More Kotlin (Racing Ambiguous Coroutines)
18 Mar 2023Shows basic examples where two or more suspended functions can be executed in parallel, the first result to complete successfully will be used and the rest will be cancelled, ending the nearest coroutine scope.
Useful for Happy Eyeballs or another fast fallback algorithm is required.
The main entry point suspend begins with an implicit Default Dispatcher (backed by a shared pool of threads on JVM).
Wrapping with coroutineScope will wait for child coroutines until completed or cancelled (when an exception is raised all children in the scope are cancelled).
Example 1 using select:
Use select to wait for the result of multiple suspending functions simultaneously. If any fail, then select produces the exception.
The awaitOn is called when a deferred value is resolved then emits the result to the enclosing select clause.
Call coroutineContext.cancelChildren() once the select has produced a result, all coroutines in the scope are cancelled.
The main function will complete after the quickest task completes - in this case task1.
import kotlinx.coroutines.*
import kotlinx.coroutines.selects.select
import kotlin.system.measureTimeMillis
suspend fun task1(): String {
delay(1000)
return "Task 1 completed"
}
suspend fun task2(): String {
delay(12000)
return "Task 2 completed"
}
suspend fun task3(): String {
delay(13000)
return "Task 3 completed"
}
suspend fun main() = coroutineScope {
val tasks = listOf(::task3, ::task2, ::task1)
val ms = measureTimeMillis {
val first =
select {
tasks.forEach { task ->
async { task() }.onAwait { it }
}
}
coroutineContext.cancelChildren()
println(first) // "Task 1 completed"
}
println("in $ms milliseconds)")
}
Example 2 using Channels:
Where channelFlow is a cold flow.
The first operator returns the first value or exception emitted by the flow and then cancels flow’s collection.
import kotlinx.coroutines.*
import kotlinx.coroutines.flow.channelFlow
import kotlinx.coroutines.flow.first
import kotlin.system.measureTimeMillis
suspend fun task1(): String {
delay(1000)
return "Task 1 completed"
}
suspend fun task2(): String {
delay(12000)
return "Task 2 completed"
}
suspend fun task3(): String {
delay(13000)
return "Task 3 completed"
}
suspend fun main() = coroutineScope {
val tasks = listOf(::task3, ::task2, ::task1)
val ms = measureTimeMillis {
val first =
channelFlow {
tasks.forEach { task ->
launch { send(task()) }
}
}.first()
println(first)
//"Task 1 completed"
}
println("in $ms milliseconds)")
}
Example 3 using Flows:
Similar to channelFlow, concurrently merge without limit on the number of simultaneously collected flows.
The first operator returns the first element emitted by the flow and then cancels flow’s collection
import kotlinx.coroutines.*
import kotlinx.coroutines.flow.*
import kotlin.system.measureTimeMillis
suspend fun task1(): String {
delay(1000)
return "Task 1 completed"
}
suspend fun task2(): String {
delay(12000)
return "Task 2 completed"
}
suspend fun task3(): String {
delay(13000)
return "Task 3 completed"
}
@OptIn(FlowPreview::class)
suspend fun main() = coroutineScope {
val tasks = listOf(::task3, ::task2, ::task1)
val ms = measureTimeMillis {
val first = tasks.map { it.asFlow() }.merge().first()
println(first)
//"Task 1 completed"
}
println("in $ms milliseconds)")
}
Example 4 Exceptions:
Wrapping with supervisorScope adds a SupervisorJob to the Coroutine Context,
allowing any Exception thrown by an async job to be handled by the User.
The first job that fails, select produces the corresponding exception as the result.
Unless cancelled, the remaining jobs will continue to run and prevent the coroutine scope from completing.
import kotlinx.coroutines.*
import kotlinx.coroutines.selects.select
import kotlin.system.measureTimeMillis
suspend fun task1(): String {
delay(1000)
error("Task 1 failed")
}
suspend fun task2(): String {
delay(12000)
return "Task 2 completed"
}
suspend fun task3(): String {
delay(13000)
return "Task 3 completed"
}
suspend fun main() = supervisorScope {
val tasks = listOf(::task3, ::task2, ::task1)
val ms = measureTimeMillis {
try {
val first =
select {
tasks.forEach { task ->
async() { task() }.onAwait { it }
}
}
println(first)
} catch (e: Exception) {
println(e.message)
// "Task 1 failed"
} finally {
coroutineContext.cancelChildren()
}
}
println("in $ms milliseconds)")
}
Example 5 Happy Eyeballs:
Hostnames on the Internet often resolve to multiple IP addresses, each of which may have different performance and connectivity characteristics. Address families (IPv4 or IPv6) may be blocked, broken, or sub-optimal on a network, clients that attempt multiple connections in parallel have a chance of establishing a connection more quickly reducing the overall client delay.
Pseudocode for the Happy Eyeballs algorithm of racing connections returning the quickest resolved ip address for e.g. fastly.com. A real example would attempt to connect with a socket and close any sockets that are not used.
