documentation/manual/working/scalaGuide/main/ws/ScalaWS.md
Sometimes we would like to call other HTTP services from within a Play application. Play supports this via its WS ("WebService") library, which provides a way to make asynchronous HTTP calls through a WSClient instance.
There are two important parts to using the WSClient: making a request, and processing the response. We'll discuss how to make both GET and POST HTTP requests first, and then show how to process the response from WSClient. Finally, we'll discuss some common use cases.
Note: In Play 2.6, Play WS has been split into two, with an underlying standalone client that does not depend on Play, and a wrapper on top that uses Play specific classes. In addition, shaded versions of AsyncHttpClient and Netty are now used in Play WS to minimize library conflicts, primarily so that Play's HTTP engine can use a different version of Netty. Please see the [[2.6 migration guide|WSMigration26]] for more information.
To use WSClient, first add ws to your build.sbt file:
libraryDependencies += ws
Play WS supports HTTP caching, but requires a JSR-107 cache implementation to enable this feature. You can add ehcache:
libraryDependencies += ehcache
Or you can use another JSR-107 compatible cache such as Caffeine.
Once you have the library dependencies, then enable the HTTP cache as shown on [[WS Cache Configuration|WsCache]] page.
Using an HTTP cache means savings on repeated requests to backend REST services, and is especially useful when combined with resiliency features such as stale-if-error and stale-while-revalidate.
Now any component that wants to use WS will have to declare a dependency on the WSClient:
We've called the WSClient instance ws, all the following examples will assume this name.
To build an HTTP request, you start with ws.url() to specify the URL.
This returns a WSRequest that you can use to specify various HTTP options, such as setting headers. You can chain calls together to construct complex requests.
You end by calling a method corresponding to the HTTP method you want to use. This ends the chain, and uses all the options defined on the built request in the WSRequest.
This returns a Future[WSResponse] where the Response contains the data returned from the server.
If you are doing any blocking work, including any kind of DNS work such as calling
java.util.URL.equals(), then you should use a custom execution context as described in [[ThreadPools]], preferably through aCustomExecutionContext. You should size the pool to leave a safety margin large enough to account for failures.
If you are calling out to an unreliable network, consider using
Futures.timeoutand a circuit breaker like Failsafe.
If you need to use HTTP authentication, you can specify it in the builder, using a username, password, and an AuthScheme. Valid case objects for the AuthScheme are BASIC, DIGEST, KERBEROS, NTLM, and SPNEGO.
If an HTTP call results in a 302 or a 301 redirect, you can automatically follow the redirect without having to make another call.
Parameters can be specified as a series of key/value tuples. Use addQueryStringParameters to add parameters, and withQueryStringParameters to overwrite all query string parameters.
Headers can be specified as a series of key/value tuples. Use addHttpHeaders to append additional headers, and withHttpHeaders to overwrite all headers.
If you are sending plain text in a particular format, you may want to define the content type explicitly.
Cookies can be added to the request by using DefaultWSCookie or by passing through play.api.mvc.Cookie. Use addCookies to append cookies, and withCookies to overwrite all cookies.
A virtual host can be specified as a string.
If you wish to specify a request timeout, you can use withRequestTimeout to set a value. An infinite timeout can be set by passing Duration.Inf.
To post url-form-encoded data a Map[String, Seq[String]] needs to be passed into post.
If the body is empty, you must pass play.api.libs.ws.EmptyBody into the post method.
To post multipart-form-encoded data a Source[play.api.mvc.MultipartFormData.Part[Source[ByteString, Any]], Any] needs to be passed into post.
To upload a file you need to pass a play.api.mvc.MultipartFormData.FilePart[Source[ByteString, Any]] to the Source:
The easiest way to post JSON data is to use the [[JSON|ScalaJson]] library.
The easiest way to post XML data is to use XML literals. XML literals are convenient, but not very fast. For efficiency, consider using an XML view template, or a JAXB library.
It's also possible to stream data in the request body using Pekko Streams.
