Configuring the Server

预计阅读时间: 19 分钟

Ktor uses HOCON (Human-Optimized Config Object Notation) format in the external configuration file. In this file, you can configure things like the port to listen to, or the modules to be loaded. This format is similar to JSON, but is optimized to be read and written by humans, and supports additional features like environment variable substitution. In this case, you configure the server engine to be used with the mainClassName pointing to a particular EngineMain.

Ktor also uses a set of lambdas with a typed DSL (Domain Specific Language) to configure the application and server engine when using embeddedServer.

Starting with Ktor 1.0.0-beta-2, the DevelopmentEngine class has been renamed to EngineMain, for older versions just rename it.

Table of contents:

The HOCON file

This is the preferred way of configuring Ktor applications as it allows you to easily change the configuration without recompiling your application.

When Ktor is started using a EngineMain, or by calling the commandLineEnvironment, it tries to load a HOCON file called application.conf from the application resources. You can change the location of the file using command line arguments.

Available development engines that you can use as mainClassName:

  • io.ktor.server.cio.EngineMain
  • io.ktor.server.tomcat.EngineMain
  • io.ktor.server.jetty.EngineMain
  • io.ktor.server.netty.EngineMain

Ktor only requires you to specify which module or modules you want it to load when starting the server using the ktor.application.modules property. All the other properties are optional.

A typical, simple HOCON file for Ktor (application.conf) would look like this:

ktor {
    deployment {
        port = 8080
    }

    application {
        modules = [ io.ktor.samples.metrics.MetricsApplicationKt.main ]
    }
}

Using dot notation it would be equivalent to:

ktor.deployment.port = 8080
ktor.application.modules = [ io.ktor.samples.metrics.MetricsApplicationKt.main ]

Ktor allows you to configure much more: from additional core configurations to Ktor features, and even custom configurations for your applications:

ktor {
    deployment {
        environment = development
        port = 8080
        sslPort = 8443
        autoreload = true
        watch = [ httpbin ]
    }

    application {
        modules = [ io.ktor.samples.httpbin.HttpBinApplicationKt.main ]
    }

    security {
        ssl {
            keyStore = build/temporary.jks
            keyAlias = mykey
            keyStorePassword = changeit
            privateKeyPassword = changeit
        }
    }
}

jwt {
    domain = "https://jwt-provider-domain/"
    audience = "jwt-audience"
    realm = "ktor sample app"
}

youkube {
  session {
    cookie {
      key = 03e156f6058a13813816065
    }
  }
  upload {
    dir = ktor-samples/ktor-samples-youkube/.video
  }
}

There is a list of the available core configurations in this document.

You can use HOCON to set properties from environment variables.

There is an IntelliJ plugin for HOCON, that you may want to install.

Command Line

When using commandLineEnvironment (any EngineMain main) there are several switches and configuration parameters you can use to configure your application module.

If you start the application from the command line with -config=anotherfile.conf, it will load the config file from the specific local file instead of from the resources.

Using switches, you can, for example, override the bound port defined by executing:

java -jar myapp-fatjar.jar -port=8080

There is a list of available command line switches in this document.

Configuring the embeddedServer

embeddedServer is a simple way to start a Ktor application. You provide your own main function, and being more explicit, it is easier to understand what happens exactly.

embeddedServer includes an optional parameter configure that allows you to set the configuration for the engine specified in the first parameter. Independent to the engine used, you will have some available properties to configure:

embeddedServer(AnyEngine, configure = {
    // Size of the event group for accepting connections
    connectionGroupSize = parallelism / 2 + 1
    // Size of the event group for processing connections,
    // parsing messages and doing engine's internal work 
    workerGroupSize = parallelism / 2 + 1
    // Size of the event group for running application code 
    callGroupSize = parallelism 
}) {
    // ...
}.start(true)

Multiple connectors

It is possible to define by code several connectors using the applicationEngineEnvironment.

Inside the applicationEngineEnvironment, you can define HTTP and HTTPS connectors:

To define a HTTP connector:

connector {
    host = "0.0.0.0"
    port = 9090
}

To define a HTTPS connector:

sslConnector(keyStore = keyStore, keyAlias = "mykey", keyStorePassword = { "changeit".toCharArray() }, privateKeyPassword = { "changeit".toCharArray() }) {
    port = 9091
    keyStorePath = keyStoreFile.absoluteFile
}

An actual example:

fun main(args: Array<String>) {
    val env = applicationEngineEnvironment {
        module {
            main()
        }
        // Private API
        connector {
            host = "127.0.0.1"
            port = 9090
        }
        // Public API
        connector {
            host = "0.0.0.0"
            port = 8080
        }
    }
    embeddedServer(Netty, env).start(true)
}

The application will handle all the connections. You have access to the local port for each ApplicationCall, so you can decide what to do based on the local port:

fun Application.main() {
    routing {
        get("/") {
            if (call.request.local.port == 8080) {
                call.respondText("Connected to public api")
            } else {
                call.respondText("Connected to private api")
            }
        }
    }
}

You can see a complete example of this in ktor-samples/multiple-connectors.

