Vert.x Service Proxy

When you compose a Vert.x application, you may want to isolate a functionality somewhere and make it available to the rest of your application. That’s the main purpose of service proxies. It lets you expose a service on the event bus, so, any other Vert.x component can consume it, as soon as they know the address on which the service is published.

A service is described with a Java interface containing methods following the async pattern. Under the hood, messages are sent on the event bus to invoke the service and get the response back. But for ease of use, it generates a proxy that you can invoke directly (using the API from the service interface).

Using Vert.x service proxies

To use Vert.x Service Proxies, add the following dependency to the dependencies section of your build descriptor:

  • Maven (in your pom.xml):

<dependency>
 <groupId>io.vertx</groupId>
 <artifactId>vertx-service-proxy</artifactId>
 <version>4.5.7</version>
</dependency>
  • Gradle (in your build.gradle file):

compile 'io.vertx:vertx-service-proxy:4.5.7'

To implement service proxies, also add:

  • Maven (in your pom.xml):

<dependency>
 <groupId>io.vertx</groupId>
 <artifactId>vertx-codegen</artifactId>
 <version>4.5.7</version>
 <classifier>processor</classifier>
 <scope>provided</scope>
</dependency>
  • Gradle < 5 (in your build.gradle file):

compileOnly 'io.vertx:vertx-codegen:4.5.7'
  • Gradle >= 5 (in your build.gradle file):

annotationProcessor 'io.vertx:vertx-codegen:4.5.7:processor'
annotationProcessor 'io.vertx:vertx-service-proxy:4.5.7'

Be aware that as the service proxy mechanism relies on code generation, so modifications to the service interface require to re-compile the sources to regenerate the code.

To generate the proxies in different languages, you will need to add the language dependency such as vertx-lang-groovy for Groovy.

Introduction to service proxies

Let’s have a look at service proxies and why they can be useful. Let’s imagine you have a database service exposed on the event bus, you should do something like this:

JsonObject message = new JsonObject();

message
  .put("collection", "mycollection")
  .put("document", new JsonObject().put("name", "tim"));

DeliveryOptions options = new DeliveryOptions().addHeader("action", "save");

vertx.eventBus()
  .request("database-service-address", message, options)
  .onSuccess(msg -> {
    // done
  }).onFailure(err -> {
    // failure
  });

When creating a service there’s a certain amount of boilerplate code to listen on the event bus for incoming messages, route them to the appropriate method and return results on the event bus.

With Vert.x service proxies, you can avoid writing all that boilerplate code and concentrate on writing your service.

You write your service as a Java interface and annotate it with the @ProxyGen annotation, for example:

@ProxyGen
public interface SomeDatabaseService {

 // A couple of factory methods to create an instance and a proxy
 static SomeDatabaseService create(Vertx vertx) {
   return new SomeDatabaseServiceImpl(vertx);
 }

 static SomeDatabaseService createProxy(Vertx vertx,
   String address) {
   return new SomeDatabaseServiceVertxEBProxy(vertx, address);
 }

// Actual service operations here...
void save(String collection, JsonObject document,
  Handler<AsyncResult<Void>> resultHandler);
}

You will also need a package-info.java file somewhere in (or above) the package your interface is defined in. That package needs to be annotated with @ModuleGen so that Vert.x CodeGen can recognize your interfaces and generate the appropriate EventBus proxy code.

package-info.java
@io.vertx.codegen.annotations.ModuleGen(groupPackage = "io.vertx.example", name = "services", useFutures = true)
package io.vertx.example;

Given the interface, Vert.x will generate all the boilerplate code required to access your service over the event bus, and it will also generate a client side proxy for your service, so your clients can use a rich idiomatic API for your service instead of having to manually craft event bus messages to send. The client side proxy will work irrespective of where your service actually lives on the event bus (potentially on a different machine).

That means you can interact with your service like this:

SomeDatabaseService service = SomeDatabaseService
  .createProxy(vertx, "database-service-address");

// Save some data in the database - this time using the proxy
service.save(
  "mycollection",
  new JsonObject().put("name", "tim"),
  res2 -> {
    if (res2.succeeded()) {
      // done
    }
  });

You can also combine @ProxyGen with language API code generation (@VertxGen) in order to create service stubs in any of the languages supported by Vert.x - this means you can write your service once in Java and interact with it through an idiomatic other language API irrespective of whether the service lives locally or is somewhere else on the event bus entirely. For this don’t forget to add the dependency on your language in your build descriptor:

@ProxyGen // Generate service proxies
@VertxGen // Generate the clients
public interface SomeDatabaseService {
 // ...
}

Async interface

To be used by the service-proxy generation, the service interface must comply to a couple of rules. First it should follow the async pattern. To return a result, the method should declare a Future<ResultType> return type. ResultType can be another proxy (and so a proxies can be factories for other proxies).

