在集群的 Java EE 应用程序中并行化作为 @Singleton 实现的计划任务

Question

I am currently developing a Java EE application which will be deployed on a Payara server 4. The payara server works on host 1 and there are also 2 instances (host 1 and host 2) available as a Payara cluster.

The problem which I have is, if we schedule a task in the singleton class, then the task will be executed thrice at the same time. The question is 2-fold.

• Why does this phenomenon happen?
• How can I avoid such multiple execution?

---

The entry point looks like this. (Not the whole.)

@ApplicationScoped
@Singleton
public class StartClass {
    public void StartClass() {}
    public void init(@Observes @Initialized(ApplicationScoped.class) ServletContext) throws ... { ... }

    @Schedule(hour="*", minute="0", persistent=false)
    public void runJob() {
        MyClass my_class = new MyClass();
        my_class.do_the_job();
    }
}

Here the scheduled task which I mentioned above is my_class.do_the_job() . This method is just to insert a line into a DB and the source code is very similar to this tutorial and has several System.out.println("DO_THE_JOB") . The target table has a column which is automatically filled with the inserted timestamp. I also put an information about the host with System.getenv("HOST") . And the result is that we insert three rows every hour: 2 rows from host 1 and 1 row from host 2. But the result of println() can be found once in the log file.

Notes:

• Usage of JTA is not the point. I would like to understand the behaviour of the application and the cluster.
• Java version is 8
• If there should be another relevant part, let me know.

Answer 1

In the JEE spec there is no way to have a @Singleton which is cluster-wide.

One possible option to get around this is to use the Payara-specific clustered singleton with the additional @Clustered annotation. It is available starting from Payara 4.182.

Read more about it here Payara clustered singleton

Answer 2

Why does this phenomenon happen?

Simply because Singleton beans are not clusterable . Meaning that each node of the cluster has its own instance of StartClass . You may find additional useful information on a similar SO question .

How can I avoid such multiple execution?

With your current setup, you can't . What you possibly need is a scheduler which supports clustering. A popular option is quartz-scheduler which supports clustering.