public sealed Interface StructuredTaskScope<T, R, R_X extends Throwable>

StructuredTaskScope defines the static open() method to create and open a new StructuredTaskScope. It defines the close() method to close it. The API is designed to be used with the try-with-resources statement where a StructuredTaskScope is opened as a resource and then closed automatically. The code inside the try block uses the fork(Callable) method to fork subtasks. Each call to the fork(Callable) method starts a new Thread (typically a virtual thread) to execute a subtask as a value-returning method. The subtask executes concurrently with the code inside the try block, and concurrently with other subtasks forked in the scope. After forking all subtasks, the code inside the block uses the join() method to wait for all subtasks to finish (or some other outcome) as a single operation. The code after the join() method processes the outcome. Execution does not continue beyond the try block (or close method) until all threads started in the scope to execute subtasks have finished.

To ensure correct usage, the fork(Callable), join() and close() methods may only be invoked by the owner thread (the thread that opened the StructuredTaskScope), the fork(Callable) method may not be called after join(), the join() method must be invoked to get the outcome after forking subtasks, and the close() method throws an exception after closing if the owner did not invoke the join() method after forking subtasks.

As a first example, consider a "main" task that splits into two subtasks to concurrently fetch values from two remote services. The main task aggregates the results of both subtasks. The example invokes fork(Callable) to fork the two subtasks. Each call to fork(Callable) returns a Subtask as a handle to the forked subtask. Both subtasks may complete successfully, one subtask may succeed and the other may fail, or both subtasks may fail.

The main task in the example is interested in the successful result from both subtasks. It waits in the join() method for both subtasks to complete successfully or for either subtask to fail. If both subtasks complete successfully then the join() method completes normally and the task uses the Subtask.get() method to get the result of each subtask. If one of the subtasks fails then the other subtask is cancelled, and the join() method throws ExecutionException with the exception from the failed subtask as the cause.

try (var scope = StructuredTaskScope.open()) {

    Subtask<String> subtask1 = scope.fork(() -> fetchFromRemoteService1());
    Subtask<Integer> subtask2 = scope.fork(() -> fetchFromRemoteService2());

    // throws ExecutionException if either subtask fails
    scope.join();

    // both subtasks completed successfully
    var result = new MyResult(subtask1.get(), subtask2.get());

} // close

The close() method always waits for threads executing subtasks to finish, even if the join() method throws, so that execution cannot continue beyond the close() method until the interrupted threads finish.

To allow for cancellation, subtasks must be coded so that they finish as soon as possible when interrupted. Subtasks that do not respond to interrupt, e.g. block on methods that are not interruptible, may delay the close() method indefinitely.

In the example, the subtasks produce results of different types (String and Integer). In other cases the subtasks may all produce results of the same type. If the example had used StructuredTaskScope.<String>open() to open the scope then it could only be used to fork subtasks that return a String result.

Joiners

The join() method in the example above completes normally, and returns null, if all subtasks succeed. It throws ExecutionException if any subtask fails. Other policies and outcomes are possible by creating a StructuredTaskScope with a Joiner that implements the desired policy and outcome. A Joiner handles subtasks as they are forked and when they complete, and produces the outcome for the join() method. Instead of null, a Joiner may cause join() to return the result of a specific subtask, a collection of results, or an object constructed from the results of some or all subtasks. When the outcome is an exception, a Joiner may cause join() to throw an exception other than ExecutionException.

A Joiner may cancel the scope (sometimes called "short-circuiting") when some condition is reached, e.g. a subtask fails, that does not require the outcome of other subtasks that are still executing. Cancelling the scope prevents new threads from being started in the scope, cancels subtasks in the scope that have not completed execution, and causes the join() method to wake up with the outcome (result or exception). In the above example, the outcome is that join() completes normally when all subtasks succeed. The scope is cancelled if any subtask fails and join() throws ExecutionException with the exception from the failed subtask as the cause. Other Joiner implementations may cancel the scope for other reasons, and may cause join() to throw a different exception when the outcome is an exception.

