JGroups uses a JChannel as the main API to connect to a cluster, send and receive messages, and to register listeners that are called when things (such as member joins) happen.

What is sent around are Messages, which contain a byte buffer (the payload), plus the sender's and receiver's address. Addresses are subclasses of org.jgroups.Address, and usually contain an IP address plus a port.

The list of instances in a cluster is called a view (org.jgroups.View), and every instance contains exactly the same View. The list of the addresses of all instances can get retrieved by calling View.getMembers().

Instances can only send or receive messages when they've joined a cluster.

When an instance wants to leave the cluster, methods JChannel.disconnect() or JChannel.close() can be called. The latter actually calls disconnect() if the channel is still connected before closing the channel.

To join a cluster, we'll use a JChannel. An instance of JChannel is created with a configuration (e.g. an XML file) which defines the properties of the channel. To actually connect to the cluster, the connect(String name) method is used. All channel instances which call connect() with the same argument will join the same cluster. So, let's actually create a JChannel and connect to a cluster called "ChatCluster":

import org.jgroups.JChannel;

public class SimpleChat {
    JChannel channel;
    String user_name=System.getProperty("user.name", "n/a");

    private void start() throws Exception {
        channel=new JChannel();
        channel.connect("ChatCluster");
    }

    public static void main(String[] args) throws Exception {
        new SimpleChat().start();
    }
}
        

First we create a channel using the empty contructor. This configures the channel with the default properties. Alternatively, we could pass an XML file to configure the channel, e.g. new JChannel("/home/bela/udp.xml").

The connect() method joins cluster "ChatCluster". Note that we don't need to explicitly create a cluster beforehand; connect() creates the cluster if it is the first instance. All instances which join the same cluster will be in the same cluster (of course!), for example if we have

, then we will have 3 clusters: "cluster-one" with instances ch1 and ch4, "cluster-two" with ch2 and ch3, and "cluster-three" with only ch5.

We now run an event loop, which reads input from stdin ('a message') and sends it to all instances currently in the cluster. When "exit" or "quit" quit are entered, we fall out of the loop and close the channel.

    private void start() throws Exception {
        channel=new JChannel();
        channel.connect("ChatCluster");
        eventLoop();
        channel.close();
    }

    private void eventLoop() {
        BufferedReader in=new BufferedReader(new InputStreamReader(System.in));
        while(true) {
            try {
                System.out.print("> "); System.out.flush();
                String line=in.readLine().toLowerCase();
                if(line.startsWith("quit") || line.startsWith("exit")) {
                    break;
                }
                line="[" + user_name + "] " + line;
                Message msg=new Message(null, null, line);
                channel.send(msg);
            }
            catch(Exception e) {
            }
        }
    }
        

We added the call to eventLoop() and the closing of the channel to the start() method, and we provided an implementation of eventLoop.

The event loop blocks until a new line is ready (from standard input), then sends a message to the cluster. This is done by creating a new Message and calling Channel.send() with it as argument.

The first argument of the Message constructor is the destination address. A null destination address means send the message to everyone in the cluster (a non-null address of an instance would send a message from us to only 1 instance).

The second argument is our own address. This is null as well, as the stack will insert the correct address anyway.

The third argument is the line that we read from stdin, this uses Java serialization to create a byte[] buffer and set the message's payload to it. Note that we could also serialize the object ourselves (which is actually the recommended way !) and use the Message contructor which takes a byte[] buffer as third argument.

The application is now fully functional, except that we don't yet receive messages or view notifications. This is done in the next section below.

Let's now register as a Receiver to receive message and view changes. To this end, we could implement org.jgroups.Receiver (with 6 methods), however, I chose to extend ReceiverAdapter which has default implementations, and only override callbacks (receive() and viewChange()) we're interested in. We now need to extend ReceiverAdapter:

public class SimpleChat extends ReceiverAdapter {
            

, set the receiver in start():

    private void start() throws Exception {
        channel=new JChannel();
        channel.setReceiver(this);
        channel.connect("ChatCluster");
        eventLoop();
        channel.close();
    }
            

, and implement receive() and viewAccepted():

   public void viewAccepted(View new_view) {
        System.out.println("** view: " + new_view);
    }

    public void receive(Message msg) {
        System.out.println(msg.getSrc() + ": " + msg.getObject());
    }
            

The viewAccepted() callback is called whenever a new instance joins the cluster, or an existing instance leaves (crashes included). Its toString() method prints out the view ID (an increasing ID) and a list of the current instances in the cluster

In receive(), we get a Message as argument. We simply get its buffer as an object (again using Java serialization) and print it to stdout. We also print the sender's address (Message.getSrc()).

Note that we could also get the byte[] buffer (the payload) by calling Message.getBuffer() and then de-serializing it ourselves, e.g. String line=new String(msg.getBuffer()).

