Skip to main content

Notice: this Wiki will be going read only early in 2024 and edits will no longer be possible. Please see: https://gitlab.eclipse.org/eclipsefdn/helpdesk/-/wikis/Wiki-shutdown-plan for the plan.

Jump to: navigation, search

Tutorial: Creating Custom Distribution Providers


Introduction

The ECF project provides an implementation of the OSGi R6 Remote Services and Remote Service Admin specifications. The RSA specification defines two major subsystems: discovery and distribution. Discovery concerns finding remote services exported by other processes on the network. The distribution subsystem is responsible for the actual communication of invoking a remote call: serializing remote method parameters, communicating with the remote service host via some network transport protocol, unmarshalling and invoking the service method with provided parameters, and returning a result to the consumer.

ECF's implementation of RSA defines an API to create new distribution providers. This API is declared in the ECF remote services API, provided by the org.eclipse.ecf.remoteservices bundle. Custom distribution providers implement a portion of this API and then will be used at runtime to supply the necessary functions of the distribution provider.

This tutorial will describe the creation of a simple custom distribution provider using the relevant portions of the ECF remote services API.

Remote Service Containers, IDs, and Namespaces

ECF has the concept of a 'container' (IContainer), which is an object instance that implements the remote services API and represents a network-accessible endpoint.

Containers have unique transport-specific ID. Some examples of transport-specific container IDs:

https://myhost.com/v1/path/to/service
ecftcp://localhost:3282/server
mqtt://mybroker.com/mytopic
jms:tcp://jmsbroker.com:6686/jmstopic
r_osgi://somehost/
hazelcast:///mytopicname
g6YuiWAjkk34je1lJlNmv==
uuid:de305d54-75b4-431b-adb2-eb6b9e546014
241

Each ID instance must be unique within a Namespace. Distribution providers must define a new Namespace that enforces the expected syntax requirements of the endpoint identifier.

Step 1: Namespace

The first thing a distribution provider must do is to register a new type of Namespace. ECF provides a number of Namespace classes that can be extended to make this easy. For example, the URIIDNamespace can handle any ID syntax that can be represented as a URI, like all of the above. Here is an example Namespace class that extends the URIIDNamespace:

public class Example1Namespace extends URIIDNamespace {
 
	private static final long serialVersionUID = 2460015768559081873L;
 
	public static final String NAME = "ecf.example1.namespace";
	private static final String SCHEME = "ecf.example1";
	private static Example1Namespace INSTANCE;
 
	/**
	 * The singleton instance of this namespace is created (and registered
	 * as a Namespace service) in the Activator class for this bundle.
	 * The singleton INSTANCE may then be used by both server and client.
	 */
	public Example1Namespace() {
		super(NAME, "Example 1 Namespace");
		INSTANCE = this;
	}
 
	public static Example1Namespace getInstance() {
		return INSTANCE;
	}
 
	@Override
	public String getScheme() {
		return SCHEME;
	}
}

In bundle Activator start, the singleton Example1Namespace is created and registered:

// Create and register the Example1Namespace
context.registerService(org.eclipse.ecf.core.identity.Namespace.class, new Example1Namespace(),  null);

The same Example1Namespace type must be used by both servers and clients, and so it typically makes sense to define the Namespace in a small common bundle, which can be deployed on both servers and clients. For the source code for this example, see the bundle here.

Note that other type of ID syntax can be easily supported by either inheriting from other Namespace classes (e.g. LongIDNamespace, StringIDNamespace, UUIDNamespace, etc.), or creating one's own Namespace subclass.

Step 2: Server

In addition to registering a new Namespace, distribution providers must implement and register the IRemoteServiceDistributionProvider interface using the whiteboard pattern.

For example, here is the code for registering a new custom distribution provider server

public static final String SERVER_ID_PARAMETER = "id";
public static final String SERVER_ID_PARAMETER_DEFAULT = "tcp://localhost:3333";
// register this remote service distribution provider
context.registerService(IRemoteServiceDistributionProvider.class,
   new RemoteServiceDistributionProvider.Builder().setName(ProviderConstants.SERVER_PROVIDER_CONFIG_TYPE)
      .setInstantiator(new RemoteServiceContainerInstantiator(ProviderConstants.SERVER_PROVIDER_CONFIG_TYPE,
         ProviderConstants.CLIENT_PROVIDER_CONFIG_TYPE) {
         @Override
         public IContainer createInstance(ContainerTypeDescription description,Map<String, ?> parameters) {
            // Create and configure an instance of our server 
            // container type
            return new Example1ServerContainer(getIDParameterValue(Example1Namespace.getInstance(), parameters,
				SERVER_ID_PARAMETER, SERVER_ID_PARAMETER_DEFAULT));
         }
   }).build(),	
null);

Notes for the above code

The RemoteServiceDistributionProvider.Builder class is used to create the IRemoteServiceDistributionProvider instance.

The setName(ProviderConstants.SERVER_PROVIDER_CONFIG_TYPE) method on the Builder sets the distribution provider's name (ProviderConstants.SERVER_PROVIDER_CONFIG_TYPE==ecf.example1.provider.dist.server).

This name corresponds to the OSGi Remote Service config type, and must uniquely identify the distribution provider. When a remote service is registered, the standard service property service.exported.configs may be specified so that the remote service will be exported via this distribution provider rather than some other distribution provider. For example:

Hashtable serviceProperties = new Hashtable();
serviceProperties.put("service.exported.interfaces","*");
serviceProperties.put("service.exported.configs","ecf.example1.provider.dist.server");
...
context.registerService(MyService.class,serviceInstance,serviceProperties);

The underlying ECF RSA implementation will export the serviceInstance via the ecf.example2.provider.dist.server distribution provider.

NOTE: As above, the IRemoteServiceDistributionProvider should be registered before any remote services using these distribution providers are exported. In other words, if you create and register a IRemoteServiceDistributionProvider with name 'ecf.example1.provider.dist.server' the provider implementation bundle should be started and the IRemoteServiceDistributionProvider service should be registered prior to registering the service to be exported using that provider. If the distribution provider is not registered prior to the remote service export, then the correct distribution provider will not be available to do the export.

The other required part of the distribution provider is the container instantiator:

      .setInstantiator(new RemoteServiceContainerInstantiator(ProviderConstants.SERVER_PROVIDER_CONFIG_TYPE,
         ProviderConstants.CLIENT_PROVIDER_CONFIG_TYPE) {
         @Override
         public IContainer createInstance(ContainerTypeDescription description,Map<String, ?> parameters) {
            // Create and configure an instance of our server 
            // container type
            return new Example1ServerContainer(getIDParameterValue(Example1Namespace.getInstance(), parameters,
				SERVER_ID_PARAMETER, SERVER_ID_PARAMETER_DEFAULT));
         }
   })

This code uses the RemoteServiceContainerInstantiator class. The two constants in the constructor (ProviderConstants.SERVER_PROVIDER_CONFIG_TYPE,ProviderConstants.CLIENT_PROVIDER_CONFIG_TYPE) sets up the instantiator so that these two config types (server and client for distribution provider) are associated with each other so at remote service consumer import time the ECF RSA implementation can select the appropriate client config type associated with the exporting server config type. For example, when ecf.example1.provider.dist.server is used as the exporting config type, then on the client/consumer the selected config type will be: ecf.example1.provider.dist.client
The createInstance method then creates an instance of the appropriate container type. This method allows the container to be created and configured based upon a Map of parameters. As per the RSA specification, if the remote service registration has service properties with keys of the form:
<config type>.<param>
then the provided parameters Map will contain entries of the form
key: <param>=<value>
For example, if the remote service is registered:

Hashtable serviceProperties = new Hashtable();
serviceProperties.put("service.exported.interfaces","*");
serviceProperties.put("service.exported.configs","ecf.example1.provider.dist.server");
serviceProperties.put("ecf.example1.provider.dist.server.id","tcp://localhost:3333");
...
context.registerService(MyService.class,serviceInstance,serviceProperties);

then the parameters Map will contain an entry with key="id" and value="tcp://localhost:3333". Such values can be used to configure the provider. In the case of the Example1 provider above, the id parameter is used to create an ID via the call to getIDParameterValue. Other distribution providers can define arbitrary parameters used to configure the underlying transport such as host, port, bind address, path, etc.
At the appropriate time during export, the createInstance method will be called and the Example1ServerContainer instance created. Here is the implementation of Example1Container

