How to cluster embedded Cassandra instances?

Question

I've written a small application to start an embedded instance of Cassandra 1.2.

I'm trying to create a cluster of 3 of these embedded instances locally, by running 3 instances of this application. Each one looks at a different cassandra.yaml on the filesystem. Each file has:

• the same cluster name
• blank initial_token
• unique listen address (all mapped to 127.0.0.1 in my hosts file)
• unique rpc, storage and ssl_storage ports
• the same seed (the listen address (no port) of the first server)
• unique -Dcom.sun.management.jmxremote.port value passed in application launch

When I launch the applications, all come up fine, and have separate storage on the filesystem. However, when I use nodetool to inspect each one, each appears to be in a cluster by itself:

C:\Program Files\DataStax Community\apache-cassandra\bin>nodetool -h 127.0.0.1 -p 7197 ring
Starting NodeTool

Datacenter: datacenter1
==========
Replicas: 1

Address    Rack        Status State   Load            Owns                Token

127.0.0.1  rack1       Up     Normal  198,15 KB       100,00%             8219116491729144532


C:\Program Files\DataStax Community\apache-cassandra\bin>nodetool -h 127.0.0.2 -p 7198 ring
Starting NodeTool

Datacenter: datacenter1
==========
Replicas: 1

Address    Rack        Status State   Load            Owns                Token

127.0.0.2  rack1       Up     Normal  152,13 KB       100,00%             -3632227916915216562

Blogs and docs online suggest this should be sufficient. Is it possible to cluster embedded instances? If so, does anyone know how my configuration or understanding is incorrect/insufficient?

Code to launch the embedded instances is below. Hope you can help, thanks.

public class EmbeddedCassandraDemo {

    private static final String CONF_PATH_FORMAT = "D:\\embedded_cassandra\\Node%d\\";

    private ExecutorService executor = Executors.newSingleThreadExecutor();
    private CassandraDaemon cassandraDaemon;
    private int nodeNumber;

    public EmbeddedCassandraDemo(int nodeNumber) {
        this.nodeNumber = nodeNumber;
    }

    public static void main(String [ ] args) throws InterruptedException, ConnectionException {
        new EmbeddedCassandraDemo(Integer.parseInt(args[0])).run();
    }

    private void run() throws InterruptedException, ConnectionException {
        setProperties();

        activateDeamon();
    }

    private void activateDeamon() {
        executor.execute( new Runnable(){

            @Override
            public void run() {
                cassandraDaemon = new CassandraDaemon();
                cassandraDaemon.activate();
            }});
    }

    private void setProperties() {
        System.setProperty("cassandra.config", String.format("file:%scassandra.yaml", String.format(CONF_PATH_FORMAT, nodeNumber)));
        System.setProperty("log4j.configuration", String.format("file:%slog4j-server.properties", String.format(CONF_PATH_FORMAT, nodeNumber)));
        System.setProperty("cassandra-foreground", "true");
    }
}

Answer 1

"blank initial_token"

Are you using virtual nodes? If not, I wonder if that could be your issue. You should have each machine defined with a different initial token. For a 3-node cluster, those initial tokens should be increments of 56,713,727,820,156,410,577,229,101,238,628,035,242 apart from each other.

Using DataStax's Python script for computing initial tokens, these values should suit your needs:

node 0: 0
node 1: 56713727820156410577229101238628035242
node 2: 113427455640312821154458202477256070485

Also, which endpoint_snitch are you using? If you are using "PropertyFileSnitch" make sure that your cassandra-topology.properties file contains a definition for each node (along with DC and rack).

Give that a try and see if it helps.

Answer 2

Sorted now. Looked through the code and intro on https://github.com/pcmanus/ccm (linked to from Run multiple cassandra nodes (a cluster) from the same machine?), and that keeps all port values the same apart from that for JMX connection.

Having made those changes, plus setting the initial token on each thanks to @BryceAtNetwork23, and specifying the IPs of all 3 servers as seeds, they now form a cluster.