Asynchronous HTTP Programming with Play Framework

1. Overview

Often our web services need to use other web services in order to do their job. It can be difficult to serve user requests while keeping a low response time. A slow external service can increase our response time and cause our system to pile up requests, using more resources. This is where a non-blocking approach can be very helpful

In this tutorial, we’ll fire multiple asynchronous requests to a service from a Play Framework application. By leveraging Java’s non-blocking HTTP capability, we’ll be able to smoothly query external resources without affecting our own main logic.

In our example we’ll explore the Play WebService Library.

2. The Play WebService (WS) Library

WS is a powerful library providing asynchronous HTTP calls using Java Action.

Using this library, our code sends these requests and carries on without blocking. To process the result of the request, we provide a consuming function, that is, an implementation of the Consumer interface.

This pattern shares some similarities with JavaScript’s implementation of callbacks, Promises, and the async/await pattern.

Let’s build a simple Consumer that logs some of the response data:

ws.url(url)
  .thenAccept(r -> 
    log.debug("Thread#" + Thread.currentThread().getId() 
      + " Request complete: Response code = " + r.getStatus() 
      + " | Response: " + r.getBody() 
      + " | Current Time:" + System.currentTimeMillis()))

Our Consumer is merely logging in this example. The consumer could do anything that we need to do with the result, though, like store the result in a database.

If we look deeper into the library’s implementation, we can observe that WS wraps and configures Java’s AsyncHttpClient, which is part of the standard JDK and does not depend on Play.

3. Prepare an Example Project

To experiment with the framework, let’s create some unit tests to launch requests. We’ll create a skeleton web application to answer them and use the WS framework to make HTTP requests.

3.1. The Skeleton Web Application

First of all, we create the initial project by using the sbt new command:

sbt new playframework/play-java-seed.g8

In the new folder, we then edit the build.sbt file and add the WS library dependency:

libraryDependencies += javaWs

Now we can start the server with the sbt run command:

$ sbt run
...
--- (Running the application, auto-reloading is enabled) ---

[info] p.c.s.AkkaHttpServer - Listening for HTTP on /0:0:0:0:0:0:0:0:9000

Once the application has started, we can check everything is ok by browsing http://localhost:9000, which will open Play’s welcome page.

3.2. The Testing Environment

To test our application, we’ll use the unit test class HomeControllerTest.

First, we need to extend WithServer which will provide the server life cycle:

public class HomeControllerTest extends WithServer {

Thanks to its parent, this class now starts our skeleton webserver in test mode and on a random port, before running the tests. The WithServer class also stops the application when the test is finished.

Next, we need to provide an application to run.

We can create it with Guice‘s GuiceApplicationBuilder:

@Override
protected Application provideApplication() {
    return new GuiceApplicationBuilder().build();
}

And finally, we set up the server URL to use in our tests, using the port number provided by the test server:

@Override
@Before
public void setup() {
    OptionalInt optHttpsPort = testServer.getRunningHttpsPort();
    if (optHttpsPort.isPresent()) {
        port = optHttpsPort.getAsInt();
        url = "https://localhost:" + port;
    } else {
        port = testServer.getRunningHttpPort()
          .getAsInt();
        url = "http://localhost:" + port;
    }
}

Now we’re ready to write tests. The comprehensive test framework lets us concentrate on coding our test requests.

4. Prepare a WSRequest

Let’s see how we can fire basic types of requests, such as GET or POST, and multipart requests for file upload.

4.1. Initialize the WSRequest Object

First of all, we need to obtain a WSClient instance to configure and initialize our requests.

In a real-life application, we can get a client, auto-configured with default settings, via dependency injection:

@Autowired
WSClient ws;

In our test class, though, we use WSTestClient, available from Play Test framework:

WSClient ws = play.test.WSTestClient.newClient(port);

Once we have our client, we can initialize a WSRequest object by calling the url method:

ws.url(url)

The url method does enough to allow us to fire a request. However, we can customize it further by adding some custom settings:

ws.url(url)
  .addHeader("key", "value")
  .addQueryParameter("num", "" + num);

As we can see, it’s pretty easy to add headers and query parameters.

After we’ve fully configured our request, we can call the method to initiate it.

4.2. Generic GET Request

To trigger a GET request we just have to call the get method on our WSRequest object:

ws.url(url)
  ...
  .get();

As this is a non-blocking code, it starts the request and then continues execution at the next line of our function.

The object returned by get is a CompletionStage instance, which is part of the CompletableFuture API.

Once the HTTP call has completed, this stage executes just a few instructions. It wraps the response in a WSResponse object.

