A simple multi-threaded web server written in Java and implementing the HTTP/1.1 specification.
- Simple file serving
- Configurable connection thread pooling
- Keep-alive (persistent) connections
- Extensible request handlers
- Access and error logging
- Automatic directory listings
- Server info summary
- Java 8
- Maven 3
To build the project just run the Maven package
goal:
mvn clean package
This will produce an executable JAR in the target
directory.
Start the server:
java -jar target/java-webserver-1.0.0.jar
This will start the server using the default settings: a randomly assigned available port, a connection timeout of 10 seconds, up to 20 worker threads, and a document root at ./docroot
.
With the server running you can browse the docroot by visiting http://localhost:<port>/
or see an overview of the current configuration at http://localhost:<port>/serverInfo
.
You can configure the port, timeout, worker threads and other settings with the corresponding option. For example, to run the server on port 80 with a document root at /var/www/html
you would use the following:
java -jar target/java-webserver-1.0.0.jar --port 80 --docroot /var/www/html
To display all the options and usage information:
java -jar target/java-webserver-1.0.0.jar --help
The server is designed to be extensible and easy to understand and use, and code is written to be expressive and well documented.
The operation of the server can be broken down into a few key components responsible for handling client connections, parsing requests, and generating responses. Besides these there are various utilities to facilitate these functions.
The centre of connection handling is the WebServerExecutor
, which maintains the main server socket bound to the port
the server is running on. It continuously listens for new client connections and as soon as one is made it delegates it
to a WebWorker
. The WebWorkers
are queued in a thread pool, to be executed as soon as a thread becomes available.
The WebServerExecutor
is designed to do the minimal amount of work needed to delegate a request to a worker to
maximise throughput.
Each WebWorker
is "disposable" and is only responsible for a single client connection. In some cases the connection
might only last for a single request-response cycle, or in the case of persistent or "keep-alive" connections (the
default for HTTP/1.1) the worker may service any number of requests. The details of how workers decide how to respond to
a request is explained further down.
Client connections are streamed through a RequestReader
, which reads a stream of HTTP request data into Request
objects for further processing.
To decide how to respond to a request WebWorkers
rely on RequestHandlers
, which are at the centre of defining the
server's behaviour. Each handler indicates if it knows how to handle a given request, and if the handler agrees to
handle a request it is then asked to provide a response. Since there could be any number of RequestHandlers
registered, WebWorkers
deal with handlers via the ResponseFactory
, rather than talking to each handler directly.
For the sake of simplicity of the implementation, the default RequestHandlers
are hand-defined in the application's
"main" class (App
). In a commercial scenario they would instead be configured via a configuration file, similar to how
Apache HTTPD modules are configured.
After the the worker has generated a response, it streams it back to the client with the help of a ResponseWriter
,
which is essentially the reverse of RequestReader
in that is writes Response
to an output stream.
This design makes implementing new server behaviours very easy, since handlers need not be concerned about either the client connection lifecycle, or the logistics of reading or writing HTTP messages.
There are several out-of-the-box Response
types for serving files (FileResponse
), sending errors (ErrorResponse
),
redirects (RedirectResponse
) and listing directory contents (DirectoryListingResponse
). The Response
uses a
stream-based system for response bodies, which aid performance particularly when serving large files since they do not
need to be read into memory and can instead be streamed directly to the client.
The entry point to the application is App
, which handles the reading of CLI options and uses them to create a new
WebServer
instance. It also creates and configures the default RequestHandler
s (see above for rationale).
Superficially, WebServer
and WebServerExecutor
may seem similar, however WebServer
is actually the controller for
WebServerExecutor
, since listening for client connections is a blocking operation and WebServer
and is designed to
be non-blocking so consumers can easily start and stop individual servers without worrying spawning and managing the
threads to run them in. Starting and stopping a new server is as simple as:
WebServer server = new WebServer(port, timeout, threads, handlers);
server.start();
// do something
server.stop();
The server handles requests by delegating them to various RequestHandler
s, which are registered at startup time. The
server behaviour can be readily extended by adding new handlers and registering them. For more information about how
handlers are used see "How it works" above.