/**
 * Instances of this class are created via the container instantiator specified
 * by the RemoteServiceDistributionProvider whiteboard service in the Activator.
 * @see Activator#start(org.osgi.framework.BundleContext)
 * 
 */
public class Example1ServerContainer extends AbstractRSAContainer {
 
	/**
	 * Create an Example1ServerContainer.  The given id must not be null,
	 * and should be created using the Example1Namespace.  The server's ID
	 * will then be automatically used to provide the ecf.endpoint.id to
	 * remote service clients via the RSA-created EndpointDescription.
	 * @param id
	 */
	public Example1ServerContainer(ID id) {
		super(id);
	}
 
 
	@Override
	/**
	 * When the ECF RSA impl is requested to export a remote service and 
	 * this container instance is selected, a remote service registration will be
	 * created by RSA and this method will then be called to actually export the given
	 * registration via some communications transport.  This should trigger the appropriate 
	 * networking initialization (e.g. opening listener on socket), given the info in the
	 * registration.
	 * The given registration will not be <code>null</code>.
	 * If the transport would like to insert properties into the EndpointDescription
	 * for this endpoint, a Map of name (String) -> Object map should be returned.  All
	 * the values in the Map should be Serializable.   This provides a mechanism for
	 * distribution providers to include arbitrary private properties for use by 
	 * clients.
	 * 
	 * Note that keys for the Map should be unique to avoid conflicting with any other properties.
	 * 
	 * If values are provided for either OSGI RemoteConstants or ECF Remote Services Constants
	 * then these new values will <b>override</b> the default values in the EndpointDescription.
	 *
	 */
	protected Map<String, Object> exportRemoteService(RSARemoteServiceRegistration registration) {
		// Here would be the code to export/make a remote service on the actual transport
                // For example, starting a listener to handle connections, and responding to
                // connection attempts, handling remote calls, etc.
                // the registration parameter contains the services interfaces, the implementation, and the
                // service properties passed in when the remote service is registered
                // This method will be called by the same thread that calls BundleContext.registerService
		return null;
	}
 
	@Override
	/**
	 * When this remote service is unregistered, this method will be called by ECF RSA
	 * to allow the underlying transport to unregister the remote service given
	 * by the registration.  Necessary clean-up/shutdown of transport should be completed
	 * before returning.
	 */
	protected void unexportRemoteService(RSARemoteServiceRegistration registration) {
		// This method will be called when the service registration unregister() 
                // method is called, and is responsible for cleaning up/shutting down
                // and network resources associated with this remote service
                // This method will be called by the same thread that is calling
                // ServiceRegistration.unregister()
	}
}

Once this container instance is created the ECF RSA implementation will then call exportRemoteService(RSARemoteServiceRegistration registration) and expect that upon successful completion that the distribution provider will have made the remote service described by the given registration will have been made available for remote access. The exportRemoteService implementation is therefore where the distribution provider should actually make the remote service available over the network (e.g. by listening on socket, joining communication group, etc). When the remote service is unregistered, the unexportRemoteService(RSARemoteServiceRegistration registration) will be called to allow the transport to close and clean up any network resources associated with the previous export.
Each method will be called for each remote service export, with a unique RSARemoteServiceRegistration passed in by the ECF RSA implementation.
For the exportRemoteService method, the implementation can return a Map of service properties (or null). If a Map instance is returned, the values in the given map will be merged with the other endpoint description properties defined/required by ECF's RSA implementation. For example, one required ECF remote service property is ecf.endpoint.id. The default value of the ecf.endpoint.id is the exporting container's ID...e.g. by default
ecf.endpoint.id will be set to: "tcp://localhost:3333" because the Example1Container's ID is set by construction to "tcp://localhost:3333". However, if the returned Map is non null, and has the key ecf.endpoint.id, the value in the returned map will override the default value. e.g.

	protected Map<String, Object> exportRemoteService(RSARemoteServiceRegistration registration) {
		// TODO setup transport and export given remote service registration
                Map<String,Object> result = new HashMap<String,Object>();
                // This will override the ecf.endpoint.id value in the RSA endpoint description
                result.put("ecf.endpoint.id","http://somehost:2222/foo/");
                // This will add a brand new property to the RSA endpoint description
                result.put("my.new.property","howdy");
		return result;
	}

The complete source for the distribution provider server is available here.