The Joiner interface defines static factory methods to create a Joiner for a number of common cases. The interface can be implemented when a more advanced or custom policy is required. A Joiner that returns a non-null result may remove the need for bookkeeping and the need to keep a reference to Subtask objects returned by the fork(Callable) method.

Now consider another example where a main task splits into two subtasks. In this example, each subtask produces a String result and the main task is only interested in the result from the first subtask to complete successfully. The example uses Joiner.anySuccessfulOrThrow() to create a Joiner that produces the result of any subtask that completes successfully.

try (var scope = StructuredTaskScope.open(Joiner.<String>anySuccessfulOrThrow())) {

    scope.fork(callable1);
    scope.fork(callable2);

    // throws ExecutionException if both subtasks fail
    String firstResult = scope.join();

} // close

In the example, the task forks the two subtasks, then waits in the join() method for either subtask to complete successfully or for both subtasks to fail. If one of the subtasks completes successfully then the Joiner causes the other subtask to be cancelled (this will interrupt the thread executing the subtask), and the join() method returns the result from the successful subtask. Cancelling the other subtask avoids the task waiting for a result that it doesn't care about. If both subtasks fail then the join() method throws ExecutionException with the exception from one of the subtasks as the cause. Joiner.anySuccessfulOrThrow(Function) can be used with a function that produces an exception other than ExecutionException to throw when all subtasks fail.

Configuration

The open() and open(Joiner) methods create a StructuredTaskScope with the default configuration. The default configuration has a ThreadFactory that creates unnamed virtual threads, does not name the scope, and has no timeout.

The open(UnaryOperator) and open(Joiner, UnaryOperator) methods can be used to create a StructuredTaskScope that uses a different ThreadFactory, is named for monitoring and management purposes, or has a timeout that cancels the scope if the timeout expires before or while waiting for subtasks to complete. The open methods are called with an operator that is applied to the default configuration and returns a Configuration for the StructuredTaskScope under construction.

The following example opens a new StructuredTaskScope with a ThreadFactory that creates virtual threads named "duke-0", "duke-1" ...

ThreadFactory factory = Thread.ofVirtual().name("duke-", 0).factory();

try (var scope = StructuredTaskScope.open(cf -> cf.withThreadFactory(factory))) {

    var subtask1 = scope.fork( .. );   // runs in a virtual thread with name "duke-0"
    var subtask2 = scope.fork( .. );   // runs in a virtual thread with name "duke-1"

    scope.join();

    var result = new MyResult(subtask1.get(), subtask2.get());

 }

A second example sets a timeout, represented by a Duration. The timeout starts when the new scope is opened. If the timeout expires before or while waiting in the join() method then the scope is cancelled (this interrupts the threads executing the subtasks that have not completed), and the join() method throws ExecutionException with a CancelledByTimeoutException as the cause.

Duration timeout = Duration.ofSeconds(10);

try (var scope = StructuredTaskScope.open(Joiner.<String>allSuccessfulOrThrow(),
                                          cf -> cf.withTimeout(timeout))) {

    scope.fork(callable1);   // subtask takes a really long time
    scope.fork(callable2);

    // throws ExecutionException with CancelledByTimeoutException as cause
    List<String> results = scope.join();

}

Exception handling

The outcome of the join() method is a result or exception. When the outcome is an exception then its cause will typically be the exception from a failed subtask or CancelledByTimeoutException if a timeout was configured.

In some cases it may be useful to add a catch block to the try-with-resources statement to handle the exception. The following example uses the open(UnaryOperator) method to open a scope with a timeout configured. The join() method in this example throws ExecutionException if any subtask fails or the timeout expires. The exception cause is the exception from a failed subtask or CancelledByTimeoutException. The example uses the switch statement to select based on the cause.

try (var scope = StructuredTaskScope.open(cf -> cf.withTimeout(timeout))) {

    ..