For example, imagine you have executed a database query that is returning a large image, and you would like to forward that data to a different endpoint for further processing. Ideally, if you can send the data as you receive it from the database, you will reduce latency and also avoid problems resulting from loading in memory a large set of data. If your database access library supports Reactive Streams (for instance, Slick does), here is an example showing how you could implement the described behavior:
The largeImageFromDB in the code snippet above is a Source[ByteString, _].
You can do additional processing on a WSRequest by adding a request filter. A request filter is added by extending the play.api.libs.ws.WSRequestFilter trait, and then adding it to the request with request.withRequestFilter(filter).
A sample request filter that logs the request in cURL format to SLF4J has been added in play.api.libs.ws.ahc.AhcCurlRequestLogger.
will output:
curl \
--verbose \
--request PUT \
--header 'Content-Type: application/x-www-form-urlencoded; charset=utf-8' \
--data 'key=value' \
http://localhost:19001/
Working with the Response is easily done by mapping inside the Future.
The examples given below have some common dependencies that will be shown once here for brevity.
Whenever an operation is done on a Future, an implicit execution context must be available - this declares which thread pool the callback to the future should run in. You can inject the default Play execution context in your DI-ed class by declaring an additional dependency to ExecutionContext in the class' constructor:
The examples also use the following case class for serialization/deserialization:
The WSResponse extends play.api.libs.ws.WSBodyReadables trait, which contains type classes for Play JSON and Scala XML conversion. You can also create your own custom type classes if you would like to convert the response to your own types, or use a different JSON or XML encoding.
You can process the response as a JSON object by calling response.json.
The JSON library has a [[useful feature|ScalaJsonCombinators]] that will map an implicit Reads[T] directly to a class:
@scalaws-process-json-with-implicit
You can process the response as an XML literal by calling response.xml.
Calling get(), post() or execute() will cause the body of the response to be loaded into memory before the response is made available. When you are downloading a large, multi-gigabyte file, this may result in unwelcome garbage collection or even out of memory errors.
WS lets you consume the response's body incrementally by using an Pekko Streams Sink. The stream() method on WSRequest returns a streaming WSResponse which contains a bodyAsSource method that returns a Source[ByteString, _]
Note: In 2.5.x, a
StreamedResponsewas returned in response to arequest.stream()call. In 2.6.x, a standardWSResponseis returned, and thebodyAsSource()method should be used to return the Source.
Here is a trivial example that uses a folding Sink to count the number of bytes returned by the response:
Alternatively, you could also stream the body out to another location. For example, a file:
Another common destination for response bodies is to stream them back from a controller's Action:
As you may have noticed, before calling stream() we need to set the HTTP method to use by calling withMethod on the request. Here follows another example that uses PUT instead of GET:
Of course, you can use any other valid HTTP verb.
Using for comprehensions is a good way to chain WSClient calls in a trusted environment. You should use for comprehensions together with Future.recover to handle possible failure.
When making a request from a controller, you can map the response to a Future[Result]. This can be used in combination with Play's Action.async action builder, as described in [[Handling Asynchronous Results|ScalaAsync]].
If a chain of WS calls does not complete in time, it may be useful to wrap the result in a timeout block, which will return a failed Future if the chain does not complete in time -- this is more generic than using withRequestTimeout, which only applies to a single request. The best way to do this is with Play's [[non-blocking timeout feature|ScalaAsync]], using play.api.libs.concurrent.Futures:
If you are using compile time dependency injection, you can access a WSClient instance by using the trait AhcWSComponents in your [[application's components|ScalaCompileTimeDependencyInjection#Using-other-components]].
We recommend that you get your WSClient instances using dependency injection as described above. WSClient instances created through dependency injection are simpler to use because they are automatically created when the application starts and cleaned up when the application stops.
However, if you choose, you can instantiate a WSClient directly from code and use this for making requests or for configuring underlying AsyncHttpClient options.
