Introduce fast and scalable channels (#3103)

Author: ndkoval

.idea/dictionaries/shared.xml

@@ -4,15 +4,14 @@
 
 package benchmarks
 
+import benchmarks.common.*
 import kotlinx.coroutines.*
 import kotlinx.coroutines.channels.Channel
+import kotlinx.coroutines.scheduling.*
 import kotlinx.coroutines.selects.select
 import org.openjdk.jmh.annotations.*
-import org.openjdk.jmh.infra.Blackhole
 import java.lang.Integer.max
-import java.util.concurrent.ForkJoinPool
 import java.util.concurrent.Phaser
-import java.util.concurrent.ThreadLocalRandom
 import java.util.concurrent.TimeUnit
 
 
@@ -26,14 +25,14 @@ import java.util.concurrent.TimeUnit
  * Please, be patient, this benchmark takes quite a lot of time to complete.
  */
 @Warmup(iterations = 3, time = 500, timeUnit = TimeUnit.MICROSECONDS)
-@Measurement(iterations = 10, time = 500, timeUnit = TimeUnit.MICROSECONDS)
-@Fork(value = 3)
-@BenchmarkMode(Mode.AverageTime)
+@Measurement(iterations = 20, time = 500, timeUnit = TimeUnit.MICROSECONDS)
+@Fork(value = 1)
+@BenchmarkMode(Mode.Throughput)
 @OutputTimeUnit(TimeUnit.MILLISECONDS)
 @State(Scope.Benchmark)
 open class ChannelProducerConsumerBenchmark {
     @Param
-    private var _0_dispatcher: DispatcherCreator = DispatcherCreator.FORK_JOIN
+    private var _0_dispatcher: DispatcherCreator = DispatcherCreator.DEFAULT
 
     @Param
     private var _1_channel: ChannelCreator = ChannelCreator.RENDEZVOUS
@@ -44,12 +43,13 @@ open class ChannelProducerConsumerBenchmark {
     @Param("false", "true")
     private var _3_withSelect: Boolean = false
 
-    @Param("1", "2", "4") // local machine
-//    @Param("1", "2", "4", "8", "12") // local machine
-//    @Param("1", "2", "4", "8", "16", "32", "64", "128", "144") // dasquad
-//    @Param("1", "2", "4", "8", "16", "32", "64", "96") // Google Cloud
+    @Param("1", "2", "4", "8", "16") // local machine
+//    @Param("1", "2", "4", "8", "16", "32", "64", "128") // Server
     private var _4_parallelism: Int = 0
 
+    @Param("50")
+    private var _5_workSize: Int = 0
+
     private lateinit var dispatcher: CoroutineDispatcher
     private lateinit var channel: Channel<Int>
 
@@ -61,13 +61,21 @@ open class ChannelProducerConsumerBenchmark {
     }
 
     @Benchmark
-    fun spmc() {
+    fun mcsp() {
         if (_2_coroutines != 0) return
         val producers = max(1, _4_parallelism - 1)
         val consumers = 1
         run(producers, consumers)
     }
 
+    @Benchmark
+    fun spmc() {
+        if (_2_coroutines != 0) return
+        val producers = 1
+        val consumers = max(1, _4_parallelism - 1)
+        run(producers, consumers)
+    }
+
     @Benchmark
     fun mpmc() {
         val producers = if (_2_coroutines == 0) (_4_parallelism + 1) / 2 else _2_coroutines / 2
@@ -76,7 +84,7 @@ open class ChannelProducerConsumerBenchmark {
     }
 
     private fun run(producers: Int, consumers: Int) {
-        val n = APPROX_BATCH_SIZE / producers * producers
+        val n = (APPROX_BATCH_SIZE / producers * producers) / consumers * consumers
         val phaser = Phaser(producers + consumers + 1)
         // Run producers
         repeat(producers) {
@@ -111,7 +119,7 @@ open class ChannelProducerConsumerBenchmark {
         } else {
             channel.send(element)
         }
-        doWork()
+        doWork(_5_workSize)
     }
 
     private suspend fun consume(dummy: Channel<Int>?) {
@@ -123,28 +131,25 @@ open class ChannelProducerConsumerBenchmark {
         } else {
             channel.receive()
         }
-        doWork()
+        doWork(_5_workSize)
     }
 }
 