} catch (ExecutionException e) {
    switch (e.getCause()) {
        case CancelledByTimeoutException ->
        case IOException ioe -> ..
        default -> ..
    }
}

In other cases it may not be useful to catch the exception but instead leave it to propagate to the configured uncaught exception handler for logging purposes.

For cases where a specific exception triggers the use of a default result then it may be more appropriate to handle this in the subtask itself rather than the subtask failing and the scope owner handling the exception.

The join() method throws InterruptedException when interrupted before or while waiting in the join() method. The Thread Interruption section of the Thread specification provides guidance on handling this exception.

Inheritance of scoped value bindings

When used in conjunction with a StructuredTaskScope, a ScopedValue can also safely and efficiently share a value to methods executed by subtasks forked in the scope. When a ScopedValue object is bound to a value in the thread executing a "main" task then that binding is inherited by the threads created to execute subtasks. The thread executing the main task does not continue beyond the close() method until all threads executing the subtasks have finished. This ensures that the ScopedValue is not reverted to being unbound (or its previous value) while subtasks are executing. In addition to providing a safe and efficient means to inherit a value into subtasks, the inheritance allows sequential code using ScopedValue to be refactored to use structured concurrency.

To ensure correctness, opening a new StructuredTaskScope captures the current thread's scoped value bindings. These are the scoped values bindings that are inherited by the threads created to execute subtasks in the scope. Forking a subtask checks that the bindings in effect at the time that the subtask is forked match the bindings when the StructuredTaskScope was created. This check ensures that a subtask does not inherit a binding that is reverted in the main task before the subtask has completed.

A ScopedValue that is shared across threads requires that the value be an immutable object or for all access to the value to be appropriately synchronized.

The following example demonstrates the inheritance of scoped value bindings. The scoped value USERNAME is bound to the value "duke" for the bounded period of a lambda expression by the thread executing it. The code in the block opens a StructuredTaskScope and forks two subtasks, it then waits in the join() method and aggregates the results from both subtasks. If code executed by the threads running subtask1 and subtask2 uses ScopedValue#get(), to get the value of USERNAME, then value "duke" will be returned.

private static final ScopedValue<String> USERNAME = ScopedValue.newInstance();

MyResult result = ScopedValue.where(USERNAME, "duke").call(() -> {

    try (var scope = StructuredTaskScope.open()) {

        Subtask<String> subtask1 = scope.fork( .. );    // inherits binding
        Subtask<Integer> subtask2 = scope.fork( .. );   // inherits binding

        scope.join();
        return new MyResult(subtask1.get(), subtask2.get());
    }

});

A scoped value inherited into a subtask may be rebound to a new value in the subtask for the bounded execution of some method executed in the subtask. When the method completes, the value of the ScopedValue reverts to its previous value, the value inherited from the thread executing the main task.

A subtask may execute code that itself opens a new StructuredTaskScope. A main task executing in thread T1 opens a StructuredTaskScope and forks a subtask that runs in thread T2. The scoped value bindings captured when T1 opens the scope are inherited into T2. The subtask (in thread T2) executes code that opens a new StructuredTaskScope and forks a (sub-)subtask that runs in thread T3. The scoped value bindings captured when T2 opens the scope are inherited into T3. These include the bindings that were inherited from T1. In effect, scoped values are inherited into a tree of subtasks, not just one level of subtask.

Memory consistency effects

Actions in the owner thread of a StructuredTaskScope prior to forking of a subtask happen-before any actions taken by the thread that executes the subtask, which in turn happen-before actions in any thread that successfully obtains the subtask outcome with Subtask.get() or Subtask.exception(). If a subtask's outcome contributes to the result or exception from join(), then any actions taken by the thread executing that subtask happen-before the owner thread returns from join() with the outcome.

General exceptions

Unless otherwise specified, passing a null argument to a method in this class will cause a NullPointerException to be thrown.