Using HA Singleton in JBoss 7

Some time ago I had to change standard clustering behavior of Quartz Scheduler, and let it work without synchronizing over database. There are of course a lot of options to do that, but because I'm big fan of simplicity I've decided to use standard Spring @Scheduled configuration and totally skip thinking about cluster at this level. The idea was to just invoke "check that I'm on master node" method at the beginning of the scheduled method. The only problem was how to write such method :) The choice was to use JBoss HA Singleton functionality. It's available in JBoss 7.x but the big lack of documentation forces some experiments... nice!

First thing which we need to do is to provide proper dependency, containing few important classes. Of course remember about right version (here I'm using 7.1.1.Final because it's available in public repositories and all next versions of 7 needs to be build manually).

  org.jboss.as
  jboss-as-clustering-singleton
  7.1.1.Final
  provided
Now it's time to do some coding! Let's start from class containing service that will be installed into application server. Its role is to set master status flag and expose it by simple static method.

package com.stackholder.jboss.ha;
import org.jboss.msc.service.*;
import org.slf4j.Logger;
import org.slf4j.LoggerFactory;
import java.io.Serializable;
import java.util.concurrent.atomic.AtomicBoolean;

public class MasterStatusHaSingleton 
    extends AbstractService<Serializable> {

  private static final Logger LOGGER = 
      LoggerFactory.getLogger(MasterStatusHaSingleton.class);

  private static AtomicBoolean masterStatus = 
      new AtomicBoolean(false);

  @Override
  public void start(StartContext startContext) {
    LOGGER.info("MasterStatusHaSingleton started");
    masterStatus.set(true);
  }

  @Override
  public void stop(StopContext stopContext) {
    LOGGER.info("MasterStatusHaSingleton stopped");
    masterStatus.set(false);
  }

  public static boolean isMaster() {
    return masterStatus.get();
  }
}
Now we need to write class that will install prepared service into container.
package com.stackholder.jboss.ha;

import org.jboss.as.clustering.singleton.SingletonService;
import org.jboss.msc.service.*;
import org.slf4j.Logger;
import org.slf4j.LoggerFactory;

public class HaSingletonActivator implements ServiceActivator {

  private final static Logger LOGGER =
      LoggerFactory.getLogger(HaSingletonActivator.class);

  public static final ServiceName SINGLETON_SERVICE_NAME =
      ServiceName.JBOSS.append("ha", "singleton");

  @Override
  public void activate(ServiceActivatorContext context) 
      throws ServiceRegistryException {
    LOGGER.info("HaSingletonActivator will be installed");

    MasterStatusHaSingleton srv = new MasterStatusHaSingleton();
    SingletonService singleton = 
        new SingletonService(srv, SINGLETON_SERVICE_NAME);

    singleton.build(new DelegatingServiceContainer(
        context.getServiceTarget(),context.getServiceRegistry()))
        .setInitialMode(ServiceController.Mode.ACTIVE).install();
    LOGGER.info("HaSingletonActivator installation SUCCESSFUL");
  }
}
Remember that JBoss 7 is using OSGi, so we've to tell application server what modules we're going to use in our application. The easiest way to do that is to add configuration into war or jar plugin:

  
    
      
        org.jboss.msc,org.jboss.as.server,
        org.jboss.as.clustering.singleton
      
    
  
Last thing we need to do in our application is to include our ServiceActivator in JBoss. To do that we need to create file org.jboss.msc.service.ServiceActivator in META-INF directory. Content is just one line with our ServiceActivator qualified name, which in my case means:
com.stackholder.jboss.ha.HaSingletonActivator
Great - almost finished! Almost :) Last thing we have to do is to activate proper modules in JBoss configuration. Just edit standalone-full-ha.xml file (or other configuration that you use of course) and add following modules into ee subsystem:

  
    <module name="org.jboss.msc" slot="main">
    <module name="org.jboss.as.clustering.singleton" slot="main">
  
And finally you can execute your server and enjoy new, cool functionality ;)

Comments

Mr. Feng said…
Thanks for the example! However, I am still missing one part - how would HaSingletonActivator be found and invoked? I have followed all your code and configuration but I do not see HaSingletonActivator is invoked, I must missed something fundamental. I am using WildFly 8.0.
StackHolder said…
I think that I have forgot some info. You have to create the file META-INF/services/org.jboss.msc.service.ServiceActivator
In that file you should add qualified name of your activator (in my case com.stackholder.jboss.ha.HaSingletonActivator). Please confirm that it works so I could extend by article.
Mr. Feng said…
Thanks for reply! Yes, that does make the hook!

However, I am getting another issue running the example now. The error is "Name segment is null for server" on singleton.build(new DelegatingServiceContainer( context.getServiceTarget(),context.getServiceRegistry())) .setInitialMode(ServiceController.Mode.ACTIVE).install();

You put a service name ServiceName.JBOSS.append("ha", "singleton"). Do I have to do some setup to use that name? Or, it could be anything I come up to use?

Thanks!
StackHolder said…
No, you don't have to configure this name. I've updated my post and now it works out of the box
Unknown said…
Hi,

I would like to know if you have any aqcuillian or test cases for an HA Singleton and it's ServiceActivator

Thanks

Piwe
StackHolder said…
Unfortunately no - I did just manual testing on two nodes in cluster. But if you'll do it in Arquillian please let me know :)
Unknown said…
Thanks for this post. I know it's been a while but do you have any idea if this would be applicable to Wildfly 10? I think they changed the way HA Singletons work. We are upgrading from JBoss 6 to Wildfly 10 and we need to know which instance is the master. Previously we used HASingletonController but it looks like that class is gone.

Popular posts from this blog

Understanding Spring Web Initialization

Overview of the circuit breaker in Hystrix

Do we really still need a 32-bit JVM?