The ordering of ip addresses is expected to be interleaved by family type.
Each suspend function is decorated with a staggered 250ms connection delay using onEach by order of input. The first ip in the list is started without delay, subsequent ips are delayed (e.g. 250ms, 500ms …) before starting.
What is missing from this naive example - tasks should be staggered by a delay or run immediately if the previous task fails. The exception handling required is quite tricky. (see gist.github.com/griffio/073e5f440971e7e19dbc3e2011c9ec07 for possible implementations)
Once again, the flows merge concurrently starting after their respective delay, the first ip to “resolve” is returned
and the rest of the tasks are cancelled.
import kotlinx.coroutines.*
import kotlinx.coroutines.flow.*
import kotlin.time.Duration
import kotlin.time.Duration.Companion.milliseconds
import kotlin.system.measureTimeMillis
suspend fun ip4a(): String {
delay(1200)
return "151.101.1.57"
}
suspend fun ip6a(): String {
delay(1100)
return "2a04:4e42:400::313"
}
suspend fun ip4b(): String {
delay(100)
return "151.101.193.57"
}
suspend fun ip6b(): String {
delay(100)
return "2a04:4e42:600::313"
}
suspend fun main(): Unit = coroutineScope {
val ms = measureTimeMillis {
val fastly = listOf(::ip6a, ::ip4a, ::ip6b, ::ip4b)
val result = happyEyeballs(fastly, 250.milliseconds)
println(result)
// 2a04:4e42:600::313
}
println("in $ms milliseconds)")
}
@OptIn(FlowPreview::class)
suspend fun <T> happyEyeballs(tasks: List<suspend () -> T>, delayBy: Duration): T = coroutineScope {
val flows = tasks.mapIndexed { ix, it ->
it.asFlow().onEach {
delay(delayBy * ix)
}
}
flows.merge().first()
}
Example 6 using channelFlow with exception handling to run upto 5 tasks.
The first task is started immediately, the second task is started after waiting for the delay or if the first fails.
The second task fails and the third is started immediately and so on.
The third task completes before the fifth task starts.
import kotlinx.coroutines.*
import kotlinx.coroutines.channels.Channel
import kotlinx.coroutines.flow.*
import kotlinx.coroutines.selects.onTimeout
import kotlinx.coroutines.selects.select
import kotlin.random.Random
import kotlin.time.Duration
import kotlin.time.Duration.Companion.milliseconds
import kotlin.time.Duration.Companion.seconds
suspend fun task1(): String {
println("<start>.task1")
//error("fail: task1")
delay(10.seconds)
println("<finish>.task1")
return "success: task1"
}
suspend fun task2(): String {
println("<start>.task2")
delay(1.seconds)
error("fail: task2")
return "success: task2"
}
suspend fun task3(): String {
println("<start>.task3")
//error("fail: task3")
delay(3.seconds)
println("<finish>.task3")
return "success: task3"
}
suspend fun task4(): String {
println("<start>.task4")
// error("fail: task4")
delay(2.seconds)
println("<finish>.task4")
return "success: task4"
}
suspend fun task5(): String {
println("<start>.task5")
delay(2.seconds)
println("<finish>.task5")
return "success: task5"
}
// wait for delay or error to be signaled first
suspend fun delayOrFail(channel: Channel<Unit>, delayBy: Duration) = channelFlow {
launch { send(delayTask(delayBy)) }
launch { send(failedTask(channel)) }
}.first()
//Another way using select
@OptIn(ExperimentalCoroutinesApi::class)
suspend fun delayOrFailAlt(failedTask: Channel<Unit>, delayBy: Duration) {
select {
failedTask.onReceive {}
onTimeout(delayBy) {}
}
}
suspend fun <T> failedTask(failed: Channel<T>) {
failed.receive()
}
suspend fun delayTask(delayBy: Duration) {
delay(delayBy)
}
suspend fun <T> happyEyeBalls(tasks: List<suspend () -> T>, delayedBy: Duration): Flow<T> = channelFlow {
val failedTask = Channel<Unit>(1)
// start the first task immediately
launch {
try {
send(tasks.first()())
} catch (e: Exception) {
failedTask.trySend(Unit) // it failed, send message to channel
}
}
// start remaining tasks only after either waiting for delay or failed channel element is received
tasks.drop(1).forEach { task ->
delayOrFail(failedTask, delayedBy) // wait for delay or failed signal to continue next task
launch {
try {
send(task())
} catch (e: Exception) {
failedTask.trySend(Unit) // task failed, send message to channel - next task will start
}
}
}
}
suspend fun main(): Unit = coroutineScope {
val tasks = listOf(::task1, ::task2, ::task3, ::task4, ::task5)
// the first task to succeed - the others are cancelled
val winner = happyEyeBalls(tasks, 2.seconds).first()
println(winner) // task3 is quickest - task5 doesn't start
}
See the repo for example kotlin-coroutine-racing