Normally, this result would be passed on to the next stage of the execution chain. In this example, we have not provided any consuming function, so the result is lost.

For this reason, this request is of type “fire-and-forget”.

4.3. Submit a Form

Submitting a form is not very different from the get example.

To trigger the request we just call the post method:

ws.url(url)
  ...
  .setContentType("application/x-www-form-urlencoded")
  .post("key1=value1&key2=value2");

In this scenario, we need to pass a body as a parameter. This can be a simple string like a file, a json or xml document, a BodyWritable or a Source.

4.4. Submit a Multipart/Form Data

A multipart form requires us to send both input fields and data from an attached file or stream.

To implement this in the framework, we use the post method with a Source.

Inside the source, we can wrap all the different data types needed by our form:

Source<ByteString, ?> file = FileIO.fromPath(Paths.get("hello.txt"));
FilePart<Source<ByteString, ?>> file = 
  new FilePart<>("fileParam", "myfile.txt", "text/plain", file);
DataPart data = new DataPart("key", "value");

ws.url(url)
...
  .post(Source.from(Arrays.asList(file, data)));

Though this approach adds some more configuration, it is still very similar to the other types of requests.

5. Process the Async Response

Up to this point, we have only triggered fire-and-forget requests, where our code doesn’t do anything with the response data.

Let’s now explore two techniques for processing an asynchronous response.

We can either block the main thread, waiting for a CompletableFuture, or consume asynchronously with a Consumer.

5.1. Process Response by Blocking With CompletableFuture

Even when using an asynchronous framework, we may choose to block our code’s execution and wait for the response.

Using the CompletableFuture API, we just need a few changes in our code to implement this scenario:

WSResponse response = ws.url(url)
  .get()
  .toCompletableFuture()
  .get();

This could be useful, for example, to provide a strong data consistency that we cannot achieve in other ways.

5.2. Process Response Asynchronously

To process an asynchronous response without blocking, we provide a Consumer or Function that is run by the asynchronous framework when the response is available.

For example, let’s add a Consumer to our previous example to log the response:

ws.url(url)
  .addHeader("key", "value")
  .addQueryParameter("num", "" + 1)
  .get()
  .thenAccept(r -> 
    log.debug("Thread#" + Thread.currentThread().getId() 
      + " Request complete: Response code = " + r.getStatus() 
      + " | Response: " + r.getBody() 
      + " | Current Time:" + System.currentTimeMillis()));

We then see the response in the logs:

[debug] c.HomeControllerTest - Thread#30 Request complete: Response code = 200 | Response: {
  "Result" : "ok",
  "Params" : {
    "num" : [ "1" ]
  },
  "Headers" : {
    "accept" : [ "*/*" ],
    "host" : [ "localhost:19001" ],
    "key" : [ "value" ],
    "user-agent" : [ "AHC/2.1" ]
  }
} | Current Time:1579303109613

It’s worth noting that we used thenAccept, which requires a Consumer function since we don’t need to return anything after logging.

When we want the current stage to return something, so that we can use it in the next stage, we need thenApply instead, which takes a Function.

These use the conventions of the standard Java Functional Interfaces.

5.3. Large Response Body

The code we’ve implemented so far is a good solution for small responses and most use cases. However, if we need to process a few hundreds of megabytes of data, we’ll need a better strategy.

We should note: Request methods like get and post load the entire response in memory.

To avoid a possible OutOfMemoryError, we can use Akka Streams to process the response without letting it fill our memory.

For example, we can write its body in a file:

ws.url(url)
  .stream()
  .thenAccept(
    response -> {
        try {
            OutputStream outputStream = Files.newOutputStream(path);
            Sink<ByteString, CompletionStage<Done>> outputWriter =
              Sink.foreach(bytes -> outputStream.write(bytes.toArray()));
            response.getBodyAsSource().runWith(outputWriter, materializer);
        } catch (IOException e) {
            log.error("An error happened while opening the output stream", e);
        }
    });

The stream method returns a CompletionStage where the WSResponse has a getBodyAsStream method that provides a Source<ByteString, ?>.

We can tell the code how to process this type of body by using Akka’s Sink, which in our example will simply write any data passing through in the OutputStream.

5.4. Timeouts

When building a request, we can also set a specific timeout, so the request is interrupted if we don’t receive the complete response in time.

This is a particularly useful feature when we see that a service we’re querying is particularly slow and could cause a pile-up of open connections stuck waiting for the response.

We can set a global timeout for all our requests using tuning parameters. For a request-specific timeout, we can add to a request using setRequestTimeout:

ws.url(url)
  .setRequestTimeout(Duration.of(1, SECONDS));

There’s still one case to handle, though: We may have received all the data, but our Consumer may be very slow processing it. This might happen if there is lots of data crunching, database calls, etc.