Let’s see an example:

@ProxyGen
public interface SomeDatabaseService {

// A couple of factory methods to create an instance and a proxy

static SomeDatabaseService create(Vertx vertx) {
  return new SomeDatabaseServiceImpl(vertx);
}

static SomeDatabaseService createProxy(Vertx vertx, String address) {
  return new SomeDatabaseServiceVertxEBProxy(vertx, address);
}

// A method notifying the completion without a result (void)
Future<Void> save(String collection, JsonObject document);

// A method providing a result (a json object)
Future<JsonObject> findOne(String collection, JsonObject query);

// Create a connection
Future<MyDatabaseConnection> createConnection(String shoeSize);

}

with:

@ProxyGen
@VertxGen
public interface MyDatabaseConnection {

void insert(JsonObject someData);

Future<Void> commit();

@ProxyClose
void close();
}

You can also declare that a particular method unregisters the proxy by annotating it with the @ProxyClose annotation. The proxy instance is disposed when this method is called.

More constraints on the service interfaces are described below.

Async interface with callbacks

Until Vert.x 4.1, service async interfaces were defined by callbacks.

You can still create service async interface using callbacks, with this module declaration:

package-info.java
@io.vertx.codegen.annotations.ModuleGen(groupPackage = "io.vertx.example", name = "services", useFutures = false)
package io.vertx.example;
Note
for backward compatibility, the default value of useFutures is false so you can also omit the declaration

The service async interface with callbacks looks like:

@ProxyGen
public interface SomeDatabaseService {

 // A method notifying the completion without a result (void)
 void save(String collection, JsonObject document,
  Handler<AsyncResult<Void>> result);

 // A method providing a result (a json object)
 void findOne(String collection, JsonObject query,
  Handler<AsyncResult<JsonObject>> result);

 // Create a connection
 void createConnection(String shoeSize,
  Handler<AsyncResult<MyDatabaseConnection>> resultHandler);

}

Return types must be one of:

  • void

  • @Fluent and return reference to the service (this):

@Fluent
SomeDatabaseService doSomething();

This is because methods must not block and it’s not possible to return a result immediately without blocking if the service is remote.

Security

Service proxies can perform basic security using a simple interceptor. A authentication provider must be provided, optionally Authorization`s can be added and in this case an `AuthorizationProvider must also be present. Note that the authentication works based on tokens, that are extracted from the auth-token header.

SomeDatabaseService service = new SomeDatabaseServiceImpl();
// Register the handler
new ServiceBinder(vertx)
  .setAddress("database-service-address")
  // Secure the messages in transit
  .addInterceptor(
    "action",
    // Tokens will be validated using JWT authentication
    AuthenticationInterceptor.create(
      JWTAuth.create(vertx, new JWTAuthOptions())))
  .addInterceptor(
    AuthorizationInterceptor.create(JWTAuthorization.create("permissions"))
      // optionally we can secure permissions too:
      // an admin
      .addAuthorization(RoleBasedAuthorization.create("admin"))
      // that can print
      .addAuthorization(PermissionBasedAuthorization.create("print")))
  .register(SomeDatabaseService.class, service);

Code generation

Service annotated with @ProxyGen annotation trigger the generation of the service helper classes:

  • The service proxy: a compile time generated proxy that uses the EventBus to interact with the service via messages

  • The service handler: a compile time generated EventBus handler that reacts to events sent by the proxy

Generated proxies and handlers are named after the service class, for example if the service is named MyService the handler is called MyServiceProxyHandler and the proxy is called MyServiceEBProxy.

In addition Vert.x Core provides a generator creating data object converters to ease data object usage in service proxies. Such converter provides a basis for the JsonObject constructor and the toJson() method that are necessary for using data objects in service proxies.

The codegen annotation processor generates these classes at compilation time. It is a feature of the Java compiler so no extra step is required, it is just a matter of configuring correctly your build:

Just add the io.vertx:vertx-codegen:processor and io.vertx:vertx-service-proxy dependencies to your build.

Here a configuration example for Maven:

<dependency>
 <groupId>io.vertx</groupId>
 <artifactId>vertx-codegen</artifactId>
 <version>4.5.7</version>
 <classifier>processor</classifier>
</dependency>
<dependency>
 <groupId>io.vertx</groupId>
 <artifactId>vertx-service-proxy</artifactId>
 <version>4.5.7</version>
</dependency>

This feature can also be used in Gradle:

compile "io.vertx:vertx-codegen:4.5.7:processor"
compile "io.vertx:vertx-service-proxy:4.5.7"

IDE provides usually support for annotation processors.

The codegen processor classifier adds to the jar the automatic configuration of the service proxy annotation processor via the META-INF/services plugin mechanism.

