Scheduler

Sponge exposes the Scheduler to allow you to designate tasks to be executed in the future. The Scheduler provides a Task.Builder with which you can specify task properties such as the delay, interval, name, (a)synchronicity, and Runnable (see Task Properties).

Task Builder

First, obtain an instance of the Task.Builder:

import org.spongepowered.api.scheduler.Task;

Task.Builder taskBuilder = Task.builder();

The only required property is the Runnable, which you can specify using Task.Builder#execute(Runnable):

taskBuilder.execute(new Runnable() {
    public void run() {
        logger.info("Yay! Schedulers!");
    }
});

or using Java 8 syntax with Task.Builder#execute(Runnable runnable)

taskBuilder.execute(
    () -> {
        logger.info("Yay! Schedulers!");
    }
);

or using Java 8 syntax with Task.Builder#execute(Consumer<Task> task)

taskBuilder.execute(
    task -> {
        logger.info("Yay! Schedulers! :" + task.getName());
    }
);

Task Properties

Using the Task.Builder, you can specify other, optional properties, as described below.

Eigenschap	Method Used	Omschrijving
delay	delayTicks(long delay) delay(long delay, TimeUnit unit)	The optional amount of time to pass before the task is to be run. The time is specified as a number of ticks with the `delayTicks()` method, or it may be provided as a number of a more convenient time unit by specifying a TimeUnit with the delay() method. Either method, but not both, can specified per task.
interval	intervalTicks( long interval) interval(long interval, TimeUnit unit)	The amount of time between repetitions of the task. If an interval is not specified, the task will not be repeated. The time is specified as a number of ticks with the `intervalTicks()` method, or it may be provided as a number of a more convenient time unit by specifying a TimeUnit with the interval() method. Either method, but not both, can specified per task.
synchronization	async()	A synchronous task is run in the game’s main loop in series with the tick cycle. If `Task.Builder#async` is used, the task will be run asynchronously. Therefore, it will run in its own thread, independently of the tick cycle, and may not safely use game state. (See Asynchronous Tasks.)
name	name(String name)	The name of the task. By default, the name of the task will be PLUGIN_ID “-” ( “A-” \| “S-” ) SERIAL_ID. For example, a default task name could look like “fooplugin-A-12”. No two active tasks will have the same serial ID for the same synchronization type. If a task name is specified, it should be descriptive and aid users in debugging your plugin.

Lastly, submit the task to the scheduler using Task.Builder#submit(Object).

And that’s it! To summarize, a fully functional scheduled task that would run asynchronously every 5 minutes after an initial delay of 100 milliseconds could be built and submitted using the following code:

import java.util.concurrent.TimeUnit;

Task task = Task.builder().execute(() -> logger.info("Yay! Schedulers!"))
    .async().delay(100, TimeUnit.MILLISECONDS).interval(5, TimeUnit.MINUTES)
    .name("ExamplePlugin - Fetch Stats from Database").submit(plugin);

To cancel a task, simply call the Task#cancel() method:

task.cancel();

If you need to cancel the task from within the runnable itself, you can instead opt to use a Consumer<Task>` in order to access the task. The below example will schedule a task that will count down from 60 and cancel itself upon reaching 0.

@Listener
public void onGameInit(GameInitializationEvent event) {
    Task task = Task.builder().execute(new CancellingTimerTask())
        .interval(1, TimeUnit.SECONDS)
        .name("Self-Cancelling Timer Task").submit(plugin);
}

private class CancellingTimerTask implements Consumer<Task> {
    private int seconds = 60;
    @Override
    public void accept(Task task) {
        seconds--;
        Sponge.getServer()
            .getBroadcastChannel()
            .send(Text.of("Remaining Time: "+seconds+"s"));
        if (seconds < 1) {
            task.cancel();
        }
    }
}

Asynchronous Tasks

Asynchronous tasks should be used primarily for code that may take a significant period of time to execute, namely requests to another server or database. If done on the main thread, a request to another server could greatly impact the performance of the game, since the next tick cannot be fired until the request is completed.

Since Minecraft is largely single-threaded, there is little you can do in an asynchronous thread. If you must run a thread asynchronously, you should execute all of the code that does not use the SpongeAPI/affect Minecraft, then register another synchronous task to handle the code that needs the API. There are a few parts of Minecraft that you can work with asynchronously, including:

Chat
Sponge’s built-in Permissions handling
Sponge’s scheduler

In addition, there are a few other operations that are safe to do asynchronously:

Independent network requests
Filesystem I/O (excluding files used by Sponge)

Waarschuwing

Accessing game objects outside of the main thread can lead to crashes, inconsistencies and various other problems and should be avoided. If this is done wrong, you can get a ConcurrentModificationException with or without a server crash at best and a corrupted player/world/server at worst.

Compatibility with other libraries

As your plugin grows in size and scope you might want to start using one of the many concurrency libraries available for Java and the JVM. These libraries do tend to support Java’s ExecutorService as a means of directing on which thread the task is executed.