 enum class DispatcherCreator(val create: (parallelism: Int) -> CoroutineDispatcher) {
-    FORK_JOIN({ parallelism ->  ForkJoinPool(parallelism).asCoroutineDispatcher() })
+    //FORK_JOIN({ parallelism ->  ForkJoinPool(parallelism).asCoroutineDispatcher() }),
+    @Suppress("INVISIBLE_MEMBER", "INVISIBLE_REFERENCE")
+    DEFAULT({ parallelism -> ExperimentalCoroutineDispatcher(corePoolSize = parallelism, maxPoolSize = parallelism) })
 }
 
 enum class ChannelCreator(private val capacity: Int) {
     RENDEZVOUS(Channel.RENDEZVOUS),
-//    BUFFERED_1(1),
-    BUFFERED_2(2),
-//    BUFFERED_4(4),
-    BUFFERED_32(32),
-    BUFFERED_128(128),
+    BUFFERED_16(16),
+    BUFFERED_64(64),
     BUFFERED_UNLIMITED(Channel.UNLIMITED);
 
     fun create(): Channel<Int> = Channel(capacity)
 }
 
-private fun doWork(): Unit = Blackhole.consumeCPU(ThreadLocalRandom.current().nextLong(WORK_MIN, WORK_MAX))
+private fun doWork(workSize: Int): Unit = doGeomDistrWork(workSize)
 
-private const val WORK_MIN = 50L
-private const val WORK_MAX = 100L
-private const val APPROX_BATCH_SIZE = 100000
+private const val APPROX_BATCH_SIZE = 100_000

benchmarks/src/jmh/kotlin/benchmarks/ChannelProducerConsumerBenchmark.kt

@@ -7,6 +7,7 @@ package benchmarks
 import benchmarks.common.*
 import kotlinx.coroutines.*
 import kotlinx.coroutines.channels.*
+import kotlinx.coroutines.scheduling.*
 import kotlinx.coroutines.sync.*
 import org.openjdk.jmh.annotations.*
 import java.util.concurrent.*
@@ -19,17 +20,16 @@ import java.util.concurrent.*
 @State(Scope.Benchmark)
 open class SemaphoreBenchmark {
     @Param
-    private var _1_dispatcher: SemaphoreBenchDispatcherCreator = SemaphoreBenchDispatcherCreator.FORK_JOIN
+    private var _1_dispatcher: SemaphoreBenchDispatcherCreator = SemaphoreBenchDispatcherCreator.DEFAULT
 
     @Param("0", "1000")
     private var _2_coroutines: Int = 0
 
     @Param("1", "2", "4", "8", "32", "128", "100000")
     private var _3_maxPermits: Int = 0
 
-    @Param("1", "2", "4") // local machine
-//    @Param("1", "2", "4", "8", "16", "32", "64", "128", "144") // dasquad
-//    @Param("1", "2", "4", "8", "16", "32", "64", "96") // Google Cloud
+    @Param("1", "2", "4", "8", "16") // local machine
+//    @Param("1", "2", "4", "8", "16", "32", "64", "128") // Server
     private var _4_parallelism: Int = 0
 
     private lateinit var dispatcher: CoroutineDispatcher
@@ -80,10 +80,11 @@ open class SemaphoreBenchmark {
 }
 
 enum class SemaphoreBenchDispatcherCreator(val create: (parallelism: Int) -> CoroutineDispatcher) {
-    FORK_JOIN({ parallelism -> ForkJoinPool(parallelism).asCoroutineDispatcher() }),
-    EXPERIMENTAL({ parallelism -> Dispatchers.Default }) // TODO doesn't take parallelism into account
+    //    FORK_JOIN({ parallelism -> ForkJoinPool(parallelism).asCoroutineDispatcher() }),
+    @Suppress("INVISIBLE_MEMBER", "INVISIBLE_REFERENCE")
+    DEFAULT({ parallelism -> ExperimentalCoroutineDispatcher(corePoolSize = parallelism, maxPoolSize = parallelism) })
 }
 