Netty

When using Netty as the engine, in addition to common properties, you can configure some other properties:

embeddedServer(Netty, configure = {
    // Size of the queue to store [ApplicationCall] instances that cannot be immediately processed
    requestQueueLimit = 16 
    // Do not create separate call event group and reuse worker group for processing calls
    shareWorkGroup = false 
    // User-provided function to configure Netty's [ServerBootstrap]
    configureBootstrap = {
        // ...
    } 
    // Timeout in seconds for sending responses to client
    responseWriteTimeoutSeconds = 10 
}) {
    // ...
}.start(true)

Jetty

When using Jetty as the engine, in addition to common properties, you can configure the Jetty server.

embeddedServer(Jetty, configure = {
    // Property to provide a lambda that will be called during Jetty
    // server initialization with the server instance as an argument.
    configureServer = {
        // ...
    } 
}) {
    // ...
}.start(true)

CIO

When using CIO (Coroutine I/O) as the engine, in addition to common properties, you can configure the connectionIdleTimeoutSeconds property.

embeddedServer(CIO, configure = {
    // Number of seconds that the server will keep HTTP IDLE connections open.
    // A connection is IDLE if there are no active requests running.
    connectionIdleTimeoutSeconds = 45
}) {
    // ...
}.start(true)

Tomcat

When using Tomcat, in addition to common properties, you can configure the Tomcat server.

embeddedServer(Tomcat, configure = {
    // Property to provide a lambda that will be called during Tomcat
    // server initialization with the server instance as argument.
    configureTomcat { // this: Tomcat ->
        // ...
    }
}) {
    // ...
}.start(true)

Those are the official engines developed for Ktor, but it is also possible to create your own engines and provide custom configurations for them.

Available configuration parameters

There is a list of properties which Ktor understands out of the box and that you can pass from the command line or the HOCON file.

Switch refers to command line arguments that you pass to the application, so you can, for example, change the bound port by:

java -jar myapp-fatjar.jar -port=8080

Parameter paths are paths inside the application.conf file:

ktor.deployment.port = 8080
ktor {
    deployment {
        port = 8080
    }
}

General switches and parameters:

Switch Parameter path Default Description
-jar=     Path to JAR file
-config=     Path to config file (instead of application.conf from resources)
-host= ktor.deployment.host 0.0.0.0 Bound host
-port= ktor.deployment.port 80 Bound port
-watch= ktor.deployment.watch [] Package paths to watch for reloading
  ktor.application.id Application Application Identifier used for logging
  ktor.deployment.rootPath / Servlet context path
  ktor.deployment.callGroupSize parallelism Event group size running application code
  ktor.deployment.connectionGroupSize parallelism / 2 + 1 Event group size accepting connections
  ktor.deployment.workerGroupSize parallelism / 2 + 1 Event group size for processing connections, parsing messages and doing engine’s internal work
  ktor.deployment.shutdown.url   URL for shutdown the application when defined. Internally uses the ShutDownUrl feature

Required when SSL port is defined:

Switch Parameter path Default Description
-sslPort= ktor.deployment.sslPort null SSL port
-sslKeyStore= ktor.security.ssl.keyStore null SSL key store
  ktor.security.ssl.keyAlias mykey Alias for the SSL key store
  ktor.security.ssl.keyStorePassword null Password for the SSL key store
  ktor.security.ssl.privateKeyPassword null Password for the SSL private key

You can use -P: to specify parameters that don’t have a specific switch. For example: -P:ktor.deployment.callGroupSize=7.

Reading the configuration from code

If you are using a EngineMain instead of an embeddedServer, the HOCON file is loaded, and you are able to access its configuration properties.

You can also define arbitrary property paths to configure your application.

val port: String = application.environment.config
    .propertyOrNull("ktor.deployment.port")?.getString()
    ?: "80"

It is possible to access the HOCON application.conf configuration too, by using a custom main with commandLineEnvironment:

embeddedServer(Netty, commandLineEnvironment(args + arrayOf("-port=8080"))).start(true)

Or by redirecting it to the specific EngineMain.main:

val moduleName = Application::module.javaMethod!!.let { "${it.declaringClass.name}.${it.name}" }
io.ktor.server.netty.main(args + arrayOf("-port=8080", "-PL:ktor.application.modules=$moduleName"))

Or with a custom applicationEngineEnvironment:

embeddedServer(Netty, applicationEngineEnvironment {
    log = LoggerFactory.getLogger("ktor.application")
    config = HoconApplicationConfig(ConfigFactory.load()) // Provide a Hocon config file

    module {
        routing {
            get("/") {
                call.respondText("HELLO")
            }
        }
    }

    connector {
        port = 8080
        host = "127.0.0.1"
    }
}).start(true)

You can also access the configuration properties by manually loading the default config file application.conf:

val config = HoconApplicationConfig(ConfigFactory.load())

Using environment variables

For HOCON, if you want to configure some parameters using environment variables, you can use environment substitution using ${ENV} syntax. For example:

ktor {
    deployment {
        port = ${PORT}
    }
}

This will look for a PORT environment variable, and if not found an exception will be thrown:

Exception in thread "main" com.typesafe.config.ConfigException$UnresolvedSubstitution: application.conf @ file:/path/to/application.conf: 3: Could not resolve substitution to a value: ${PORT}

In case you want to provide a default value for a property because the environment doesn’t exist, you can set the property with the default value, and then set it again with the ${?ENV} syntax:

ktor {
    deployment {
        port = 8080
        port = ${?PORT}
    }
}

If you are using embeddedServer you can still use System.getenv from Java. For example:

val port = System.getenv("PORT")?.toInt() ?: 8080

Custom configuration systems

Ktor provides an interface that you can implement the configuration in, available at application.environment.config. You can construct and set the configuration properties inside an applicationEngineEnvironment.

interface ApplicationConfig {
    fun property(path: String): ApplicationConfigValue
    fun propertyOrNull(path: String): ApplicationConfigValue?
    fun config(path: String): ApplicationConfig
    fun configList(path: String): List<ApplicationConfig>
}

interface ApplicationConfigValue {
    fun getString(): String
    fun getList(): List<String>
}

class ApplicationConfigurationException(message: String) : Exception(message)

Ktor provides two implementations. One based on a map (MapApplicationConfig), and other based on HOCON (HoconApplicationConfig).

You can create and compose config implementations and set them at applicationEngineEnvironment, so it is available to all the application components.