RPC

Description

Light-weight RPC package for creating RESTful server-side Dart APIs. The package supports the Google Discovery Document format for message encoding and HTTP REST for routing of requests.

The discovery documents for the API are automatically generated and are compatible with existing Discovery Document client stub generators (see the "Calling the API" section below for more details). This makes it easy to create a server side API that can be called by any client language for which there is a Discovery Document client stub generator.

Simple Example

Getting started is simple! The example below gives a quick overview of how to create an API and in the following sections a more elaborate description follows of how to build the API and setup an API server.

@ApiClass(version: 'v1')
class Cloud {
  @ApiMethod(method: 'GET', path: 'resource/{name}')
  ResourceMessage getResource(String name) {
    ... find resource of name {resourceName} ...
    return new ResourceMessage
        ..id = resource.id
        ..name = resource.name
        ..capacity = resource.capacity;
  }

  @ApiMethod(method: 'POST', path: 'resource/{name}/update')
  VoidMessage updateResource(String name, UpdateMessage request) {
    ... process request, throw on error ...
  }
}

class ResourceMessage {
  int id;
  String name;
  int capacity;
}

class UpdateMessage {
  int newCapacity;
}

Two complete examples using respectively dart:io and shelf can be found at Example.

Usage

Terminology

We use the following concepts below when describing how to build your API.

• Top-level class - This is the API entry-point. It describes the API name and version. The top-level class is defined via the ApiClass annotation.
• Resource - Resources are used to group methods together for a cleaner API structure. Class fields annotated with @ApiResource are exposed as resources.
• Method - Methods are what can be invoked. Only methods annotated with @ApiMethod are exposed as remotely accessible methods.
• Schema - Schemas are used to describe response and the request messages passed in the body of the HTTP request.
• Properties - A schema contains properties. Each property can optionally be further restricted by the ApiProperty annotation.

Main API Class

Defining an API starts with annotating a class with the @ApiClass annotation. It must specify at least the version field. The API name can optionally be specified via the name field and will default to the class name in camel-case if omitted.

@ApiClass(
  name: 'cloud',  // Optional (default is 'cloud' since class name is Cloud).
  version: 'v1',
  description: 'My Dart server side API' // optional
)
class Cloud  {
  (...)
}

The above API would be available at the path /cloud/v1. E.g. if the server was serving on http://localhost:8080 the API base url would be http://localhost:8080/cloud/v1.

Methods

Inside of your API class you can define public methods that will correspond to methods that can be called on your API.

For a method to be exposed as a remote API method it must be annotated with the @ApiMethod annotation specifying a unique path used for routing requests to the method.

The @ApiMethod annotation also supports specifying the HTTP method used to invoke the method. The method field is used for this. If omitted the HTTP method defaults to GET.

A description of the method can also be specified using the description field. If omitted it defaults to the empty string.

Response message (return value)

A method must always return a response. The response can be either an instance of a class or a future of the instance. In the case where a method has no response the predefined VoidMessage class should be returned.

Example method returning nothing:

@ApiMethod(path: 'voidMethod')
VoidMessage myVoidMethod() {
  ...
  return null;
}

Example method returning class:

class MyResponse {
  String result;
}

@ApiMethod(path: 'someMethod')
MyResponse myMethod() {
  ...
  return new MyResponse();
}

Example method returning a future:

@ApiMethod(path: 'futureMethod')
Future<MyResponse> myFutureMethod() {
  ...
    completer.complete(new MyResponse();
  ...
  return completer.future;
}

The MyResponse class must be a non-abstract class with an unnamed constructor taking no required parameters. The RPC backend will automatically serialize all public fields of the the MyResponse instance into JSON corresponding to the generated Discovery Document schema.

Parameters

Method parameters can be passed in three different ways.

• As a path parameter in the method path (supported on all HTTP methods)
• As a query string parameter (supported for GET and DELETE)
• As the request body (supported for POST and PUT)

Path parameters and the request body parameter are required. The query string parameters are optional named parameters.

Example of a method using POST with both path parameters and a request body:

@ApiMethod(
  method: 'POST',
  path: 'resource/{name}/type/{type}')
MyResponse myMethod(String name, String type, MyRequest request) {
  ...
  return new MyResponse();
}

The curly brackets specify path parameters and must appear as positional parameters in the same order as on the method signature. The request body parameter is always specified as the last parameter.

