A few days ago I've started developing a Backend for Mobile Applications using Google App Engine and Google Cloud Endpoints. This tutorial shows how the endpoints are getting generated automatically, as well as the client library for Android.

So we have our Entity:

@Entity
public class Person implements IsSerializable{

    @Id
    @GeneratedValue(strategy = GenerationType.IDENTITY)
    private Key key;

    private String name;
//...
} 

And the endpoint for this class:

@Api(name = "personendpoint")
public class PersonEndpoint {

    @ApiMethod(name = "getPerson")
    public Person getPerson(@Named("id") Long id) {
...

Besides, using the generated Android Endpoint Library (that uses the REST API) I would like to add a User Interface on the Server, build with Google Web Toolkit (GWT). But how should I manipulate the date on Server Side? I can see different approaches...

Option A1: Adding an RPC Service in GWT

public interface PersonServiceAsync {

    void insertPerson(Person person, AsyncCallback<Person> callback);

}

@RemoteServiceRelativePath("api")
public interface PersonService extends RemoteService {

    public Person insertPerson(Person person);

}
public class PersonServiceImpl extends RemoteServiceServlet implements PersonService{
    public Person insertPerson(Person person) {
        EntityManager mgr = getEntityManager();
        try {
            if (containsPerson(person)) {
                throw new EntityExistsException("Object already exists");
            }
            mgr.persist(person);
        } finally {
            mgr.close();
        }
        return person;
    }

        //...
}

But now my PersonServiceImpl and PersonEndpoint do roughly the same. So we didn't follow DRY :) Besides, that Person is not allowed to have com.google.appengine.api.datastore.Key so we would have to change our Entities.

Option A2: Service calls endpoint class

@Override
public Person insertPerson(Person person) {
    return new PersonEndpoint().insertPerson(person);
}

Should work , but still no com.google.appengine.api.datastore.Key Type in the Entity and since Endpoints are using CollectionResponse<Person> we would have to transform this into an Collection<Person> in case of listPerson().

Option B1: Using the Java Endpoint Client Library

We could split the GWT Client from our App Engine API Backend and use the generated Endpoint Client Libraries for Java. So we call the REST/Endpoint-API from within a RemoteServiceServlet. But wouldn't this end up in two Requests, even if the GWT client and Endpoints are on the same server or even in the same project?

GWT Client --(RPC)--> GWT Server --(HTTP Request)--> App Engine Backend Server

Option B2: Using the JavaScript Endpoint Client Library

Might be the best approach but would end up in massive JSNI.


So what's best practice? I can't find any example projects using Google Cloud Endpoints AND GWT in one project :)


Solution 1:

The good old DTO dilemma. There is no right or wrong, just what is good enough for you.

Repeating yourself can be a good thing. Right now you are exposing your data model through your endpoint, which means that any change of your Entities will impact your mobile app users. Let's say you rename an attribute on the server side -> every client who has not updated the app goes down.

Security is also an issue : if your User entity has an "email" property, serializing it through GWT RPC will make your user's email virtually available to any javascript debugger.

Is it really what you want?

Don't get me wrong, I'm not a fan of those "onion layers" monster apps where 80% of the code seems to be made to transform objects into other objets with virtually the same properties.

I think the right solution is in between: having a "client" model (DTOs), made of serializable POJOs (no datastore, ORM, JAXB, whatever annotation) that you expose through both GWT RPC and Client Endpoints. Your GWT servlet implementation and the Endpoint server would call the same service that will transform your client model into entities and process/persist them.

This way you can reuse your code, still keep it simple, have a uniform interface accross your APIs, and allow your inner plumbering to evolve without altering the client interfaces.