Scala command line app with background tasks - multithreading

I'm working on a command line program in Scala. Some of the commands will spawn background thread to perform some work. Some commands will spawn threads that will run the same task repetitively with some delay. I also need a stop command that will prevent existing background task from repeating (doesn't need to kill it, just quit after finishing the iteration). What are the best primitives to construct this sort of program?
I'm thinking about using Futures, e.g. below. What do you guys think of this design? How would you implement this sort of functionality?
case class OneTimeTaskCommand(arg: String)
case class StopTaskCommand(name: String)
case class RepeatingTaskCommand(name: String, delay: Long, arg: String)
def runOneTimeTask(arg: String): Unit = { ... }
def runRepeatingTaskCommand(arg: String): Unit = { ... }
trait Scheduler {
def schedule(name: String, delay: Long): Unit
def unschedule(name: String): Unit
def isScheduled(name: String): Boolean = repeatDelay(name).isDefined
def repeatDelay(name: String): Option[Long]
}
def runCommand(command)(implicit scheduler: Scheduler): Future[Unit] = {
command match {
case OneTimeTaskCommand(arg) => Future(runOneTimeTask(arg))
case StopTaskCommand(name) =>
if (scheduler.isScheduled(name)) {
scheduler.unschedule(name)
Future.successful(())
} else {
Future.failure(new CommandException(s"task $name is not running"))
}
case RepeatingTaskCommand(name, delay, arg) =>
/* function to generate repeating future */
def createFuture(): Future[Unit] = {
runRepeatingTaskCommand(arg)).flatMap { _ =>
scheduler.repeatDelay(name) match {
case Some(d) =>
Thread.sleep(d)
Future(createFuture())
case None =>
Future.successful(())
}
}
}
/* spin off repeating task */
if (scheduler.isScheduled(name)) {
Future.failure(new CommandException(s"task $name is already running"))
} else {
scheduler.schedule(name, delay)
createFuture()
}
}
}
The code above will be integrated into some REPL that reads users input and invokes some of the following calls:
implicit val scheduler: Scheduler = new SchedulerImpl
runCommand(OneTimeTaskCommand(someArg))
// Run backgrounded task to be repeated every second
runCommand(RepeatingTaskCommand("backgrounded-task", 1000L, someArg))
// Stop backgrounded task
runCommand(StopTaskCommand("backgrounded-task"))

Related

Get return value from thread, is this Kotlin code thread safe?

I would like to run some treads, wait till all of them are finished and get the results.
Possible way to do that would be in the code below. Is it thread safe though?
import kotlin.concurrent.thread
sealed class Errorneous<R>
data class Success<R>(val result: R) : Errorneous<R>()
data class Fail<R>(val error: Exception) : Errorneous<R>()
fun <R> thread_with_result(fn: () -> R): (() -> Errorneous<R>) {
var r: Errorneous<R>? = null
val t = thread {
r = try { Success(fn()) } catch (e: Exception) { Fail(e) }
}
return {
t.join()
r!!
}
}
fun main() {
val tasks = listOf({ 2 * 2 }, { 3 * 3 })
val results = tasks
.map{ thread_with_result(it) }
.map{ it() }
println(results)
}
P.S.
Are there better built-in tools in Kotlin to do that? Like process 10000 tasks with pool of 10 threads?
It should be threads, not coroutines, as it will be used with legacy code and I don't know if it works well with coroutines.
Seems like Java has Executors that doing exactly that
fun <R> execute_in_parallel(tasks: List<() -> R>, threads: Int): List<Errorneous<R>> {
val executor = Executors.newFixedThreadPool(threads)
val fresults = executor.invokeAll(tasks.map { task ->
Callable<Errorneous<R>> {
try { Success(task()) } catch (e: Exception) { Fail(e) }
}
})
return fresults.map { future -> future.get() }
}

Kotlin Coroutines : Waiting for multiple threads to finish

So looking at Coroutines for the first time, I want to process a load of data in parallel and wait for it to finish. I been looking around and seen RunBlocking and Await etc but not sure how to use it.
I so far have
val jobs = mutableListOf<Job>()
jobs += GlobalScope.launch { processPages(urls, collection) }
jobs += GlobalScope.launch { processPages(urls, collection2) }
jobs += GlobalScope.launch { processPages(urls, collection3) }
I then want to know/wait for these to finish
You don't need to manually keep track of your cuncurrent jobs if you use the concept of structured concurrency. Assuming that your processPages function performs some kind of blocking IO, you can encapsulate your code into the following suspending function, which executes your code in an IO dispatcher designed for this kind of work:
suspend fun processAllPages() = withContext(Dispatchers.IO) {
// withContext waits for all children coroutines
launch { processPages(urls, collection) }
launch { processPages(urls, collection2) }
launch { processPages(urls, collection3) }
}
Now, from if a topmost function of your application is not already a suspending function, then you can use runBlocking to call processAllPages:
runBlocking {
processAllPages()
}
You can use async builder function to process a load of data in parallel:
class Presenter {
private var job: Job = Job()
private var scope = CoroutineScope(Dispatchers.Main + job) // creating the scope to run the coroutine. It consists of Dispatchers.Main (coroutine will run in the Main context) and job to handle the cancellation of the coroutine.
fun runInParallel() {
scope.launch { // launch a coroutine
// runs in parallel
val deferredList = listOf(
scope.asyncIO { processPages(urls, collection) },
scope.asyncIO { processPages(urls, collection2) },
scope.asyncIO { processPages(urls, collection3) }
)
deferredList.awaitAll() // wait for all data to be processed without blocking the UI thread
// do some stuff after data has been processed, for example update UI
}
}
private fun processPages(...) {...}
fun cancel() {
job.cancel() // invoke it to cancel the job when you don't need it to execute. For example when UI changed and you don't need to process data
}
}
Extension function asyncIO:
fun <T> CoroutineScope.asyncIO(ioFun: () -> T) = async(Dispatchers.IO) { ioFun() } // CoroutineDispatcher - runs and schedules coroutines
GlobalScope.launch is not recommended to use unless you want the coroutine to be operating on the whole application lifetime and not cancelled prematurely.
Edit: as mentioned by Roman Elizarov you can try not to use awaitAll() function unless you want to update UI or do something else right away after all data are processed.
Following approach can be used.
fun myTask() {
GlobalScope.launch {
val task = listOf(
async {
},
async {
}
)
task.awaitAll()
}
}

how to cap kotlin coroutines maximum concurrency

I've got a Sequence (from File.walkTopDown) and I need to run a long-running operation on each of them. I'd like to use Kotlin best practices / coroutines, but I either get no parallelism, or way too much parallelism and hit a "too many open files" IO error.
File("/Users/me/Pictures/").walkTopDown()
.onFail { file, ex -> println("ERROR: $file caused $ex") }
.filter { ... only big images... }
.map { file ->
async { // I *think* I want async and not "launch"...
ImageProcessor.fromFile(file)
}
}
This doesn't seem to run it in parallel, and my multi-core CPU never goes above 1 CPU's worth. Is there a way with coroutines to run "NumberOfCores parallel operations" worth of Deferred jobs?
I looked at Multithreading using Kotlin Coroutines which first creates ALL the jobs then joins them, but that means completing the Sequence/file tree walk completly bfore the heavy processing join step, and that seems... iffy! Splitting it into a collect and a process step means the collection could run way ahead of the processing.
val jobs = ... the Sequence above...
.toSet()
println("Found ${jobs.size}")
jobs.forEach { it.await() }
This isn't specific to your problem, but it does answer the question of, "how to cap kotlin coroutines maximum concurrency".
EDIT: As of kotlinx.coroutines 1.6.0 (https://github.com/Kotlin/kotlinx.coroutines/issues/2919), you can use limitedParallelism, e.g. Dispatchers.IO.limitedParallelism(123).
Old solution: I thought to use newFixedThreadPoolContext at first, but 1) it's deprecated and 2) it would use threads and I don't think that's necessary or desirable (same with Executors.newFixedThreadPool().asCoroutineDispatcher()). This solution might have flaws I'm not aware of by using Semaphore, but it's very simple:
import kotlinx.coroutines.async
import kotlinx.coroutines.awaitAll
import kotlinx.coroutines.coroutineScope
import kotlinx.coroutines.sync.Semaphore
import kotlinx.coroutines.sync.withPermit
/**
* Maps the inputs using [transform] at most [maxConcurrency] at a time until all Jobs are done.
*/
suspend fun <TInput, TOutput> Iterable<TInput>.mapConcurrently(
maxConcurrency: Int,
transform: suspend (TInput) -> TOutput,
) = coroutineScope {
val gate = Semaphore(maxConcurrency)
this#mapConcurrently.map {
async {
gate.withPermit {
transform(it)
}
}
}.awaitAll()
}
Tests (apologies, it uses Spek, hamcrest, and kotlin test):
import kotlinx.coroutines.ExperimentalCoroutinesApi
import kotlinx.coroutines.delay
import kotlinx.coroutines.launch
import kotlinx.coroutines.runBlocking
import kotlinx.coroutines.test.TestCoroutineDispatcher
import org.hamcrest.MatcherAssert.assertThat
import org.hamcrest.Matchers.greaterThanOrEqualTo
import org.hamcrest.Matchers.lessThanOrEqualTo
import org.spekframework.spek2.Spek
import org.spekframework.spek2.style.specification.describe
import java.util.concurrent.atomic.AtomicInteger
import kotlin.test.assertEquals
#OptIn(ExperimentalCoroutinesApi::class)
object AsyncHelpersKtTest : Spek({
val actionDelay: Long = 1_000 // arbitrary; obvious if non-test dispatcher is used on accident
val testDispatcher = TestCoroutineDispatcher()
afterEachTest {
// Clean up the TestCoroutineDispatcher to make sure no other work is running.
testDispatcher.cleanupTestCoroutines()
}
describe("mapConcurrently") {
it("should run all inputs concurrently if maxConcurrency >= size") {
val concurrentJobCounter = AtomicInteger(0)
val inputs = IntRange(1, 2).toList()
val maxConcurrency = inputs.size
// https://github.com/Kotlin/kotlinx.coroutines/issues/1266 has useful info & examples
runBlocking(testDispatcher) {
print("start runBlocking $coroutineContext\n")
// We have to run this async so that the code afterwards can advance the virtual clock
val job = launch {
testDispatcher.pauseDispatcher {
val result = inputs.mapConcurrently(maxConcurrency) {
print("action $it $coroutineContext\n")
// Sanity check that we never run more in parallel than max
assertThat(concurrentJobCounter.addAndGet(1), lessThanOrEqualTo(maxConcurrency))
// Allow for virtual clock adjustment
delay(actionDelay)
// Sanity check that we never run more in parallel than max
assertThat(concurrentJobCounter.getAndAdd(-1), lessThanOrEqualTo(maxConcurrency))
print("action $it after delay $coroutineContext\n")
it
}
// Order is not guaranteed, thus a Set
assertEquals(inputs.toSet(), result.toSet())
print("end mapConcurrently $coroutineContext\n")
}
}
print("before advanceTime $coroutineContext\n")
// Start the coroutines
testDispatcher.advanceTimeBy(0)
assertEquals(inputs.size, concurrentJobCounter.get(), "All jobs should have been started")
testDispatcher.advanceTimeBy(actionDelay)
print("after advanceTime $coroutineContext\n")
assertEquals(0, concurrentJobCounter.get(), "All jobs should have finished")
job.join()
}
}
it("should run one at a time if maxConcurrency = 1") {
val concurrentJobCounter = AtomicInteger(0)
val inputs = IntRange(1, 2).toList()
val maxConcurrency = 1
runBlocking(testDispatcher) {
val job = launch {
testDispatcher.pauseDispatcher {
inputs.mapConcurrently(maxConcurrency) {
assertThat(concurrentJobCounter.addAndGet(1), lessThanOrEqualTo(maxConcurrency))
delay(actionDelay)
assertThat(concurrentJobCounter.getAndAdd(-1), lessThanOrEqualTo(maxConcurrency))
it
}
}
}
testDispatcher.advanceTimeBy(0)
assertEquals(1, concurrentJobCounter.get(), "Only one job should have started")
val elapsedTime = testDispatcher.advanceUntilIdle()
print("elapsedTime=$elapsedTime")
assertThat(
"Virtual time should be at least as long as if all jobs ran sequentially",
elapsedTime,
greaterThanOrEqualTo(actionDelay * inputs.size)
)
job.join()
}
}
it("should handle cancellation") {
val jobCounter = AtomicInteger(0)
val inputs = IntRange(1, 2).toList()
val maxConcurrency = 1
runBlocking(testDispatcher) {
val job = launch {
testDispatcher.pauseDispatcher {
inputs.mapConcurrently(maxConcurrency) {
jobCounter.addAndGet(1)
delay(actionDelay)
it
}
}
}
testDispatcher.advanceTimeBy(0)
assertEquals(1, jobCounter.get(), "Only one job should have started")
job.cancel()
testDispatcher.advanceUntilIdle()
assertEquals(1, jobCounter.get(), "Only one job should have run")
job.join()
}
}
}
})
Per https://play.kotlinlang.org/hands-on/Introduction%20to%20Coroutines%20and%20Channels/09_Testing, you may also need to adjust compiler args for the tests to run:
compileTestKotlin {
kotlinOptions {
// Needed for runBlocking test coroutine dispatcher?
freeCompilerArgs += "-Xuse-experimental=kotlin.Experimental"
freeCompilerArgs += "-Xopt-in=kotlin.RequiresOptIn"
}
}
testImplementation 'org.jetbrains.kotlinx:kotlinx-coroutines-test:1.4.1'
The problem with your first snippet is that it doesn't run at all - remember, Sequence is lazy, and you have to use a terminal operation such as toSet() or forEach(). Additionally, you need to limit the number of threads that can be used for that task via constructing a newFixedThreadPoolContext context and using it in async:
val pictureContext = newFixedThreadPoolContext(nThreads = 10, name = "reading pictures in parallel")
File("/Users/me/Pictures/").walkTopDown()
.onFail { file, ex -> println("ERROR: $file caused $ex") }
.filter { ... only big images... }
.map { file ->
async(pictureContext) {
ImageProcessor.fromFile(file)
}
}
.toList()
.forEach { it.await() }
Edit:
You have to use a terminal operator (toList) befor awaiting the results
I got it working with a Channel. But maybe I'm being redundant with your way?
val pipe = ArrayChannel<Deferred<ImageFile>>(20)
launch {
while (!(pipe.isEmpty && pipe.isClosedForSend)) {
imageFiles.add(pipe.receive().await())
}
println("pipe closed")
}
File("/Users/me/").walkTopDown()
.onFail { file, ex -> println("ERROR: $file caused $ex") }
.forEach { pipe.send(async { ImageFile.fromFile(it) }) }
pipe.close()
This doesn't preserve the order of the projection but otherwise limits the throughput to at most maxDegreeOfParallelism. Expand and extend as you see fit.
suspend fun <TInput, TOutput> (Collection<TInput>).inParallel(
maxDegreeOfParallelism: Int,
action: suspend CoroutineScope.(input: TInput) -> TOutput
): Iterable<TOutput> = coroutineScope {
val list = this#inParallel
if (list.isEmpty())
return#coroutineScope listOf<TOutput>()
val brake = Channel<Unit>(maxDegreeOfParallelism)
val output = Channel<TOutput>()
val counter = AtomicInteger(0)
this.launch {
repeat(maxDegreeOfParallelism) {
brake.send(Unit)
}
for (input in list) {
val task = this.async {
action(input)
}
this.launch {
val result = task.await()
output.send(result)
val completed = counter.incrementAndGet()
if (completed == list.size) {
output.close()
} else brake.send(Unit)
}
brake.receive()
}
}
val results = mutableListOf<TOutput>()
for (item in output) {
results.add(item)
}
return#coroutineScope results
}
Example usage:
val output = listOf(1, 2, 3).inParallel(2) {
it + 1
} // Note that output may not be in same order as list.
Why not use the asFlow() operator and then use flatMapMerge?
someCoroutineScope.launch(Dispatchers.Default) {
File("/Users/me/Pictures/").walkTopDown()
.asFlow()
.filter { ... only big images... }
.flatMapMerge(concurrencyLimit) { file ->
flow {
emit(runInterruptable { ImageProcessor.fromFile(file) })
}
}.catch { ... }
.collect()
}
Then you can limit the simultaneous open files while still processing them concurrently.
To limit the parallelism to some value there is limitedParallelism function starting from the 1.6.0 version of the kotlinx.coroutines library. It can be called on CoroutineDispatcher object. So to limit threads for parallel execution we can write something like:
val parallelismLimit = Runtime.getRuntime().availableProcessors()
val limitedDispatcher = Dispatchers.Default.limitedParallelism(parallelismLimit)
val scope = CoroutineScope(limitedDispatcher) // we can set limitedDispatcher for the whole scope
scope.launch { // or we can set limitedDispatcher for a coroutine launch(limitedDispatcher)
File("/Users/me/Pictures/").walkTopDown()
.onFail { file, ex -> println("ERROR: $file caused $ex") }
.filter { ... only big images... }
.map { file ->
async {
ImageProcessor.fromFile(file)
}
}.toList().awaitAll()
}
ImageProcessor.fromFile(file) will be executed in parallel using parallelismLimit number of threads.
This will cap coroutines to workers. I'd recommend watching https://www.youtube.com/watch?v=3WGM-_MnPQA
package com.example.workers
import kotlinx.coroutines.*
import kotlinx.coroutines.channels.ReceiveChannel
import kotlinx.coroutines.channels.produce
import kotlin.system.measureTimeMillis
class ChannellibgradleApplication
fun main(args: Array<String>) {
var myList = mutableListOf<Int>(3000,1200,1400,3000,1200,1400,3000)
runBlocking {
var myChannel = produce(CoroutineName("MyInts")) {
myList.forEach { send(it) }
}
println("Starting coroutineScope ")
var time = measureTimeMillis {
coroutineScope {
var workers = 2
repeat(workers)
{
launch(CoroutineName("Sleep 1")) { theHardWork(myChannel) }
}
}
}
println("Ending coroutineScope $time ms")
}
}
suspend fun theHardWork(channel : ReceiveChannel<Int>)
{
for(m in channel) {
println("Starting Sleep $m")
delay(m.toLong())
println("Ending Sleep $m")
}
}

Scala future and its callback works in the same execution context

I call def activateReward by Akka actors and execution OracleClient.rewardActivate(user) sometimes is very slow (the database is outside of my responsibility and belongs to another company).
When database is slow the thread pool is exhausted and can not effectively allocate more threads to run callbacks future.onComplete because callbacks and futures works in the same execution context.
Please advise how to execute code in the callback asynchronously from threads which allocated for futures OracleClient.rewardActivate(user)
class RewardActivatorHelper {
private implicit val ec = new ExecutionContext {
val threadPool = Executors.newFixedThreadPool(1000)
def execute(runnable: Runnable) {threadPool.submit(runnable)}
def reportFailure(t: Throwable) {throw t}
}
case class FutureResult(spStart:Long, spFinish:Long)
def activateReward(msg:Msg, time:Long):Unit = {
msg.users.foreach {
user =>
val future:Future[FutureResult] = Future {
val (spStart, spFinish) = OracleClient.rewardActivate(user)
FutureResult(spStart, spFinish)
}
future.onComplete {
case Success(futureResult:FutureResult) =>
futureResult match {
case res:FutureResult => Logger.writeToLog(Logger.LogLevel.DEBUG,s"started:${res.spStart}finished:${res.spFinish}")
case _ => Logger.writeToLog(Logger.LogLevel.DEBUG, "some error")
}
case Failure(e:Throwable) => Logger.writeToLog(Logger.LogLevel.DEBUG, e.getMessage)
}
}
}
}
You can specify the execution context explicitly instead of implicitly for the onComplete callback by doing something along these lines:
import java.util.concurrent.Executors
import scala.concurrent.duration.Duration
object Example extends App {
import scala.concurrent._
private implicit val ec = new ExecutionContext {
val threadPool = Executors.newFixedThreadPool(1000)
def execute(runnable: Runnable) {threadPool.submit(runnable)}
def reportFailure(t: Throwable) {throw t}
}
val f = Future {
println("from future")
}
f.onComplete { _ =>
println("I'm done.")
}(scala.concurrent.ExecutionContext.Implicits.global)
Await.result(f, Duration.Inf)
}
This will of course not solve the underlying problem of a database not keeping up, but might be good to know anyway.
To clarify: I let the onComplete callback be handled by the standard global execution context. You might want to create a separate one.

Make an actor sleep

I want to make an actor sleep for a while, specifically it should decide whether to sleep itself depending on a condition:
class MyActor extends Actor {
def receive {
case "doWork" => doWork()
}
def doWork(): Unit = {
// doing some work
val condition = calculateCondition
if (condition) {
// sleep for 5 seconds
// Thread.sleep(5000)
}
}
}
I'm pretty much sure it's not a good thing to call Thread.sleep(5000) inside an actor and there should be another way. Therefore, how do I make it sleep?
I would look to do this using changes of state/behaviour for the Actor. Akka gives you a couple of means of doing this: you can implement a full-on state machine, or make use of context.become (and mix in akka.actor.Stash), and have the actor pass (scheduled) messages to itself. The former feels like overkill for this case, so here is how I would look to code it up:
import akka.actor._
import scala.concurrent.duration._
class MySleepyActor(duration: FiniteDuration = (5 seconds)) extends Actor with Stash {
import context._
override def preStart() { become(running) }
def receive = PartialFunction.empty
def running: Actor.Receive = {
case "doWork" =>
if (doWork()) {
scheduleReactivate
become(paused)
}
case "wakeUp" => // already awake
}
def paused: Actor.Receive = {
case "doWork" => stash()
case "wakeUp" =>
unstashAll()
become(running)
}
def scheduleReactivate: Unit = {
system.scheduler.scheduleOnce(duration, self, "wakeUp")
}
def doWork(): Boolean = {
// doing some work, then:
calculateCondition
}
}
Note: I have not tested this code! Should give you some ideas to work with, though.

Resources