Edit Page

Table of contents

Composing suspending functions

This section covers various approaches to composition of suspending functions.

Sequential by default

Assume that we have two suspending functions defined elsewhere that do something useful like some kind of remote service call or computation. We just pretend they are useful, but actually each one just delays for a second for the purpose of this example:

suspend fun doSomethingUsefulOne(): Int {
    delay(1000L) // pretend we are doing something useful here
    return 13
}

suspend fun doSomethingUsefulTwo(): Int {
    delay(1000L) // pretend we are doing something useful here, too
    return 29
}

What do we do if need to invoke them sequentially – first doSomethingUsefulOne and then doSomethingUsefulTwo and compute the sum of their results? In practice we do this if we use the results of the first function to make a decision on whether we need to invoke the second one or to decide on how to invoke it.

We use a normal sequential invocation, because the code in the coroutine, just like in the regular code, is sequential by default. The following example demonstrates it by measuring the total time it takes to execute both suspending functions:

fun main(args: Array<String>) = runBlocking<Unit> {
    val time = measureTimeMillis {
        val one = doSomethingUsefulOne()
        val two = doSomethingUsefulTwo()
        println("The answer is ${one + two}")
    }
    println("Completed in $time ms")
}

You can get full code here

It produces something like this:

The answer is 42
Completed in 2017 ms

Concurrent using async

What if there are no dependencies between invocation of doSomethingUsefulOne and doSomethingUsefulTwo and we want to get the answer faster, by doing both concurrently? This is where async comes to help.

Conceptually, async is just like launch. It starts a separate coroutine which is a light-weight thread that works concurrently with all the other coroutines. The difference is that launch returns a Job and does not carry any resulting value, while async returns a Deferred – a light-weight non-blocking future that represents a promise to provide a result later. You can use .await() on a deferred value to get its eventual result, but Deferred is also a Job, so you can cancel it if needed.

fun main(args: Array<String>) = runBlocking<Unit> {
    val time = measureTimeMillis {
        val one = async { doSomethingUsefulOne() }
        val two = async { doSomethingUsefulTwo() }
        println("The answer is ${one.await() + two.await()}")
    }
    println("Completed in $time ms")
}

You can get full code here

It produces something like this:

The answer is 42
Completed in 1017 ms

This is twice as fast, because we have concurrent execution of two coroutines. Note, that concurrency with coroutines is always explicit.

Lazily started async

There is a laziness option to async using an optional start parameter with a value of CoroutineStart.LAZY. It starts coroutine only when its result is needed by some await or if a start function is invoked. Run the following example:

fun main(args: Array<String>) = runBlocking<Unit> {
    val time = measureTimeMillis {
        val one = async(start = CoroutineStart.LAZY) { doSomethingUsefulOne() }
        val two = async(start = CoroutineStart.LAZY) { doSomethingUsefulTwo() }
        // some computation
        one.start() // start the first one
        two.start() // start the second one
        println("The answer is ${one.await() + two.await()}")
    }
    println("Completed in $time ms")
}

You can get full code here

It produces something like this:

The answer is 42
Completed in 1017 ms

So, here the two coroutines are defined but not executed as in the previous example, but the control is given to the programmer on when exactly to start the execution by calling start. We first start one, then start two, and then await for the individual coroutines to finish.

Note, that if we have called await in println and omitted start on individual coroutines, then we would have got the sequential behaviour as await starts the coroutine execution and waits for the execution to finish, which is not the intended use-case for laziness. The use-case for async(start = CoroutineStart.LAZY) is a replacement for the standard lazy function in cases when computation of the value involves suspending functions.

Async-style functions

We can define async-style functions that invoke doSomethingUsefulOne and doSomethingUsefulTwo asynchronously using async coroutine builder with an explicit GlobalScope reference. We name such functions with "Async" suffix to highlight the fact that they only start asynchronous computation and one needs to use the resulting deferred value to get the result.

// The result type of somethingUsefulOneAsync is Deferred<Int>
fun somethingUsefulOneAsync() = GlobalScope.async {
    doSomethingUsefulOne()
}

// The result type of somethingUsefulTwoAsync is Deferred<Int>
fun somethingUsefulTwoAsync() = GlobalScope.async {
    doSomethingUsefulTwo()
}

Note, that these xxxAsync functions are not suspending functions. They can be used from anywhere. However, their use always implies asynchronous (here meaning concurrent) execution of their action with the invoking code.

The following example shows their use outside of coroutine:

// note, that we don't have `runBlocking` to the right of `main` in this example
fun main(args: Array<String>) {
    val time = measureTimeMillis {
        // we can initiate async actions outside of a coroutine
        val one = somethingUsefulOneAsync()
        val two = somethingUsefulTwoAsync()
        // but waiting for a result must involve either suspending or blocking.
        // here we use `runBlocking { ... }` to block the main thread while waiting for the result
        runBlocking {
            println("The answer is ${one.await() + two.await()}")
        }
    }
    println("Completed in $time ms")
}

You can get full code here

This programming style with async functions is provided here only for illustration, because it is a popular style in other programming languages. Using this style with Kotlin coroutines is strongly discouraged for the reasons that are explained below.

Consider what happens if between val one = somethingUsefulOneAsync() line and one.await() expression there is some logic error in the code and the program throws an exception and the operation that was being performed by the program aborts. Normally, a global error-handler could catch this exception, log and report the error for developers, but the program could otherwise continue doing other operations. But here we have somethingUsefulOneAsync still running in background, despite the fact, that operation that had initiated it aborts. This problem does not happen with structured concurrency, as shown in the section below.

Structured concurrency with async

Let us take Concurrent using async example and extract a function that concurrently performs doSomethingUsefulOne and doSomethingUsefulTwo and returns the sum of their results. Because async coroutines builder is defined as extension on CoroutineScope we need to have it in the scope and that is what coroutineScope function provides:

suspend fun concurrentSum(): Int = coroutineScope {
    val one = async { doSomethingUsefulOne() }
    val two = async { doSomethingUsefulTwo() }
     one.await() + two.await()
}

This way, if something goes wrong inside the code of concurrentSum function and it throws an exception, all the coroutines that were launched in its scope are cancelled.

fun main(args: Array<String>) = runBlocking<Unit> {
    val time = measureTimeMillis {
        println("The answer is ${concurrentSum()}")
    }
    println("Completed in $time ms")
}

You can get full code here

We still have concurrent execution of both operations as evident from the output of the above main function:

The answer is 42
Completed in 1017 ms

Cancellation is always propagated through coroutines hierarchy:

fun main(args: Array<String>) = runBlocking<Unit> {
    try {
        failedConcurrentSum()
    } catch(e: ArithmeticException) {
        println("Computation failed with ArithmeticException")
    }
}

suspend fun failedConcurrentSum(): Int = coroutineScope {
    val one = async<Int> { 
        try {
            delay(Long.MAX_VALUE) // Emulates very long computation
            42
        } finally {
            println("First child was cancelled")
        }
    }
    val two = async<Int> { 
        println("Second child throws an exception")
        throw ArithmeticException()
    }
        one.await() + two.await()
}

You can get full code here

Note, how both first async and awaiting parent are cancelled on the one child failure:

Second child throws an exception
First child was cancelled
Computation failed with ArithmeticException