HTTP Client Engines

Estimated reading time: 4 minutes

Ktor HTTP Client has a common interface for performing HTTP Requests, but allows to specify an engine that does the internal job. Different engines has different configurations, dependencies an supporting features.

Table of contents:

Default engine

By calling to the HttpClient method without specifying an engine, it uses a default engine.

val client = HttpClient()

In the case of the JVM, the default engine is resolved with a ServiceLoader, getting the first one available. Thus depends on the artifacts you have included.

For native, it uses the predefined one.

Configuring engines

Ktor HttpClient lets you configure the parameters of each engine by calling Engine.config { }, but since 0.9.4, the preferred way is to use HttpClient(MyHttpEngine) { engine { ... } } instead.

Every engine config has two common properties that can be set:

  • The dispatcher property is the CoroutineDispatcher used when processing client requests.
  • The sslContext is a allowing you to set custom keys, a trust manager or custom source for secure random data.
val client = HttpClient(MyHttpEngine) {
    engine {
        sslContext = SSLContext.getDefault()

You can also adjust maximum total connections and maximum connections per route in Apache and CIO clients (but not Jetty).



Apache is the most configurable HTTP client about right now. It supports HTTP/1.1 and HTTP/2. It is the only one that supports following redirects and allows you to configure timeouts, proxies among other things it is supported by org.apache.httpcomponents:httpasyncclient.

A sample configuration would look like:

val client = HttpClient(Apache) {
    engine {
        followRedirects = true  // Follow HTTP Location redirects - default false. It uses the default number of redirects defined by Apache's HttpClient that is 50.

        // For timeouts: 0 means infinite, while negative value mean to use the system's default value
        socketTimeout = 10_000  // Max time between TCP packets - default 10 seconds
        connectTimeout = 10_000 // Max time to establish an HTTP connection - default 10 seconds
        connectionRequestTimeout = 20_000 // Max time for the connection manager to start a request - 20 seconds

        customizeClient {
            // Apache's HttpAsyncClientBuilder
            setProxy(HttpHost("", 8080))
            setMaxConnTotal(1000) // Maximum number of socket connections.
            setMaxConnPerRoute(100) // Maximum number of requests for a specific endpoint route.
        customizeRequest {
            // Apache's RequestConfig.Builder
This engine is defined in the class io.ktor.client.engine.apache.Apache in the artifact org.apache.httpcomponents:httpasyncclient.
dependencies { compile "org.apache.httpcomponents:httpasyncclient" }
<project> ... <dependencies> <dependency> <groupId>org.apache.httpcomponents</groupId> <artifactId>httpasyncclient</artifactId> <version>${ktor.version}</version> <scope>compile</scope> </dependency> </dependencies> </project>


CIO (Coroutine-based I/O) is a Ktor implementation with no additional dependencies and is fully asynchronous. It only supports HTTP/1.x for now.

CIO provides maxConnectionsCount and a endpointConfig for configuring.

A sample configuration would look like:

val client = HttpClient(CIO) {
    engine {
        maxConnectionsCount = 1000 // Maximum number of socket connections.
        endpoint.apply {
            maxConnectionsPerRoute = 100 // Maximum number of requests for a specific endpoint route.
            pipelineMaxSize = 20 // Max number of opened endpoints.
            keepAliveTime = 5000 // Max number of milliseconds to keep each connection alive.
            connectTimeout = 5000 // Number of milliseconds to wait trying to connect to the server.
            connectRetryAttempts = 5 // Maximum number of attempts for retrying a connection.
This engine is defined in the class io.ktor.client.engine.cio.CIO in the artifact io.ktor:ktor-client-cio:$ktor_version.
dependencies { compile "io.ktor:ktor-client-cio:$ktor_version" }
<project> ... <dependencies> <dependency> <groupId>io.ktor</groupId> <artifactId>ktor-client-cio</artifactId> <version>${ktor.version}</version> <scope>compile</scope> </dependency> </dependencies> </project>


Jetty provides an additional sslContextFactory for configuring. It only supports HTTP/2 for now.

A sample configuration would look like:

val client = HttpClient(Jetty) {
    engine {
        sslContextFactory = SslContextFactory()
This engine is defined in the class io.ktor.client.engine.jetty.Jetty in the artifact io.ktor:ktor-client-jetty:$ktor_version. It includes the org.eclipse.jetty.http2:http2-client transitive dependency.
dependencies { compile "io.ktor:ktor-client-jetty:$ktor_version" }
<project> ... <dependencies> <dependency> <groupId>io.ktor</groupId> <artifactId>ktor-client-jetty</artifactId> <version>${ktor.version}</version> <scope>compile</scope> </dependency> </dependencies> </project>

JVM and Android


Since Ktor 0.9.4, there is a engine based on OkHttp.

val client = HttpClient(OkHttp) {
    engine {
        config { // this: OkHttpClient.Builder ->
            // ...
            // ...

This engine is defined in the class io.ktor.client.engine.okhttp.OkHttp in the artifact io.ktor:ktor-client-okhttp:$ktor_version. It includes the com.squareup.okhttp3:okhttp transitive dependency.
dependencies { compile "io.ktor:ktor-client-okhttp:$ktor_version" }
<project> ... <dependencies> <dependency> <groupId>io.ktor</groupId> <artifactId>ktor-client-okhttp</artifactId> <version>${ktor.version}</version> <scope>compile</scope> </dependency> </dependencies> </project>


The Android engine, doesn’t have additional dependencies, and uses a ThreadPool with a normal HttpURLConnection, to perform the requests. And can be configured like this:

val client = HttpClient(Android) {
    engine {
        connectTimeout = 100_000
        socketTimeout = 100_000
This engine is defined in the class in the artifact io.ktor:ktor-client-android:$ktor_version.
dependencies { compile "io.ktor:ktor-client-android:$ktor_version" }
<project> ... <dependencies> <dependency> <groupId>io.ktor</groupId> <artifactId>ktor-client-android</artifactId> <version>${ktor.version}</version> <scope>compile</scope> </dependency> </dependencies> </project>


The iOS engine, uses the asynchronous NSURLSession internally. And have no additional configuration.

val client = HttpClient(Ios) {
This engine is defined in the class io.ktor.client.engine.ios.Ios in the artifact io.ktor:ktor-client-ios:$ktor_version.
dependencies { compile "io.ktor:ktor-client-ios:$ktor_version" }
<project> ... <dependencies> <dependency> <groupId>io.ktor</groupId> <artifactId>ktor-client-ios</artifactId> <version>${ktor.version}</version> <scope>compile</scope> </dependency> </dependencies> </project>



There is a engine specific for testing described in its own page: MockEngine for testing.