The handler specification is simple:
public interface RequestHandler {
/**
* Checks if the handler is able to provide a response to a particular request.
*
* @param request the request to be handled.
* @return Returns true if the handler can provide a response to this request, or false if it cannot.
*/
boolean canHandle(Request request);
/**
* Provides a response to a request.
*
* @param request the request to provide a response for
* @return Returns the response to the request.
*/
Response handle(Request request);
}
Implementations simply answer "yes" or "no" when asked if they know how to handle a particular request, then provide a response.
There are several out-the-box Response
types like FileResponse
, RedirectResponse
and ErrorResponse
, and it is
straightforward to add others for more specific applications.
The handlers are registered in the main loop of App
in order of priority (highest first).
Let's create a new handler that will print the current server time when someone visits /time
.
First, we'll create a new Response
type that outputs the current time as plain text. We could use HttpResponse
directly, but this way we can keep the code cleaner and easier to test.
import au.id.deejay.webserver.api.HttpStatus;
import au.id.deejay.webserver.api.HttpVersion;
import au.id.deejay.webserver.api.Response;
import au.id.deejay.webserver.headers.Headers;
import au.id.deejay.webserver.headers.HttpHeader;
import au.id.deejay.webserver.headers.HttpHeaders;
import java.io.ByteArrayInputStream;
import java.io.InputStream;
import java.time.LocalDateTime;
import java.time.format.DateTimeFormatter;
public class CurrentTimeResponse implements Response {
private final String timeString;
public CurrentTimeResponse() {
// Convert the current time into a string
DateTimeFormatter formatter = DateTimeFormatter.ofPattern("h:mm a");
timeString = "The time is " + LocalDateTime.now().format(formatter);
}
@Override
public HttpStatus status() {
return HttpStatus.OK_200;
}
@Override
public InputStream stream() {
return new ByteArrayInputStream(timeString.getBytes());
}
@Override
public Headers headers() {
return new HttpHeaders(
new HttpHeader("Content-type", "text/plain"),
new HttpHeader("Content-length", String.valueOf(timeString.length())));
}
@Override
public HttpVersion version() {
return HttpVersion.HTTP_1_0;
}
}
Now we'll create a RequestHandler
to capture requests to /time
and respond using our new response type:
import au.id.deejay.webserver.api.HttpMethod;
import au.id.deejay.webserver.api.Request;
import au.id.deejay.webserver.api.RequestHandler;
import au.id.deejay.webserver.api.Response;
import au.id.deejay.webserver.response.CurrentTimeResponse;
public class CurrentTimeHandler implements RequestHandler {
private static final String PATH = "/time";
@Override
public boolean canHandle(Request request) {
return request.method() == HttpMethod.GET && PATH.equals(request.uri().getPath());
}
@Override
public Response handle(Request request) {
return new CurrentTimeResponse();
}
}
Finally, we'll register the new handler in App#main()
, along with the handlers already there:
// Configure request handlers
RequestHandler serverInfoHandler = new ServerInfoHandler(...);
RequestHandler docrootHandler = new DocrootHandler(...);
RequestHandler timeHandler = new CurrentTimeHandler(); // our new handler
List<RequestHandler> requestHandlers = Arrays.asList(timeHandler, serverInfoHandler, docrootHandler);
Now if we start the server and visit http://localhost/time
we should see something like:
The time is 5:35 PM
And that's all there to adding new server functionality. For a more full-featured example, check out DocrootHandler
.
- Sample content courtesy of Twitter Bootstrap, licensed under The MIT License.
- Favicon courtesy of FamFamFam Silk icons collection, licensed under Creative Commons Attribution 3.0 License
- And, of course, all Maven dependencies are courtesy of their respective owners.