To allow these libraries to work with Sponge’s Scheduler the following methods can be used:

Scheduler#createSyncExecutor(Object) creates a SpongeExecutorService which executes tasks through Sponge’s synchronous scheduler.
Scheduler#createAsyncExecutor(Object) creates a SpongeExecutorService which executes tasks through Sponge’s asynchronous scheduler. Tasks are subject to the restrictions mentioned in Asynchronous Tasks.

One thing to keep in mind is that any tasks that interacts with Sponge outside of the interactions listed in Asynchronous Tasks need to be executed on the ExecutorService created with Scheduler#createSyncExecutor(Object) to be thread-safe.

import org.spongepowered.api.scheduler.SpongeExecutorService;

SpongeExecutorService minecraftExecutor = Sponge.getScheduler().createSyncExecutor(plugin);

minecraftExecutor.submit(() -> { ... });

minecraftExecutor.schedule(() -> { ... }, 10, TimeUnit.SECONDS);

Almost all libraries have some way of adapting the ExecutorService to natively schedule tasks. As an example the following paragraphs will explain how the ExecutorService is used in a number of libraries.

CompletableFuture (Java 8)

With Java 8 the CompletableFuture object was added to the standard library. Compared to the Future object, this allows for the developer to provide a callback that is called when the future completes rather than blocking the thread until the future eventually completes.

CompletableFuture is a fluent interface which usually has the following three variations for each of its functions:

CompletableFuture#<function>Async(..., Executor ex) Executes this function through ex
CompletableFuture#<function>Async(...) Executes this function through ForkJoinPool.commonPool()
CompletableFuture#<function>(...) Executes this function on whatever thread the previous CompletableFuture was completed on.

import java.util.concurrent.CompletableFuture;

SpongeExecutorService minecraftExecutor = Sponge.getScheduler().createSyncExecutor(plugin);

CompletableFuture.supplyAsync(() -> {
    // ASYNC: ForkJoinPool.commonPool()
    return 42;
}).thenAcceptAsync((awesomeValue) -> {
    // SYNC: minecraftExecutor
}, minecraftExecutor).thenRun(() -> {
    // SYNC: minecraftExecutor
});

RxJava

RxJava is an implementation of the Reactive Extensions concept for the JVM.

Multithreading in Rx is managed through various Schedulers. Using the Schedulers#from(Executor executor) function the Executor provided by Sponge can be turned into a Scheduler.

Much like CompletableFuture by default actions are executed on the same thread that completed the previous part of the chain. Use Observable#observeOn(Scheduler scheduler) to move between threads.

One important thing to bear in mind is that the root Observable gets invoked on whatever thread Observable#subscribe() was called on. If the root observable interacts with Sponge it should be forced to run synchronously using Observable#subscribeOn(Scheduler scheduler).

import rx.Observable;
import rx.Scheduler;
import rx.schedulers.Schedulers;

SpongeExecutorService executor = Sponge.getScheduler().createSyncExecutor(plugin);
Scheduler minecraftScheduler = Schedulers.from(executor);

Observable.defer(() -> Observable.from(Sponge.getServer().getOnlinePlayers()))
          .subscribeOn(minecraftScheduler) // defer -> SYNC: minecraftScheduler
          .observeOn(Schedulers.io()) // -> ASYNC: Schedulers.io()
          .filter(player -> {
              // ASYNC: Schedulers.io()
              return "Flards".equals(player.getName());
          })
          .observeOn(minecraftScheduler) // -> SYNC: minecraftScheduler
          .subscribe(player -> {
              // SYNC: minecraftScheduler
              player.kick(Text.of("Computer says no"));
          });

Scala

Scala comes with a built-in Future object which a lot of scala framework mirror in design. Most methods of the Future accept an ExecutionContext which determined where that part of the operation is executed. This is different from the CompletableFuture or RxJava since they default to executing on the same thread on which the previous operation ended.

The fact that all these operation try to implicitly find an ExecutionContext means that you can easily use the default ExecutionContext.global and specifically run the parts that need to be thread-safe on the Sponge server thread.

To avoid accidentally scheduling work on through the Sponge ExecutorContext another context should be implicitly defined so it acts as the default choice. To maintain thread safety only the functions that actually interact with Sponge will need to have the Sponge executor specified.

import scala.concurrent.ExecutionContext

val executor = Sponge.getScheduler().createSyncExecutor(plugin)

import ExecutionContext.Implicits.global
val ec = ExecutionContext.fromExecutorService(executor)

val future = Future {
    // ASYNC: ExecutionContext.Implicits.global
}

future foreach {
    case value => // SYNC: ec
}(ec)

future map {
    case value => 42 // SYNC: ec
}(ec).foreach {
    case value => println(value) // ASYNC: ExecutionContext.Implicits.global
}