If you create a WSClient manually then you must call
client.close()to clean it up when you've finished with it. Each client creates its own thread pool. If you fail to close the client or if you create too many clients then you will run out of threads or file handles -— you'll get errors like "Unable to create new native thread" or "too many open files" as the underlying resources are consumed.
You need an instance of an pekko.stream.Materializer to create a play.api.libs.ws.ahc.AhcWSClient instance directly. Usually you'll inject this into the service using dependency injection:
Creating a client directly means that you can also change configuration at the AsyncHttpClient and Netty configuration layers as well:
You can also use play.api.test.WsTestClient.withTestClient to create an instance of WSClient in a functional test. See [[ScalaTestingWebServiceClients]] for more details.
Or, you can run the WSClient completely standalone without involving a running Play application at all:
This can be useful in cases where there is a specific HTTP client option that isn't accessible from config.
Again, once you are done with your custom client work, you must close the client:
Ideally, you should close a client after you know all requests have been completed. Be careful of using an automatic resource management pattern to close the client, because WSClient logic is asynchronous and many ARM solutions may be designed for a single threaded synchronous solution.
If you want to call WS outside of the context of Play altogether, you can use the standalone version of Play WS, which does not depend on any Play libraries. You can do this by adding play-ahc-ws-standalone to your project:
libraryDependencies += "org.playframework" %% "play-ahc-ws-standalone" % playWSStandalone
Please see https://github.com/playframework/play-ws and the [[2.6 migration guide|WSMigration26]] for more information.
Play WS comes with rich type support for bodies in the form of play.api.libs.ws.WSBodyWritables, which contains type classes for converting input such as JsValue or XML in the body of a WSRequest into a ByteString or Source[ByteString, _], and play.api.libs.ws.WSBodyReadables, which aggregates type classes that read the body of a WSResponse from a ByteString or Source[ByteString, _] and return the appropriate type, such as JsValue or XML. These type classes are automatically in scope when you import the ws package, but you can also create custom types. This is especially useful if you want to use a custom library, i.e. you would like to stream XML through STaX API or use another JSON library such as Argonaut or Circe.
You can create a custom readable by accessing the response body:
You can create a custom body writable to a request as follows, using an BodyWritable and an InMemoryBody. To specify a custom body writable with streaming, use a SourceBody.
You can get access to the underlying AsyncHttpClient from a WSClient.
Use the following properties in application.conf to configure the WSClient:
play.ws.followRedirects: Configures the client to follow 301 and 302 redirects (default is true).play.ws.useProxyProperties: To use the JVM system's HTTP proxy settings (http.proxyHost, http.proxyPort) (default is true).play.ws.useragent: To configure the User-Agent header field.play.ws.compressionEnabled: Set it to true to use gzip/deflater encoding (default is false).To configure WS for use with HTTP over SSL/TLS (HTTPS), please see [[Configuring WS SSL|WsSSL]].
To configure WS for use with HTTP caching, please see [[Configuring WS Cache|WsCache]].
There are 3 different timeouts in WSClient. Reaching a timeout causes the WSClient request to interrupt.
play.ws.timeout.connection: The maximum time to wait when connecting to the remote host (default is 120 seconds).play.ws.timeout.idle: The maximum time the request can stay idle (connection is established but waiting for more data) (default is 120 seconds).play.ws.timeout.request: The total time you accept a request to take (it will be interrupted even if the remote host is still sending data) (default is 120 seconds).The request timeout can be overridden for a specific connection with withRequestTimeout() (see "Making a Request" section).
The following advanced settings can be configured on the underlying AsyncHttpClientConfig.
Please refer to the AsyncHttpClientConfig Documentation for more information.
play.ws.ahc.keepAliveplay.ws.ahc.maxConnectionsPerHostplay.ws.ahc.maxConnectionsTotalplay.ws.ahc.maxConnectionLifetimeplay.ws.ahc.idleConnectionInPoolTimeoutplay.ws.ahc.maxNumberOfRedirectsplay.ws.ahc.maxRequestRetryplay.ws.ahc.disableUrlEncoding