RSocket LoadBalancing does not seem to work with Spring Boot.

Hi @OlegDokuka

I was looking at this example and it works just fine.

https://github.com/rsocket/rsocket-java/blob/master/rsocket-examples/src/main/java/io/rsocket/examples/transport/tcp/loadbalancer/RoundRobinRSocketLoadbalancerExample.java

I was trying to implement this with Spring Boot. However it seems to always use the first instance it connects to.

I have a project here. I also have provided the steps as part of README https://github.com/vinsguru/rsocket-load-balancing

Server-App:

• Navigate to server-app cd server-app
• Build the jar mvn clean package
• Run 2 instances of this app as shown below.

java -jar -Dspring.rsocket.server.port=6565 target/server-app-0.0.1-SNAPSHOT.jar
java -jar -Dspring.rsocket.server.port=6566 target/server-app-0.0.1-SNAPSHOT.jar

Client-app

• application.yaml - by default uses the above server instance details - localhost:6565 and localhost:6566

• Runt the main application.

• Check the server console output. Only one of the instance is processing all the requests!!

Does it require any other config?

[OlegDokuka]

@vinsguru the problem is still in configuration

private List<LoadbalanceTarget> toLoadBalanceTarget(List<RSocketServer> rSocketServers){
        System.out.println(rSocketServers);
        return rSocketServers.stream()
                .map(server -> LoadbalanceTarget.from(server.getHost() + server.getPort(), TcpClientTransport.create(server.getHost(), server.getPort())))
                .collect(Collectors.toList());
    }

As it was mentioned in the stackoverflow answer, the LoadbalanceTarget has 1 important parameter which is key using key the RSocketPool is capable of making difference between given targets and identify which target is new and which has already been there. In your setup, you always use the localhost as a key, hence for RSocketPool all the new LoadbalancerTargets seemed to be the same as before, and therefore you observed traffic to only a single target.

Also, please note, that if one of the targets disappears, your connection will be immediately disposed (that is something that we have to improve) which means if the connection is still alive and has ongoing traffic, that traffic will be immediately interrupted and your request will be signaled with onError. To fix that, I suggest putting extra retry logic. For example:

    @Override
    public void run(String... args) throws Exception {
        Flux.range(1, 10000)
            .delayElements(Duration.ofMillis(10))
            .flatMap(i -> rSocketRequester.route("square-calculator").data(i).retrieveMono(Integer.class).retry())
            .doOnNext(i -> System.out.println("Response : " + i))
            .blockLast();
    }

[OlegDokuka]

You are welcome @vinsguru. And no worries!