In low throughput systems, we can simply let the code run until it completes. However, we may wish to abort long-running activities.

To achieve that, we have to wrap our code with some futures handling.

Let’s simulate a very long process in our code:

ws.url(url)
  .get()
  .thenApply(
    result -> { 
        try { 
            Thread.sleep(10000L); 
            return Results.ok(); 
        } catch (InterruptedException e) { 
            return Results.status(SERVICE_UNAVAILABLE); 
        } 
    });

This will return an OK response after 10 seconds, but we don’t want to wait that long.

Instead, with the timeout wrapper, we instruct our code to wait for no more than 1 second:

CompletionStage<Result> f = futures.timeout(
  ws.url(url)
    .get()
    .thenApply(result -> {
        try {
            Thread.sleep(10000L);
            return Results.ok();
        } catch (InterruptedException e) {
            return Results.status(SERVICE_UNAVAILABLE);
        }
    }), 1L, TimeUnit.SECONDS);

Now our future will return a result either way: the computation result if the Consumer finished in time, or the exception due to the futures timeout.

5.5. Handling Exceptions

In the previous example, we created a function that either returns a result or fails with an exception. So, now we need to handle both scenarios.

We can handle both success and failure scenarios with the handleAsync method.

Let’s say that we want to return the result, if we’ve got it, or log the error and return the exception for further handling:

CompletionStage<Object> res = f.handleAsync((result, e) -> {
    if (e != null) {
        log.error("Exception thrown", e);
        return e.getCause();
    } else {
        return result;
    }
});

The code should now return a CompletionStage containing the TimeoutException thrown.

We can verify it by simply calling an assertEquals on the class of the exception object returned:

Class<?> clazz = res.toCompletableFuture().get().getClass();
assertEquals(TimeoutException.class, clazz);

When running the test, it will also log the exception we received:

[error] c.HomeControllerTest - Exception thrown
java.util.concurrent.TimeoutException: Timeout after 1 second
...

6. Request Filters

Sometimes, we need to run some logic before a request is triggered.

We could manipulate the WSRequest object once initialized, but a more elegant technique is to set a WSRequestFilter.

A filter can be set during initialization, before calling the triggering method, and is attached to the request logic.

We can define our own filter by implementing the WSRequestFilter interface, or we can add a ready-made one.

A common scenario is logging what the request looks like before executing it.

In this case, we just need to set the AhcCurlRequestLogger:

ws.url(url)
  ...
  .setRequestFilter(new AhcCurlRequestLogger())
  ...
  .get();

The resulting log has a curl-like format:

[info] p.l.w.a.AhcCurlRequestLogger - curl \
  --verbose \
  --request GET \
  --header 'key: value' \
  'http://localhost:19001'

We can set the desired log level, by changing our logback.xml configuration.

7. Caching Responses

WSClient also supports the caching of responses.

This feature is particularly useful when the same request is triggered multiple times and we don’t need the freshest data every time.

It also helps when the service we’re calling is temporarily down.

7.1. Add Caching Dependencies

To configure caching we need first to add the dependency in our build.sbt:

libraryDependencies += ehcache

This configures Ehcache as our caching layer.

If we don’t want Ehcache specifically, we can use any other JSR-107 cache implementation.

7.2. Force Caching Heuristic

By default, Play WS won’t cache HTTP responses if the server doesn’t return any caching configuration.

To circumvent this, we can force the heuristic caching by adding a setting to our application.conf:

play.ws.cache.heuristics.enabled=true

This will configure the system to decide when it’s useful to cache an HTTP response, regardless of the remote service’s advertised caching.

8. Additional Tuning

Making requests to an external service may require some client configuration. We may need to handle redirects, a slow server, or some filtering depending on the user-agent header.

To address that, we can tune our WS client, using properties in our application.conf:

play.ws.followRedirects=false
play.ws.useragent=MyPlayApplication
play.ws.compressionEnabled=true
# time to wait for the connection to be established
play.ws.timeout.connection=30
# time to wait for data after the connection is open
play.ws.timeout.idle=30
# max time available to complete the request
play.ws.timeout.request=300

It’s also possible to configure the underlying AsyncHttpClient directly.

The full list of available properties can be checked in the source code of AhcConfig.

9. Conclusion

In this article, we explored the Play WS library and its main features. We configured our project, learned how to fire common requests and to process their response, both synchronously and asynchronously.

We worked with large data downloads and saw how to cut short long-running activities.

Finally, we looked at caching to improve performance, and how to tune the client.

As always, the source code for this tutorial is available over on GitHub.