Assuming the above method was part of the Cloud class defined above the url to the method would be:

http://localhost:8080/cloud/v1/resource/foo/type/storage

where the first parameter name would get the value foo and the type parameter would get the value storage.

The MyRequest class must be a non-abstract class with an unnamed constructor taking no arguments. The RPC backend will automatically create an instance of the MyRequest class, decode the JSON request body, and set the class instance's fields to the values found in the decoded request body.

If the request body is not needed it is possible to use the VoidMessage class or change it to use the GET HTTP method. If using GET the method signature would instead become.

@ApiMethod(path: '/resource/{name}/type/{type}')
MyResponse myMethod(String name, String type) {
   ...
   return new MyResponse(); 
}

When using GET it is possible to use optional named parameters as below.

@ApiMethod(path: '/resource/{name}/type/{type}')
MyResponse myMethod(String name, String type, {String filter}) {
   ...
   return new MyResponse(); 
}

in which case the caller can pass the filter as part of the query string. E.g.

http://localhost:8080/cloud/v1/resource/foo/type/storage?filter=fast

More about Request/Response Messages

The data sent either as a request (using HTTP POST and PUT) or as a response body corresponds to a non-abstract class as described above.

The RPC backend will automatically decode HTTP request bodies into class instances and encode method results into an HTTP response's body. This is done according to the generated Discovery Document schemas.

Only the public fields of the classes are encoded/decoded. Currently supported types for the public fields are int, double, bool, String, DateTime, List, Map<String, SomeType>, and another message class.

A field can be further annotated using the @ApiProperty annotation to specify default values, format of an int or double specifying how to handle it in the backend, min/max value of an int property, and whether a property is required.

For int properties the format field is used to specify the size of the integer. It can take the values int32, uint32, int64 or uint64. The 64-bit variants will be represented as String in the JSON objects.

For int properties the minValue and maxValue fields can be used to specify the min and max value of the integer.

For double properties the format parameter can take the value double or float.

The defaultValue field is used to specify a default value. The required fields is used to specify whether a field is required.

Example schema:

class MyRequest {
   @ApiProperty(
     format: 'uint32',
     defaultValue: 40,
     minValue: 0,
     maxValue: 150)
   int age;

   @ApiProperty(format: 'float')
   double averageAge;
}

Resources

Resources can be used to provide structure to your API by grouping certain API methods together under a resource. To create an API resource you will add a field to the class annotated with the @ApiClass annotation. The field must point to another class (the resource) containing the methods that should be exposed together for this resource. The field must be annotated with the @ApiResource annotation. By default the name of the resource will be the field name in camel-case. If another name is desired the name field can be used in the @ApiResource annotation.

Example resource API:

@ApiClass(version: 'v1')
class Cloud {

  @ApiResource(name: 'myResource')
  MyResource aResource = new MyResource();
  
  ...
}

class MyResource {
  
  @ApiMethod(path: 'someMethod')
  MyResponse myResourceMethod() { return new MyResponse(); }
}

Notice the @ApiResource annotation is on the field rather than the resource class. This allows for a resource class to be used in multiple places (e.g. different versions) of the API.

Also notice the path of the MyResource.myResourceMethod method is independent from the resource. E.g. if MyResource was used in the previous mentioned Cloud API the method would be exposed at the url http://<server ip>:<port>/cloud/v1/someMethod.

API Server

When having annotated your classes, resources, and methods you must create an ApiServer to route the HTTP requests to your methods. Creating a RPC API server is done by first creating an instance of the ApiServer class and calling the addApi method with an instance of the class annotated with the @ApiClass annotation.

You can choose to use any web server framework you prefer for serving HTTP requests. The rpc-examples github repository (https://github.com/dart-lang/rpc-examples) includes examples for both the standard dart:io HttpServer as well as an example using the shelf middleware.

