Skip to content

Latest commit

 

History

History
137 lines (98 loc) · 6.24 KB

README.md

File metadata and controls

137 lines (98 loc) · 6.24 KB

Klite

Release Build & Test

Klite: a very light-weight (lite) non-blocking http framework for Kotlin coroutines on JVM. Probably the most sustainable JVM web framework (low resource usage and CO₂ emissions).

Inspired by SparkJava, Jooby, etc, but smaller, simpler and better.

Includes very light json and jdbc & migrations submodules for Kotlin, which can be used independently.

Please star the repo to let us know you are interested.

See the Tutorial to grasp the basics quickly.

Goals

  • Minimal amount of code
    • Simple to maintain & change
    • Performance is also important, but simplicity is preferred
  • Zero dependencies - Java built-in jdk.httpserver is used under the hood
    • Perfect for microservices
    • But still possible to easily add support for other servers if needed
    • Sample docker image is about 50-70Mb thanks to jlink, depending on used modules
    • Production applications can run with as low as 50Mb of heap memory, suitable for very cheap plans at Fly.io or Heroku
  • 12-factor apps by default
  • Most behaviour can be overridden if necessary
  • Both route builder and annotated classes
  • Very easy to do simple things, e.g. @GET suspend fun route() = provider.fetchData()
  • Proper Kotlin coroutine support with working before/after filters for e.g. transactions and logging
  • Shared type-safe value classes across http parameters, database columns and json fields
  • Most app code will not depend on the framework, easy to switch
  • Not much need for documentation - the source code is short and readable.

Dependencies

  • Java 6+ built-in non-blocking jdk.httpserver
  • Re-routable Java 9+ System.Logger

Modules

  • core - some reusable classes, e.g. Config. Don't depend on it directly.
  • server - the main server module. See it's documentation. Zero external dependencies.
  • json - lightweight and easily configurable json parsing/rendering (usable standalone)
  • csv - simple CSV parsing/generation (usable standalone)
  • i18n - simple server-side translations (for emails, etc)
  • jdbc - provides JDBC extensions for DB access, transaction handling and migrations (usable standalone)
  • jdbc-test - provides a way of testing your DB code using a real DB
  • jobs - provides a simple scheduled JobRunner
  • oauth - implements OAuth 2.0 login with several providers
  • openapi - generates OpenAPI 3.0 spec for all routes in a context, viewable with Swagger UI

Integrations

These integrate with external libraries. All of this functionality is available in Klite's own modules.

  • slf4j - redirects server logs to slf4j and configures it
  • jackson - json parsing/rendering using Jackson
  • serialization - json parsing/rendering using kotlinx-serialization
  • liquibase - allows to use Liquibase for DB migration

Status

The main server module is only ~1000 lines of code.

Klite powers a number of known production apps already. Publicly announced at KKON 2022, see the slides.

Performance

Klite (including jdk.httpserver) has sub-1ms overhead per request after warmup. Can be verified with Apache Benchmark after the sample project is launched:

This simple route produces ~19000 rps, with 99% taking less than 1ms:
ab -n 10000 -c 10 http://localhost:8080/api/hello

JDBC access route produces ~11000 rps, with 99% taking less than 1ms:
ab -n 10000 -c 10 http://localhost:8080/api/hello/user/9725b054-426b-11ee-92a5-0bd2a151eea2

Coroutine suspension test with 1000 concurrent requests, ~7000 rps, 80% of requests complete within the specified delay of 100ms:
ab -n 10000 -c 1000 http://localhost:8080/api/hello/suspend

Tests ran on Ubuntu, Java 20, i9-9900T CPU from 2019.

Usage

See the sample project on how to build apps with Klite and run them in Docker.

Klite builds are available from jitpack, see also changelog

  repositories {
    mavenCentral()
    maven { url = uri("https://jitpack.io") }
  }

  dependencies {
    val kliteVersion = "master-SNAPSHOT" // you can put a released tag or commit hash here
    implementation("com.github.codeborne.klite:klite-server:$kliteVersion")
    // Plus any optional components with their own external dependencies, see above for list
    implementation("com.github.codeborne.klite:klite-jackson:$kliteVersion")
    implementation("com.github.codeborne.klite:klite-jdbc:$kliteVersion")
    testImplementation("com.github.codeborne.klite:klite-jdbc-test:$kliteVersion")
    ...
  }

Using unreleased commits

Jitpack builds requested versions on the fly, so it is also good if you want to fork Klite and customize for your own needs - you will still be able to add your fork as a Maven/Gradle dependency in your apps.

But pull-requests are welcome if you want to improve something for everybody!

Depending on a local build

Publish to ~/.m2/repository by running ./gradlew publishToMavenLocal

Then add mavenLocal() repository to your project and use Klite version of master-SNAPSHOT.

Depenending on the Git repository (source) directly

If there is a problem with Jitpack, then it's possible to add the following to your settings.gradle.kts:

sourceControl {
  gitRepository(java.net.URI("https://github.com/codeborne/klite.git")) {
    producesModule("com.github.codeborne.klite:server")
    producesModule("com.github.codeborne.klite:jdbc")
    // list all subprojects you depend on in build.gradle.kts, use their unprefixed names, e.g. server, not klite-server
  }
}

Gradle will clone and build Klite for you automatically during your project's build. Tagged version numbers are supported this way, but not commit hashes.