Spring Cloud Sidecar - Initialization of Nodes

In the last blog post I had described how the Sidecar application can be used for registering the Cassandra nodes with Eureka and more generally can be used for registering any non-JVM application with Eureka.

In this post I will cover how an application can go about querying the Sidecar registered nodes.

Discovering Registered Nodes - Post Initialization

If the registered nodes are not required during the bean initialization phase then discovering the nodes is fairly straightforward along these lines:

@Component
public class SampleCommandLineRunner implements CommandLineRunner {

    @Autowired
    private DiscoveryClient discoveryClient;

    @PostConstruct
    public void postConstruct() {
//        System.out.println("Printing from postConstruct");
//        printDiscoveredNodes();
    }

    @Override
    public void run(String... strings) throws Exception {
        System.out.println("Printing from run method");
        printDiscoveredNodes();
    }

    public void printDiscoveredNodes() {
        System.out.println(" Printing Discovered Nodes ");

        for (ServiceInstance instance: discoveryClient.getInstances("samplecassandra.vip")) {
            System.out.println("Host: Port = " + instance.getHost() + ":" + instance.getPort());
        }
    }
}

These would print the nodes registered with a name of "samplecasssandra.vip" VIP.

Note that the nodes are being printed from the run method which gets called past the initialization of Spring container. If however the nodes were attempted to be listed from one of the lifecycle stages say the postConstruct method then very likely an exception will be thrown (this behavior is seen with "Angel.SR3" release of Spring Cloud, but appears to work cleanly with "Brixton.*" versions)

Discovering Registered Nodes - During Initialization

Now if an application needs to discover the nodes during initialization the flow is a little more complicated, for a potential issue look at this ticket.

The DiscoveryClient is initialized very late in the Spring Lifecycle and if DiscoveryClient is used in any post-processing activity of a bean it is likely to give an exception.

As an example, say the Cassandra nodes registered using Sidecar is now used by an application to initialize Cassandra connectivity, a way to do it would be to create a wrapper around Cassandra connectivity this way:

import org.springframework.data.cassandra.core.CassandraTemplate;


public class CassandraTemplateWrapper extends CassandraTemplate {

    @Override
    public void afterPropertiesSet() {
        
    }
}

Here CassandraTemplate is being overridden to prevent the check in afterPropertiesSet method that a Cassandra session exists, as a session will be established much later in the start-up cycle.

A Cassandra session can be injected into this custom CassandraTemplate lazily in a bean that implements SmartLifecyle along these lines:

package mvctest.cassandra;

import com.datastax.driver.core.Cluster;
import com.datastax.driver.core.Session;
import org.slf4j.Logger;
import org.slf4j.LoggerFactory;
import org.springframework.beans.factory.FactoryBean;
import org.springframework.beans.factory.annotation.Autowired;
import org.springframework.context.SmartLifecycle;
import org.springframework.core.Ordered;
import org.springframework.data.cassandra.core.CassandraTemplate;
import org.springframework.stereotype.Component;

@Component("cassandraTemplate")
public class EurekaCassandraTemplateFactoryBean implements SmartLifecycle, FactoryBean<CassandraTemplate>, Ordered {

    ....


    @Override
    public boolean isAutoStartup() {
        return true;
    }

    @Override
    public void start() {
        LOGGER.info("About to start Discovery client lookup of Cassandra Cluster!!!");
        final Cluster cluster = this.eurekaClusterBuilder.build();
        Session session = cluster.connect(this.cassandraProperties.getKeyspace());
        this.cassandraTemplateWrapper.setSession(session);
        LOGGER.info("Completed Discovery client lookup of Cassandra Cluster!!!");
        running = true;
    }

    @Override
    public boolean isRunning() {
        return this.running;
    }

    @Override
    public int getPhase() {
        return Integer.MAX_VALUE;
    }

    @Override
    public int getOrder() {
        return 1;
    }
}

This way the Cassandra session can be created very late in the cycle. Somewhat rough, but the approach works.

If you are interested in exploring this sample further I have this code available in my github repo here.