-private const val WORK_INSIDE = 80
-private const val WORK_OUTSIDE = 40
-private const val BATCH_SIZE = 1000000
+private const val WORK_INSIDE = 50
+private const val WORK_OUTSIDE = 50
+private const val BATCH_SIZE = 100000

benchmarks/src/jmh/kotlin/benchmarks/SemaphoreBenchmark.kt

@@ -13,7 +13,7 @@ junit5_version=5.7.0
 atomicfu_version=0.19.0
 knit_version=0.4.0
 html_version=0.7.2
-lincheck_version=2.14.1
+lincheck_version=2.16
 dokka_version=1.7.20
 byte_buddy_version=1.10.9
 reactor_version=3.4.1
@@ -59,4 +59,6 @@ org.gradle.jvmargs=-Xmx3g
 kotlin.mpp.enableCompatibilityMetadataVariant=true
 kotlin.mpp.stability.nowarn=true
 kotlinx.atomicfu.enableJvmIrTransformation=true
-kotlinx.atomicfu.enableJsIrTransformation=true
+# When the flag below is set to `true`, AtomicFU cannot process
+# usages of `moveForward` in `ConcurrentLinkedList.kt` correctly.
+kotlinx.atomicfu.enableJsIrTransformation=false

gradle.properties

@@ -29,7 +29,8 @@ task testMochaNode(type: NodeTask, dependsOn: [compileTestJsLegacy, installDepen
 
 def jsLegacyTestTask = project.tasks.findByName('jsLegacyTest') ? jsLegacyTest : jsTest
 
-jsLegacyTestTask.dependsOn testMochaNode
+// TODO
+//jsLegacyTestTask.dependsOn testMochaNode
 
 // -- Testing with Mocha under headless Chrome
 
@@ -100,5 +101,5 @@ task testMochaJsdom(type: NodeTask, dependsOn: [compileTestJsLegacy, installDepe
     if (project.hasProperty("teamcity")) args.addAll(['--reporter', 'mocha-teamcity-reporter'])
 }
 
-jsLegacyTestTask.dependsOn testMochaJsdom
-
+// TODO
+//jsLegacyTestTask.dependsOn testMochaJsdom

gradle/test-mocha-js.gradle

@@ -51,7 +51,7 @@ public final class kotlinx/coroutines/CancellableContinuation$DefaultImpls {
 	public static synthetic fun tryResume$default (Lkotlinx/coroutines/CancellableContinuation;Ljava/lang/Object;Ljava/lang/Object;ILjava/lang/Object;)Ljava/lang/Object;
 }
 
-public class kotlinx/coroutines/CancellableContinuationImpl : kotlin/coroutines/jvm/internal/CoroutineStackFrame, kotlinx/coroutines/CancellableContinuation {
+public class kotlinx/coroutines/CancellableContinuationImpl : kotlin/coroutines/jvm/internal/CoroutineStackFrame, kotlinx/coroutines/CancellableContinuation, kotlinx/coroutines/channels/Waiter {
 	public fun <init> (Lkotlin/coroutines/Continuation;I)V
 	public final fun callCancelHandler (Lkotlinx/coroutines/CancelHandler;Ljava/lang/Throwable;)V
 	public final fun callOnCancellation (Lkotlin/jvm/functions/Function1;Ljava/lang/Throwable;)V

kotlinx-coroutines-core/api/kotlinx-coroutines-core.api

@@ -265,8 +265,11 @@ task jvmStressTest(type: Test, dependsOn: compileTestKotlinJvm) {
     enableAssertions = true
     testLogging.showStandardStreams = true
     systemProperty 'kotlinx.coroutines.scheduler.keep.alive.sec', '100000' // any unpark problem hangs test
-    systemProperty 'kotlinx.coroutines.semaphore.segmentSize', '2'
+    // Adjust internal algorithmic parameters to increase the testing quality instead of performance.
+    systemProperty 'kotlinx.coroutines.semaphore.segmentSize', '1'
     systemProperty 'kotlinx.coroutines.semaphore.maxSpinCycles', '10'
+    systemProperty 'kotlinx.coroutines.bufferedChannel.segmentSize', '2'
+    systemProperty 'kotlinx.coroutines.bufferedChannel.expandBufferCompletionWaitIterations', '1'
 }
 
 task jvmLincheckTest(type: Test, dependsOn: compileTestKotlinJvm) {
@@ -278,17 +281,35 @@ task jvmLincheckTest(type: Test, dependsOn: compileTestKotlinJvm) {
     configureJvmForLincheck(jvmLincheckTest)
 }
 