Step 3: Client

Like the server, the distribution provider client is registered via the whiteboard pattern using the IRemoteServiceDistributionProvider service interface. Here is the code for registering the Example1 client distribution provider

context.registerService(IRemoteServiceDistributionProvider.class,
	new RemoteServiceDistributionProvider.Builder().setName(ProviderConstants.CLIENT_PROVIDER_CONFIG_TYPE)
		.setInstantiator(
			new RemoteServiceContainerInstantiator(ProviderConstants.SERVER_PROVIDER_CONFIG_TYPE,
				ProviderConstants.CLIENT_PROVIDER_CONFIG_TYPE) {
				@Override
				public IContainer createInstance(ContainerTypeDescription description,
								Map<String, ?> parameters) {
				        // Create and configure an instance of our client
				        // container type (below)
					// This is called by RSA when a client container of this type
					// is needed (e.g. to import a discovered remote service)
					return new Example1ClientContainer();
				}
			}).build(),
null);

As with the server, the setName and setInstantiator builder methods are required. For this distribution provider all that's necessary is to return a new Example1ClientContainer which is declared

public class Example1ClientContainer extends AbstractRSAClientContainer {
 
	public Example1ClientContainer() {
		super(Example1Namespace.getInstance()
				.createInstance(new Object[] { "uuid:" + java.util.UUID.randomUUID().toString() }));
	}
 
	@Override
	// This is called when when a remote service instance is actually needed by consumers
	// The remote service getProxy method will be called the first time the proxy is 
	// to be created for a given consumer
	protected IRemoteService createRemoteService(RemoteServiceClientRegistration registration) {
		return new AbstractRSAClientService(this, registration) {
 
			@Override
			public Object getProxy(ClassLoader cl, @SuppressWarnings("rawtypes") Class[] interfaces) throws ECFException {
				return super.getProxy(cl, interfaces);
			}
 
			@Override
			protected Object invokeAsync(RSARemoteCall remoteCall) throws ECFException {
				// TODO Auto-generated method stub
				return null;
			}
			// invokeSync will be called when a client needs to make a remote call on 
			// one of the proxy methods.
			@Override
			protected Object invokeSync(RSARemoteCall remoteCall) throws ECFException {
				// TODO Auto-generated method stub
				return null;
			}
		};
	}
}

When the ECF RSA discovers a remote service and attempts to import it, the createRemoteService method is called with the associated RemoteServiceClientRegistration provided.
In the above case an instance of AbstractRSAClientService is created for the import. Sometime later when the proxy is actually required by a consumer, the getProxy method will be called. The code above overrides the implementation of getProxy to show that this process can be easily customized for a distribution provider, but in most cases the AbstractRSAClientService proxy creation can be used.
The invokeSync and invokeAsync method are called when the proxy methods are actually invoked by the consumer. The invokeSync method is called when the proxy methods are called synchronously (normal case), and the invokeAsync is called when the ECF async proxy methods are invoked. For the invokeSync method, the distribution provider should implement this method to make the appropriate remote call by serializing remote call arguments, sending over the distribution provider's transport, getting and deserializing the result, and returning the Object that represents the appropriate type for proxy method defined in the RSARemoteCall instance. In invokeSync the calling thread should block until the remote call is made and the response is received, an exception occurs, or any timeout occurs.
The complete source for the distribution provider client is available here.

Background and Related Articles

Getting Started with ECF's OSGi Remote Services Implementation

OSGi Remote Services and ECF

Asynchronous Proxies for Remote Services

Static File-based Discovery of Remote Service Endpoints

Download ECF Remote Services/RSA Implementation

How to Add Remote Services/RSA to Your Target Platform

Back to the top