Now that the demo chat application is fully functional, let's try it out. Start an instances of SimpleChat:

                [mac] /Users/bela$ java SimpleChat
                -------------------------------------------------------
                GMS: address is 192.168.0.6:49963
                -------------------------------------------------------
                ** view: [192.168.0.6:49963|0] [192.168.0.6:49963]
                >
            

The address of this instance is 192.168.0.6:49963 (IP address:port). It is the only instance so far. So let's start the second instance and type something:

                [mac] /Users/bela$ java SimpleChat
                -------------------------------------------------------
                GMS: address is 192.168.0.6:49964
                -------------------------------------------------------
                ** view: [192.168.0.6:49963|1] [192.168.0.6:49963, 192.168.0.6:49964]
                >
            

The cluster list is now [192.168.0.6:49963, 192.168.0.6:49964], showing the first and second instance that joined the cluster. Note that the first instance (192.168.0.6:49963) also received the same view, so both instances have the exact same view with the same ordering of its instances in the list. The instances are listed in order of joining the cluster, with the oldest instance as first element.

Sending messages is now as simple as typing a message after the prompt and pressing return. The message will be sent to the cluster and therefore it will be received by both instances, including the sender.

If the word "exit" or "quit" is entered, then the instance will leave the cluster gracefully. This means, a new view will be installed immediately.

To simulate a crash, simply kill an instance (e.g. via CTRL-C, or from the process manager). The other surviving instance will receive a new view, with only 1 instance (itself) and excluding the crashed instance.

One of the uses of JGroups is for maintaining state that is replicated across a cluster. For example, state could be all the HTTP sessions in a web server. If those sessions are replicated across a cluster, then clients can access any server in the cluster after a server which hosted the client's session crashed, and the user sessions will still be available.

Any update to a session is replicated across the cluster, e.g. by serializing the attribute that was modified and sending the modification to every server in the cluster via JChannel.send(). This is needed so that all servers have the same state.

However, what happens when a new server is started ? That server has to somehow get the existing state (e.g. all HTTP sessions) from an existing server in the cluster. This is called state transfer.

State transfer in JGroups is done by implementing 2 (getState() and setState()) callbacks and calling the JChannel.getState() method. method. Note that, in order to be able to use state transfer in an application, the protocol stack has to have a state transfer protocol (the default stack used by the demo app does).

The start() method is now modified to include the call to JChannel.getState():

    private void start() throws Exception {
        channel=new JChannel();
        channel.setReceiver(this);
        channel.connect("ChatCluster");
        channel.getState(null, 10000);
        eventLoop();
        channel.close();
    }
            

The getState() method actually returns a boolean, which is false for the first instance in a cluster, and should be true for subsequent instances.

The Receiver interface defines a callback getState() which is called on an existing instance to fetch the cluster state. In our demo application, we define the state to be the chat conversation. This is a simple list, to the tail of which we add every message we receive. (Note that this is probably not the best example for state, as this state always grows. As a workaround, we could have a bounded list, which is not done here though).

The list is defined as an instance variable:

    final List<String> state=new LinkedList<String>();
            

The getState() callback implementation is

    public byte[] getState() {
        synchronized(state) {
            try {
                return Util.objectToByteBuffer(state);
            }
            catch(Exception e) {
                e.printStackTrace();
                return null;
            }
        }
    }
            

The getState() method is called in the state provider, ie. an existing instance, to return the shared cluster state.

Since access to state may be concurrent, we synchronize it. Then we call Util.objectToByteBuffer() which is a JGroups utility method using simple serialization to generate a byte buffer from an object.

The setState() method is called on the state requester, ie. the instance which called JChannel.getState(). Its task is to deserialize the byte buffer and set its state accordingly:

    public void setState(byte[] new_state) {
        try {
            List<String> list=(List<String>)Util.objectFromByteBuffer(new_state);
            synchronized(state) {
                state.clear();
                state.addAll(list);
            }
            System.out.println("received state (" + list.size() + " messages in chat history):");
            for(String str: list) {
                System.out.println(str);
            }
        }
        catch(Exception e) {
            e.printStackTrace();
        }
    }
            

We again call a JGroups utility method (Util.objectFromByteBuffer()) to create an object from a byte buffer (using Java serialization).

Then we synchronize on state, and set its contents from the received state.

We also print the number of messages in the received chat history to stdout. Note that this is not feasible with a large chat history, but - again - we could have a bounded chat history list.

In this tutorial, we showed how to create a channel, join and leave a cluster, send and receive messages, get notified of view changes and implement state transfer. This is the core functionality provided by JGroups through the JChannel and Receiver APIs.

JGroups has two more areas that weren't covered: building blocks and the protocol stack.

Building blocks are classes residing on top of a JChannel and provide a higher abstraction level, e.g. request-response correlators, cluster-wide method calls, replicated hashmaps and so forth.

The protocol stack allows for complete customization of JGroups: protocols can be configured, removed, replaced, enhanced, or new protocols can be written and added to the stack.

We'll cover the protocol stack and available protocols in a later article.

The code for SimpleChat can be found here.

Here are some links for further information about JGroups: