2019-01-29 - Demystifying Kotlin Coroutines

training@instil.co
January 2019
© Instil Software 2019
Demystifying Kotlin Coroutines
GDG Dublin
@BoyleEamonn
Eamonn Boyle

We’ve really bought into Kotlin
Kotlin Conf 2018

We held a workshop at the conference
• React Web App with Kotlin on Server & Browser
Kotlin Conf 2018

Long Story Short - Synchronous is Easy
fun doWork() {
disableUI()
val exchangeRates = readExchangeRate()
val report = buildReport(exchangeRates)
saveReport(report)
enableUI()
}
1
2
3
4
5
imdb.com
return

Long Story Short - Synchronous is Easy
fun doWork() {
disableUI()
saveReport(report)
enableUI()
}
1
2
3
4
5
We can’t escape asynchronous programming
Typically we can’t block due to limited resources
But we want ways to hide the complexity of async

Long Story Short – Callback Hell
fun doWork() {
disableUI()
readExchangeRate { exchangeRate ->
buildReport(exchangeRate) { report ->
saveReport(report) {
enableUI()
}
}
}
}
imdb.com
return

Long Story Short – Futures Help
fun doWork() {
disableUI()
readExchangeRate()
.thenCompose(::buildReport)
.thenCompose(::saveReport)
.thenApply {
enableUI()
}
}
imdb.com
Once you’ve learned the
complete API, then you have my
permission to die
return

Long Story Short – Coroutines Are Better
suspend fun doWork() {
disableUI()
saveReport(report)
enableUI()
}
imdb.com
suspend
suspend
suspend
return

Long Story Short – Coroutines Are Better
disableUI()
saveReport(report)
enableUI()
}
fun doWork() {
disableUI()
saveReport(report)
enableUI()
}

Long Story Short – Control Dispatch
suspend fun doWorkWithThreadControl() {
disableUI()
withContext(Dispatchers.IO) {
saveReport(report)
}
enableUI()
}

Generator Use Case
1 2 3 4 5 6 7 1 2 3 4 5 6 7
Reading
Processing
fun readData(filename: String): List<Int> {
val data = mutableListOf<Int>()
File(filename).useLines {lines ->
lines.forEach {
val datum = it.toInt()
log("Read $datum")
data.add(datum)
}
}
return data
}
fun processData(filename: String) {
for (datum in readData(filename)) {
log("Processing data $datum")
}
}

lines.forEach {
log("Read $datum")
data.add(datum)
}
}
return data
}
val data = readData(filename)
for (datum in data) {
}
}

lines.forEach {
log("Read $datum")
data.add(datum)
}
}
return data
}
val data = readData(filename)
}
}
What if the read takes too
long?
What if the amount of data
read is too large?

We may need to interleave reading and processing
How should we implement this? Perhaps hand code the interleaving
• Perhaps use standard interfaces to simplify
Generator Use Case
1 2 3 4 5 6 7 1 2 3 4 5 6 7
Reading Function
Processing Function
1 1 2 2 3 3 4 4 5 5 6 6 7 7

class CustomInterleaved(filename: String) {
val file = File(filename)
.inputStream()
.bufferedReader()
fun readData(): Int {
val line = file.readLine()
if (line == null) {
file.close()
return -1
}
val datum = line.toInt()
log("Read $datum")
return datum
}
}
Manual Interleaving
val interleaved = CustomInterleaved(filename)
while (true) {
val datum = interleaved.readData()
if (datum < 0) {
break
}
}
}

class CustomIterable(val filename: String) : Iterable<Int>, AutoCloseable {
val file = File(filename).inputStream().bufferedReader()
class CustomIterator(val file: BufferedReader) : Iterator<Int> {
override fun next(): Int {
val datum = file.readLine().toInt()
log("Read $datum")
return datum
}
override fun hasNext(): Boolean = file.ready()
}
override fun iterator(): Iterator<Int> = CustomIterator(file)
override fun close() = log("Closing file")
}
Standard Idioms – Iterable Interface

CustomIterable(filename).use {data ->
}
}
}
Benefit – Simpler Usage

lines.forEach {
log("Read $datum")
data.add(datum)
}
}
return data
}
}
}
fun readData(filename: String): Sequence<Int> {
return sequence {
lines.forEach {
log("Read $datum")
yield(datum)
}
}
}
}
}
}
Coroutine Solution

lines.forEach {
log("Read $datum")
data.add(datum)
}
}
return data
}
Coroutine Solution
fun readData(filename: String): Sequence<Int> {
return sequence {
lines.forEach {
log("Read $datum")
yield(datum)
}
}
}
} return
suspend

“Coroutines are computer-program components
that generalize subroutines for non-preemptive
multitasking, by allowing multiple entry points for
suspending and resuming execution at certain
locations.”
Wikipedia
Coroutine

“Coroutines simplify writing code that needs to
suspend and resume. e.g. generators,
asynchronous code, event loops etc
You write it like standard sequential code (more or
less) but under the hood it returns/suspends and
reenters/resumes. ”
Eamonn Boyle
Coroutine
..using Magic!!

They do not help us write faster code
• Although they can help us write more efficient systems
• This may make the whole system more responsive
It is easier to write as it is very similar to standard code
We get to use all the standard coding elements we are used to
• if, for, while
• try-catch
• function calls
• Etc
Benefits of Coroutines

Consider 3 independent jobs of work
Event Loop with Asynchronous Blocks
Start 5 6EndBlocking Start 5 6EndBlocking Start 5 6EndBlocking
fun workSync(message: String) {
log("$message started")
Thread.sleep(sleepTime)
log("$message ended")
}
jobs.forEach {
workSync(it)
}
[1]: Running Synchronous
[1]: Job 1 started
[1]: Job 1 ended
[1]: Job 2 started
[1]: Job 2 ended
[1]: Job 3 started
[1]: Job 3 ended
[1]: Synchronous took 15003 ms

Consider 3 independent jobs of work
Event Loop with Asynchronous Blocks
Start 5 6EndBlocking Start 5 6EndBlocking Start 5 6EndBlocking
Start 5 6EndBlocking
If we want to run these in parallel we could create multiple threads

jobs.map {
val thread = Thread {
workSync(it)
}
thread.start()
thread
}.forEach {
log("Joining thread ${it.id}")
it.join()
}
[1]: Running Parallel Threads
[1]: Joining with thread 12
[12]: Job 1 started
[14]: Job 3 started
[13]: Job 2 started
[13]: Job 2 ended
[12]: Job 1 ended
[14]: Job 3 ended
[1]: Parallel Threads took 5004 ms
There are scalability issues - threads are heavy
Startup times can be slow
The syntax is clunky

Optimal Usage of Fewer Threads
Start End
Blocking
Start End
Blocking
Start 5 6End
Blocking

Asynchronous Use Case – Thread Pools
Start End
Blocking
Start End
Blocking
Start 5 6End
Blocking

Asynchronous Use Case – Thread Dispatching
Start
EndBlocking
Start
EndBlocking
Start
5 6EndBlocking

Asynchronous Use Case – Common Characteristics
Start EndStart EndStart 5 6End
Quite often we must use certain threads for certain work
• E.g. UI typically has a single main thread
Quite often the asynchronous work can be optimised
• Event interrupts for I/O
• Single polling thread for multiple elements
• Pooling certain work on dedicated worker threads

fun mySleep(time: Long, action: () -> Unit): CompletableFuture<Unit> {
return CompletableFuture.runAsync {
Thread.sleep(time)
}.thenApply { action() }
}
fun workSync(message: String): CompletableFuture<Unit> {
return mySleep(sleepTime) {
}
}
jobs.map {
workSync(it)
}.forEach {
it.join()
}
Callbacks & CompletableFuture Solution

fun mySleep(time: Long, action: () -> Unit): CompletableFuture<Unit> {
return CompletableFuture.runAsync {
Thread.sleep(time)
}.thenApply { action() }
}
fun workSync(message: String): CompletableFuture<Unit> {
return mySleep(sleepTime) {
}
}
jobs.map {
workSync(it)
}.forEach {
it.join()
}
[1]: Running Synchronous
[1]: Job 1 started
[1]: Job 2 started
[1]: Job 3 started
[12]: Job 1 ended
[13]: Job 2 ended
[14]: Job 3 ended
[1]: Synchronous took 5018 ms
CompletableFuture Solution

CompletableFuture Solution
Start
EndBlocking
Start
EndBlocking
Start
5 6EndBlocking

}
jobs.forEach {
workSync(it)
}
Coroutine Solution
suspend fun workAsync(message: String) {
delay(sleepTime)
}
runBlocking {
jobs.map {
launch {
workAsync(it)
}
}.joinAll()
}

Coroutine Solution
delay(sleepTime)
}
runBlocking {
jobs.map {
launch {
workAsync(it)
}
}.joinAll()
}
[1]: Running Coroutines
[1]: Job 1 started
[1]: Job 2 started
[1]: Job 3 started
[1]: Job 1 ended
[1]: Job 2 ended
[1]: Job 3 ended
[1]: Coroutines took 5161 ms

Asynchronous Coroutines
Start End
Blocking
Start End
Blocking
Start 5 6End
Blocking

}
delay(sleepTime)
}
Suspend Functions

}
Suspend Functions
delay(sleepTime)
}

}
Suspend Functions
delay(sleepTime)
}
This is a suspension
point

Coroutines as Light Weight Threads
Start EndBlocking
Start EndBlocking

Coroutines Scale
Start EndBlocking
Start EndBlocking
Start EndBlocking
Start EndBlocking
Start EndBlocking

val a = step1()
step2()
val b = step3()
val total = a + b
log("The sum is $total")
}
Another Example

val a = step1()
step2()
val b = step3()
val total = a + b
}
fun doWork(): CompletableFuture<Unit> {
return step1().thenCompose { a ->
step2().thenCompose {
step3().thenApply { b ->
val total = a + b
}
}
}
}
Coroutine as Completable

val a = step1()
step2()
val b = step3()
val total = a + b
}
class SimpleCouroutineDoWork {
var label = 0
var a: Int? = null
var b: Int? = null
var total: Int? = null
fun resumeWith(...) {
}
}
Coroutine Implementation – State Machine
This class is the continuation
It represents the state of a suspending function

val a = step1()
step2()
val b = step3()
val total = a + b
}
fun resumeWith(result: Result<Any>) {
var newResult = result.getOrThrow()
if (label == 0) {
newResult = step1(this)
label = 1
if (newResult == COROUTINE_SUSPENDED) return
}
if (label == 1) {
a = newResult as Int
label = 2
}
if (label == 2) {
b = newResult as Int
label = 3
}
if (label == 4) {
total = a!! + b!!
}
}

val a = step1()
step2()
val b = step3()
val total = a + b
}
fun resumeWith(result: Result<Any>) {
var newResult = result.getOrThrow()
if (label == 0) {
label = 1
if (newResult == COROUTINE_SUSPENDED) {
return
}
}
if (label == 1) {
a = newResult as Int
...
This is mainly for efficiency reasons
Less objects created to handle the context switching

val a = step1()
step2()
val b = step3()
val total = a + b
}
suspend fun step1() {
log("Step 1 started")
delay(1000)
log("Step 1 ended")
}
Coroutine – Suspend Functions Calling Each Other

Note, a suspend function is not a coroutine
• They are used to write parts of a coroutine
• They cannot be called from normal functions
A corountine is suspendable, sequential computation
It is similar to a thread but is not bound to any particular thread
• It can be created, started and ended
• It can suspend on one thread and resume on another
• It can return a result
Coroutine Builders

The compiler translates our suspend functions
The standard library provides only a primitive API
• This is very much by design
Using these fundamental building blocks, libraries add functionality
This add a lot of flexibility
• Consider different platforms – JVM, JS, Native
• Consider different use cases – generators, async await, actors
• We can add various different threading strategies
Coroutine APIs

kotlinx.coroutines adds the API entitles for useful coroutines
Scope – Encloses a coroutines and its child coroutines
Context – A key/value collection of values to control execution
Dispatcher – Controls coroutine execution e.g. which threads to use
Builders – functions to create a coroutine
Interceptor – Wraps/intercepts coroutine execution
Bridge – A builder that can be called from a non-suspend function
Coroutine Builders

Controls which thread is used to execute a continuation
• A scope with its context will define a dispatcher
These will be platform dependent but exists in Dispatchers object
Coroutine Dispatchers
Dispatcher Description
Default Executes on a default background thread pool
Unconfined Executes using any available thread
Main Executes on the Main (UI) thread
IO Executes blocking I/O calls. Shares threads with Default
but will create and shutdown additional threads

Any standard Java Executor Service can become a dispatcher
• Simply invoke the ‘asCoroutineDispatcher’ extension method
Other dedicated functions exist but these are going to be replaced
Using a Custom Coroutine Dispatcher
fun runDemoViaCustomThreadPool(name: String) {
val threadPool = Executors.newFixedThreadPool(2)
val dispatcher = threadPool.asCoroutineDispatcher()
runBlocking(dispatcher) {
// ...
}
threadPool.shutdown()
}

These are functions that are supplied a suspend function/lambda
and creates a coroutine
Coroutine Builders
Builder Description
launch Returns a Job object representing the coroutine but is largely fire
and forget. It is synchronised with another coroutine using join.
async Returns a Deferred<T> object which is an extension of Job that
includes a result. The result is acquired using await.
runBlocking Runs a coroutine and blocks the current thread until it completes
sequence Creates generators using a suspend function

These suspending functions execute against a Coroutine Scope
• They are extension methods of CoroutineScope
Suspending Functions
Builder Description
delay Executes a non-blocking sleep
yield Yields the thread in single thread disaptchers
withContext Like runBlocking within a coroutine i.e. it suspends
withTimeout Sets a time limit on execution of a function
await, awaitAll Waits for one or multiple Deferred to complete and returns
the result/s
join, joinAll Waits for one or multiple jobs to complete

class MyActivity : AppCompatActivity(), CoroutineScope {
lateinit var job: Job
override val coroutineContext: CoroutineContext
get() = Dispatchers.Main + job
override fun onCreate(savedInstanceState: Bundle?) {
super.onCreate(savedInstanceState)
job = Job()
}
override fun onDestroy() {
super.onDestroy()
job.cancel() // Cancel job on activity destroy. After Destroy
// all children jobs will be cancelled automatically
}
https://kotlin.github.io/kotlinx.coroutines/kotlinx-coroutines-core/kotlinx.coroutines/-coroutine-scope/
Android Activity

fun loadDataFromUI() = launch { // this (Main) Dispatcher
val ioData = async(Dispatchers.IO) { // IO Dispatcher
// blocking I/O operation
}
// Work here will execute in parallel to IO
// It will run on the Main thread
val data = ioData.await() // Wait for result of I/O
// Complete work in Main thread
}
Android Activity
https://kotlin.github.io/kotlinx.coroutines/kotlinx-coroutines-core/kotlinx.coroutines/-coroutine-scope/

fun voidPayment(tender: CheckTender) = launch(Dispatchers.Default) {
when (tender.type) {
CREDIT, MANUAL_CREDIT -> voidCreditPayment(tender)
GIFT_CARD, MANUAL_GIFT_CARD -> voidGiftCardPayment(tender)
else -> onVoidFailed()
}
}
suspend fun voidCreditPayment(tender: CheckTender) {
try {
val voidPaymentResponse = voidService.voidPayment().await()
when (voidPaymentResponse.result) {
SUCCESS -> onCreditVoidSuccess(tender, voidPaymentResponse)
else -> onCreditVoidFailed(tender, voidPaymentResponse)
}
} catch (throwable: Throwable) {
onVoidFailed()
}
}

Coroutines went stable in 1.3
Similar things exist in other languages
• E.g. async await has been in C# since 2014, F# since 2012
• JavaScript, Python have very similar syntax to C#
• Golang has goroutines & channels
• They are all very useful at simplifying your code
We now have a great solution for the JVM (and others)
The Kotlin solution is built from primitives that open up flexibility
Summary

https://github.com/Kotlin/KEEP/blob/master/proposals/coroutines.md
https://github.com/Kotlin/kotlinx.coroutines
https://www.imdb.com
References

2019-01-29 - Demystifying Kotlin Coroutines

More Related Content

What's hot

Similar to 2019-01-29 - Demystifying Kotlin Coroutines

More from Eamonn Boyle

Recently uploaded

2019-01-29 - Demystifying Kotlin Coroutines

Editor's Notes