6 Replies Latest reply on Apr 26, 2004 6:00 PM by lordbritish

Round robin clustering for HA-RMI on 2 nodes is not even !!!

lordbritish Apr 16, 2004 9:30 PM

I've created a cluster on 2 nodes using the round robin clustering policy for a HA-RMI service.

According to the clustering docs, this should alternate the requests from one server to the other one in the cluster.

This is NOT happening.

Quite frequently, one of the nodes is favored and all requests (for a while) go to it. Then inexplicably, the requests are balanced among the 2 nodes for a little while. And then later, we're back to having all the request go to the same "favored" server.

It's not behaving as advertised.

Please fix this, thx.

This is with JBoss 3.2.3

1. Re: Round robin clustering for HA-RMI on 2 nodes is not even

lordbritish Apr 20, 2004 10:21 PM (in response to lordbritish)
This is my HA-RMI service below:

InitialContext ctx = new InitialContext(); Object obj = ctx.lookup("/HAPartition/DefaultPartition"); RMISocketFactory rmiSocketFactry = RMISocketFactory.getSocketFactory(); HAPartition myPartition = (HAPartition) obj; MyService myService = new MyServiceImpl(); HARMIServerImpl rmiserver = new HARMIServerImpl(myPartition, "MyService", MyService.class, myService, 3433, rmiSocketFactry, rmiSocketFactry); MyService stub = (MyService) rmiserver.createHAStub(new RoundRobin()); ctx.bind("RMI_STUB", stub);

As you noticed I'm explicitly using "RoundRobin"

However, the 2 nodes in my cluster are not using "Round Robin" as I understand it. It's my understanding that "Round Robin" clustering means going from one sever to the next in an even manner (not factoring system load). I'm not sure what load balancing technique is actually being used but it aint "Round Robin" as I've explained it.

Fail-over is working - when I bring down one of my nodes, all request goes to the remaining node.

I believe other folks here have noticed this anomaly.

Is this a bug, or intended behvior ??
Actions
2. Re: Round robin clustering for HA-RMI on 2 nodes is not even

i_n_g Apr 26, 2004 1:36 PM (in response to lordbritish)

Can you paste the client code?
Actions

3. Re: Round robin clustering for HA-RMI on 2 nodes is not even

lordbritish Apr 26, 2004 4:16 PM (in response to lordbritish)

Here's what I'm doing with the client

Properties props = new Properties();
props.setProperty("java.naming.factory.initial", "org.jnp.interfaces.NamingContextFactory");
props.setProperty("java.naming.factory.url.pkgs", "jboss.naming:org.jnp.interfaces");
props.setProperty("java.naming.provider.url, "node1:2000,node2:2000");

InitialContext ctx = new InitialContext(props);
MyService myService = (MyService)ctx.lookup("RMI_STUB");
myService.helloWorld();

I get the stub every time I call the the remote method - perhaps that's the problem. May be I should be reusing the stub??

4. Re: Round robin clustering for HA-RMI on 2 nodes is not even

i_n_g Apr 26, 2004 5:56 PM (in response to lordbritish)

yeah, the HA-RMI only works when you call myService.helloWorld() repeatedly:

myService = ctx.lookup("RMI_STUB");
// get the remote stub from first available node

while (...) {
myService.helloWorld();
// the "smart-stub" will distribute the work to the clustering nodes
}
Actions
5. Re: Round robin clustering for HA-RMI on 2 nodes is not even

i_n_g Apr 26, 2004 5:57 PM (in response to lordbritish)

The "smart guy" is the "stub", not the namingservice. :-)
Actions
6. Re: Round robin clustering for HA-RMI on 2 nodes is not even

lordbritish Apr 26, 2004 6:00 PM (in response to lordbritish)

Oh, ok, thanks a $1,000,000 - i don't really have that but u know what I mean ;)
Actions

Go to original post