If you want you can use it too with the regular jar but you need then to declare the annotation processor explicitly, for instance in Maven:

<plugin>
 <artifactId>maven-compiler-plugin</artifactId>
 <configuration>
   <annotationProcessors>
     <annotationProcessor>io.vertx.codegen.CodeGenProcessor</annotationProcessor>
   </annotationProcessors>
 </configuration>
</plugin>

Exposing your service

Once you have your service interface, compile the source to generate the stub and proxies. Then, you need some code to "register" your service on the event bus:

SomeDatabaseService service = new SomeDatabaseServiceImpl();
// Register the handler
new ServiceBinder(vertx)
  .setAddress("database-service-address")
  .register(SomeDatabaseService.class, service);

This can be done in a verticle, or anywhere in your code.

Once registered, the service becomes accessible. If you are running your application on a cluster, the service is available from any host.

To withdraw your service, use the unregister method:

ServiceBinder binder = new ServiceBinder(vertx);

// Create an instance of your service implementation
SomeDatabaseService service = new SomeDatabaseServiceImpl();
// Register the handler
MessageConsumer<JsonObject> consumer = binder
  .setAddress("database-service-address")
  .register(SomeDatabaseService.class, service);

// ....

// Unregister your service.
binder.unregister(consumer);

Proxy creation

Now that the service is exposed, you probably want to consume it. For this, you need to create a proxy. The proxy can be created using the ServiceProxyBuilder class:

ServiceProxyBuilder builder = new ServiceProxyBuilder(vertx)
  .setAddress("database-service-address");

SomeDatabaseService service = builder.build(SomeDatabaseService.class);
// or with delivery options:
SomeDatabaseService service2 = builder.setOptions(options)
  .build(SomeDatabaseService.class);

The second method takes an instance of DeliveryOptions where you can configure the message delivery (such as the timeout).

Alternatively, you can use the generated proxy class. The proxy class name is the service interface class name followed by VertxEBProxy. For instance, if your service interface is named SomeDatabaseService, the proxy class is named SomeDatabaseServiceVertxEBProxy.

Generally, service interface contains a createProxy static method to create the proxy. But this is not required:

@ProxyGen
public interface SomeDatabaseService {

// Method to create the proxy.
static SomeDatabaseService createProxy(Vertx vertx, String address) {
  return new SomeDatabaseServiceVertxEBProxy(vertx, address);
}

// ...
}

Error Handling

Service methods may return errors to the client by passing a failed Future containing a ServiceException instance to the method’s Handler. A ServiceException contains an int failure code, a message, and an optional JsonObject containing any extra information deemed important to return to the caller. For convenience, the ServiceException.fail factory method can be used to create an instance of ServiceException already wrapped in a failed Future. For example:

public class SomeDatabaseServiceImpl implements SomeDatabaseService {

 private static final BAD_SHOE_SIZE = 42;
 private static final CONNECTION_FAILED = 43;

 // Create a connection
 public Future<MyDatabaseConnection> createConnection(String shoeSize) {
   if (!shoeSize.equals("9")) {
     return Future.failedFuture(ServiceException.fail(BAD_SHOE_SIZE, "The shoe size must be 9!",
       new JsonObject().put("shoeSize", shoeSize)));
    } else {
       return doDbConnection().recover(err -> Future.failedFuture(ServiceException.fail(CONNECTION_FAILED, result.cause().getMessage())));
    }
 }
}

The client side can then check if the Throwable it receives from a failed Future is a ServiceException, and if so, check the specific error code inside. It can use this information to differentiate business logic errors from system errors (like the service not being registered with the Event Bus), and to determine exactly which business logic error occurred.

public Future<JsonObject> foo(String shoeSize) {
 SomeDatabaseService service = SomeDatabaseService.createProxy(vertx, SERVICE_ADDRESS);
 server.createConnection("8")
   .compose(connection -> {
     // Do success stuff.
     return doSuccessStuff(connection);
   })
   .recover(err -> {
     if (err instanceof ServiceException) {
       ServiceException exc = (ServiceException) err;
       if (exc.failureCode() == SomeDatabaseServiceImpl.BAD_SHOE_SIZE) {
         return Future.failedFuture(
           new InvalidInputError("You provided a bad shoe size: " +
             exc.getDebugInfo().getString("shoeSize")));
       } else if (exc.failureCode() == SomeDatabaseServiceImpl.CONNECTION) {
         return Future.failedFuture(new ConnectionError("Failed to connect to the DB"));
       }
     } else {
       // Must be a system error (e.g. No service registered for the proxy)
       return Future.failedFuture(new SystemError("An unexpected error occurred: + " result.cause().getMessage()));
     }
   });
}