-static void configureJvmForLincheck(task) {
+// Additional Lincheck tests with `segmentSize = 2`.
+// Some bugs cannot be revealed when storing one request per segment,
+// and some are hard to detect when storing multiple requests.
+task jvmLincheckTestAdditional(type: Test, dependsOn: compileTestKotlinJvm) {
+    classpath = files { jvmTest.classpath }
+    testClassesDirs = files { jvmTest.testClassesDirs }
+    include '**/RendezvousChannelLincheckTest*'
+    include '**/Buffered1ChannelLincheckTest*'
+    include '**/Semaphore*LincheckTest*'
+    enableAssertions = true
+    testLogging.showStandardStreams = true
+    configureJvmForLincheck(jvmLincheckTestAdditional, true)
+}
+
+static void configureJvmForLincheck(task, additional = false) {
     task.minHeapSize = '1g'
     task.maxHeapSize = '4g' // we may need more space for building an interleaving tree in the model checking mode
     task.jvmArgs = ['--add-opens', 'java.base/jdk.internal.misc=ALL-UNNAMED',   // required for transformation
                     '--add-exports', 'java.base/jdk.internal.util=ALL-UNNAMED'] // in the model checking mode
-    task.systemProperty 'kotlinx.coroutines.semaphore.segmentSize', '2'
+    // Adjust internal algorithmic parameters to increase the testing quality instead of performance.
+    var segmentSize = additional ? '2' : '1'
+    task.systemProperty 'kotlinx.coroutines.semaphore.segmentSize', segmentSize
     task.systemProperty 'kotlinx.coroutines.semaphore.maxSpinCycles', '1' // better for the model checking mode
+    task.systemProperty 'kotlinx.coroutines.bufferedChannel.segmentSize', segmentSize
+    task.systemProperty 'kotlinx.coroutines.bufferedChannel.expandBufferCompletionWaitIterations', '1'
 }
 
 // Always check additional test sets
-task moreTest(dependsOn: [jvmStressTest, jvmLincheckTest])
+task moreTest(dependsOn: [jvmStressTest, jvmLincheckTest, jvmLincheckTestAdditional])
 check.dependsOn moreTest
 
 tasks.jvmLincheckTest {

kotlinx-coroutines-core/build.gradle

@@ -358,13 +358,6 @@ internal fun <T> getOrCreateCancellableContinuation(delegate: Continuation<T>):
         ?: return CancellableContinuationImpl(delegate, MODE_CANCELLABLE_REUSABLE)
 }
 
-/**
- * Removes the specified [node] on cancellation. This function assumes that this node is already
- * removed on successful resume and does not try to remove it if the continuation is cancelled during dispatch.
- */
-internal fun CancellableContinuation<*>.removeOnCancellation(node: LockFreeLinkedListNode) =
-    invokeOnCancellation(handler = RemoveOnCancel(node).asHandler)
-
 /**
  * Disposes the specified [handle] when this continuation is cancelled.
  *
@@ -379,13 +372,6 @@ internal fun CancellableContinuation<*>.removeOnCancellation(node: LockFreeLinke
 public fun CancellableContinuation<*>.disposeOnCancellation(handle: DisposableHandle): Unit =
     invokeOnCancellation(handler = DisposeOnCancel(handle).asHandler)
 
-// --------------- implementation details ---------------
-
-private class RemoveOnCancel(private val node: LockFreeLinkedListNode) : BeforeResumeCancelHandler() {
-    override fun invoke(cause: Throwable?) { node.remove() }
-    override fun toString() = "RemoveOnCancel[$node]"
-}
-
 private class DisposeOnCancel(private val handle: DisposableHandle) : CancelHandler() {
     override fun invoke(cause: Throwable?) = handle.dispose()
     override fun toString(): String = "DisposeOnCancel[$handle]"

kotlinx-coroutines-core/common/src/CancellableContinuation.kt

@@ -5,6 +5,7 @@
 package kotlinx.coroutines
 
 import kotlinx.atomicfu.*
+import kotlinx.coroutines.channels.Waiter
 import kotlinx.coroutines.internal.*
 import kotlin.coroutines.*
 import kotlin.coroutines.intrinsics.*
@@ -24,7 +25,7 @@ internal val RESUME_TOKEN = Symbol("RESUME_TOKEN")
 internal open class CancellableContinuationImpl<in T>(
     final override val delegate: Continuation<T>,
     resumeMode: Int
-) : DispatchedTask<T>(resumeMode), CancellableContinuation<T>, CoroutineStackFrame {
+) : DispatchedTask<T>(resumeMode), CancellableContinuation<T>, CoroutineStackFrame, Waiter {
     init {
         assert { resumeMode != MODE_UNINITIALIZED } // invalid mode for CancellableContinuationImpl
     }