E.g. to use dart:io you would do something like:

final ApiServer _apiServer = new ApiServer();

main() async {
  _apiServer.addApi(new Cloud());
  HttpServer server = await HttpServer.bind(InternetAddress.ANY_IP_V4, 8080);
  server.listen(_apiServer.httpRequestHandler);
}

The above example uses the default provided ApiServer HTTP request handler which converts the HttpRequest to a HttpApiRequest and forwards it along. A custom HTTP request handler doing the conversion to the HttpApiRequest class and calling the ApiServer.handleHttpApiRequest method itself can also be used if more flexibility is needed.

Notice that the ApiServer is agnostic of the HTTP server framework being used by the application. The RPC package does provide a request handler for the standard dart:io HttpRequest class. There is also a shelf_rpc package which provides the equivalent for shelf (see the example for how this is done). However as the RPC ApiServer is using its own HttpApiRequest class any framework can be used as long as it converts the HTTP request to a corresponding HttpApiRequest and calls the ApiServer.handleHttpApiRequest method.

The result of calling the handleHttpApiRequest method is returned as an HttpApiResponse which contains a stream with the encoded response or in the case of an error it contains the encoded JSON error as well as the exception thrown internally.

Errors

There are a couple of predefined error classes that can be used to return an error from the server to the client. They are:

• any RpcError(HTTP status code, Error name, Any message)
• 400 BadRequestError('You sent some data we don't understand.')
• 404 NotFoundError('We didn't find the api or method you are looking for.')
• 500 ApplicationError('The invoked method failed with an exception.')
• 500 Some internal exception occurred and it was not due to a method invocation.

If one of the above exceptions are thrown by the server API implementation it will be sent back as a serialized json response as described below. Any other exception thrown will be wrapped in the ApplicationError exception containing the toString() version of the internal exception as the method.

The JSON format for errors is:

{
  "error": {
    "code": <http status code>,
    "message": <error message>
  }
}      

In addition to the basic way of returning an http status code and an error message, you can attach RpcErrorDetail objects to your RpcError (as specified in the Google JSON style guide):

throw new RpcError(403, 'InvalidUser', 'User does not exist')
    ..errors.add(new RpcErrorDetail(reason: 'UserDoesNotExist'));

This will return the JSON:

{
  "error": {
    "code": 403,
    "message": "User does not exist",
    "errors": [
      {"reason": "UserDoesNotExist"}
    ]
  }
}      

Generating client stub code

Once your server API is written you can generate a Discovery Document describing the API and use it to generate a client stub library to call the server from your client.

There are two ways to generate a Discovery Document from your server API.

• Use the rpc:generate script to generate it from the commandline
• Retrieve it from a running server instance

Using the rpc:generate script you can generate a Discovery Document by running the script on the file where you put the class annotated with @ApiClass. Assuming your @ApiClass class is in a file 'lib/server/cloudapi.dart' you would write:

pub global activate rpc
cd <your package directory>
mkdir json
pub global run rpc:generate discovery -i lib/server/cloudapi.dart > json/cloud.json

In order for the rpc:generate script to work the API class (@ApiClass class) must have a default constructor taking no required arguments.

The other way to retrive a Discovery Document if from a running server instance. This requires the Discovery Service to be enabled. This is done by calling the ApiServer.enableDiscoveryApi() method on the ApiServer, see Example. for details.

After enabling the Discovery Service deploy the server and download the Discovery Document. For example if we have the 'cloud' API from the above example the Discovery Document can be retrieved from the deployed server by:

URL='https://your_app_server/discovery/v1/apis/cloud/v1/rest'
mkdir json
curl -o json/cloud.json $URL

Once you have the Discovery Document you can generate a client stub library using a Discovery Document client API generator. For Dart we have the Discovery API Client Generator. Discovery Document generators for other languages can also be used to call your API from e.g Python or Java.

If you want to generate a standalone client library for calling your server do:

pub global activate discoveryapis_generator
pub global run discoveryapis_generator:generate package -i json -o client

This will create a new Dart package with generated client stubs for calling each of your API methods. The generated library can be used like any of the other Google Client API libraries, some samples here.

If you want to generate a client stub code that should be integrated into an existing client you can instead do:

pub global activate discoveryapis_generator
pub global run discoveryapis_generator:generate files -i json -o <path to existing client package>

This will just generate a file in the directory specified by the '-o' option. NOTE: you might have to modify the existing client's pubspec.yaml file to include the packages required by the generated client stub code.