Does the Cluster-level failover of pulsar have any requirements on the version of pulsar? Is version 2.10 or higher required? 2.9.4 Can it be used?

[tianshimoyi]

I would like to know if there is a version requirement for the Cluster-level failover function of pulsar? What is the minimum supported version?

[merlimat]

Auto cluster failover was introduced in the 2.10 release. It's a client-side feature, so you just need to have the client library to 2.10+, but you can use it with older brokers.

[tianshimoyi]

@merlimat I tried to test with pulsar version 2.9.4, client library version 2.10.3, but it didn't work!
producer code

public class PulsarTest {

    public static void main(String[] args) {

        String primaryUrl = "pulsar://10.0.87.203:6650";
        String secondaryUrl = "pulsar://10.0.88.116:6650";
        Map<String, Authentication> m = new HashMap<>();
        m.put(secondaryUrl, AuthenticationFactory.token("xxxxx"));
        ServiceUrlProvider failover = AutoClusterFailover.builder()
                .primary(primaryUrl)
                .secondary(List.of(secondaryUrl))
                .failoverDelay(10, TimeUnit.SECONDS)
                .switchBackDelay(30, TimeUnit.SECONDS)
                .checkInterval(1000, TimeUnit.MILLISECONDS)
                .secondaryAuthentication(m)
                .build();
        try {
            PulsarClient client = PulsarClient.builder()
                    .serviceUrlProvider(failover)
                    .authentication(AuthenticationFactory.token("xxxxx"))
                    .build();
            Producer<String> producer = client.newProducer(Schema.STRING)
                    .topic("public/default/t1")
                    .create();
            long index = 0;
            while (true) {
                String msg = new String("hello-" + index);
                producer.send(msg);
                Thread.sleep(1000);
                index = index + 1;
                System.out.println(msg + " send ok");
            }
        } catch (Exception e) {
            System.out.println("exit");
            e.printStackTrace();
        }
    }
}

Then I shut down all the Broker nodes of the main cluster, and no switchover occurred. Is there something wrong with the code? Or Broker requires special configuration ？

[tisonkun]

@tianshimoyi

1. You should desensitize ip address and token when share it online - never paste your private token.
2. What do you observe "no switchover occurred"? Generally, the switch happened after the delay you configure elapse, instead of immediately.

[hezhangjian]

@tianshimoyi Can you provide details log during the tests?

[tianshimoyi]

@shoothzj @tisonkun Hi, My steps are as follows:

1. Open the producer consumption message
2. Turn off all promary broker nodes
   Then it was found that there was no master-slave switchover, and the failure time had exceeded failoverDelay=10s

producer code:

public class PulsarTest {

    public static void main(String[] args) {
        String primaryUrl = "pulsar://10.0.68.235:6650";
        String secondaryUrl = "pulsar://10.0.64.134:6650";
        Map<String,Authentication> m = new HashMap<>();
        m.put(secondaryUrl,AuthenticationFactory.token("xxxx"));
        ServiceUrlProvider failover = AutoClusterFailover.builder()
                .primary(primaryUrl)
                .secondary(List.of(secondaryUrl))
                .failoverDelay(10, TimeUnit.SECONDS)
                .switchBackDelay(30, TimeUnit.SECONDS)
                .checkInterval(1000, TimeUnit.MILLISECONDS)
                .secondaryAuthentication(m)
                .build();
        try {
            PulsarClient client = PulsarClient.builder()
                    .serviceUrlProvider(failover)
                    .authentication(AuthenticationFactory.token("xxxx"))
                    .build();
            Producer<String> producer = client.newProducer(Schema.STRING)
                    .topic("public/tcloud-log/test")
                    .create();
            long index = 0;
            while (true) {
                String msg = "hello-" + index;
                producer.send(msg);
                Thread.sleep(1000);
                index = index + 1;
                System.out.println(msg + " send ok");
            }
        } catch (Exception e) {
            System.out.println("exit");
            e.printStackTrace();
        }
    }
}

client version: 2.10.3

[tianshimoyi]

Then I restart the producer code directly, and it will not switch.