If desired, service implementations may also return a sub-class of ServiceException, as long as a default MessageCodec is registered for it . For example, given the following ServiceException sub-class:

class ShoeSizeException extends ServiceException {
 public static final BAD_SHOE_SIZE_ERROR = 42;

 private final String shoeSize;

 public ShoeSizeException(String shoeSize) {
   super(BAD_SHOE_SIZE_ERROR, "In invalid shoe size was received: " + shoeSize);
   this.shoeSize = shoeSize;
 }

 public String getShoeSize() {
   return extra;
 }

 public static <T> Future<T> fail(int failureCode, String message, String shoeSize) {
   return Future.failedFuture(new MyServiceException(failureCode, message, shoeSize));
 }
}

As long as a default MessageCodec is registered, the Service implementation can return the custom exception directly to the caller:

public class SomeDatabaseServiceImpl implements SomeDatabaseService {
 public SomeDataBaseServiceImpl(Vertx vertx) {
   // Register on the service side. If using a local event bus, this is all
   // that's required, since the proxy side will share the same Vertx instance.
 SomeDatabaseService service = SomeDatabaseService.createProxy(vertx, SERVICE_ADDRESS);
   vertx.eventBus().registerDefaultCodec(ShoeSizeException.class,
     new ShoeSizeExceptionMessageCodec());
 }

 // Create a connection
 Future<MyDatabaseConnection> createConnection(String shoeSize) {
   if (!shoeSize.equals("9")) {
     return ShoeSizeException.fail(shoeSize);
   } else {
     // Create the connection here
     return Future.succeededFuture(myDbConnection);
   }
 }
}

Finally, the client can now check for the custom exception:

public Future<JsonObject> foo(String shoeSize) {
 // If this code is running on a different node in the cluster, the
 // ShoeSizeExceptionMessageCodec will need to be registered with the
 // Vertx instance on this node, too.
 SomeDatabaseService service = SomeDatabaseService.createProxy(vertx, SERVICE_ADDRESS);
 service.createConnection("8")
   .compose(connection -> {
     // Do success stuff.
     return doSuccessStuff(connection);
   })
   .recover(err -> {
     if (result.cause() instanceof ShoeSizeException) {
       ShoeSizeException exc = (ShoeSizeException) result.cause();
       return Future.failedFuture(
         new InvalidInputError("You provided a bad shoe size: " + exc.getShoeSize()));
     } else {
       // Must be a system error (e.g. No service registered for the proxy)
       return Future.failedFuture(
         new SystemError("An unexpected error occurred: + " result.cause().getMessage())
       );
     }
   });
}

Note that if you’re clustering Vertx instances, you’ll need to register the custom Exception’s MessageCodec with each Vertx instance in the cluster.

Restrictions for service interface

There are restrictions on the types and return values that can be used in a service method so that these are easy to marshall over event bus messages and so they can be used asynchronously. They are:

Data types

Let JSON = JsonObject | JsonArray Let PRIMITIVE = Any primitive type or boxed primitive type

Parameters can be any of:

  • JSON

  • PRIMITIVE

  • List<JSON>

  • List<PRIMITIVE>

  • Set<JSON>

  • Set<PRIMITIVE>

  • Map<String, JSON>

  • Map<String, PRIMITIVE>

  • Any Enum type

  • Any class annotated with @DataObject

Asynchronous results are modelled as

  • Future<R>

  • Handler<AsyncResult<R>> for callback style

R can be any of:

  • JSON

  • PRIMITIVE

  • List<JSON>

  • List<PRIMITIVE>

  • Set<JSON>

  • Set<PRIMITIVE>

  • Any Enum type

  • Any class annotated with @DataObject

  • Another proxy

Overloaded methods

There must be no overloaded service methods. (i.e. more than one with the same name, regardless the signature).

Convention for invoking services over the event bus (without proxies)

Service Proxies assume that event bus messages follow a certain format so they can be used to invoke services.

Of course, you don’t have to use client proxies to access remote service if you don’t want to. It’s perfectly acceptable to interact with them by just sending messages over the event bus.

In order for services to be interacted with a consistent way the following message formats must be used for any Vert.x services.

The format is very simple:

  • There should be a header called action which gives the name of the action to perform.

  • The body of the message should be a JsonObject, there should be one field in the object for each argument needed by the action.

For example to invoke an action called save which expects a String collection and a JsonObject document:

Headers:
   "action": "save"
Body:
   {
       "collection", "mycollection",
       "document", {
           "name": "tim"
       }
   }

The above convention should be used whether or not service proxies are used to create services, as it allows services to be interacted with consistently.

In the case where service proxies are used the "action" value should map to the name of an action method in the service interface and each [key, value] in the body should map to a [arg_name, arg_value] in the action method.

For return values the service should use the message.reply(…​) method to send back a return value - this can be of any type supported by the event bus. To signal a failure the method message.fail(…​) should be used.

If you are using service proxies the generated code will handle this for you automatically.