a special operator to escape HTML characters in the string, import it from klite.html.unaryPlus.
Instead, modern applications would use a frontend framework, such as Svelte or React, and serve the frontend as a static site.
Klite has a built-in static file server, which you can use to serve your frontend files.
assets("/", AssetsHandler(Path.of("public"), useIndexForUnknownPaths = true))This adds its own context, it is a good idea to use / as the path, so that files are available from shorter URLs.
The latter parameter is useful for SPA (single-page applications), where the frontend router handles the URL.