Disruptor 高性能队列

T-note

2019-08-13

背景

记录一下 disruptor 的学习笔记

Disruptor 例子

import java.util.concurrent.ThreadFactory
import com.lmax.disruptor.dsl.{Disruptor, ProducerType}
import com.lmax.disruptor.{BlockingWaitStrategy,EventFactory,EventHandler,EventTranslatorOneArg,WaitStrategy}

object DisruptorTest {

  val disruptor = {
    val factory = new EventFactory[Event] {
      override def newInstance(): Event = Event(-1)
    }

    val threadFactory = new ThreadFactory(){
      override def newThread(r: Runnable): Thread = new Thread(r)
    }
    
    val disruptor = new Disruptor[Event](factory, 4, threadFactory, ProducerType.SINGLE, new BlockingWaitStrategy())

    disruptor.handleEventsWith(TestHandler).`then`(ThenHandler)
    
    disruptor
  }
  
  val translator = new EventTranslatorOneArg[Event, Int]() {
    override def translateTo(event: Event, sequence: Long, arg: Int): Unit = {
      event.id = arg
      println(s"translator: ${event}, sequence: ${sequence}, arg: ${arg}")
    }
  }

  def main(args: Array[String]): Unit = {
    disruptor.start()
    (0 until 100).foreach { i =>
      disruptor.publishEvent(translator, i)
    }
    disruptor.shutdown()
  }
}

case class Event(var id: Int) {
  override def toString: String = s"event: ${id}"
}

object TestHandler extends EventHandler[Event] {
  override def onEvent(event: Event, sequence: Long, endOfBatch: Boolean): Unit = {
    println(s"${this.getClass.getSimpleName} ${System.nanoTime()} ${event}")
  }
}

object ThenHandler extends EventHandler[Event] {
  override def onEvent(event: Event, sequence: Long, endOfBatch: Boolean): Unit = {
    println(s"${this.getClass.getSimpleName} ${System.nanoTime()} ${event}")
  }
}

源码阅读

disrutpor 初始化

先看 Disruptor 构造方法

public Disruptor(final EventFactory<T> eventFactory, 
  final int ringBufferSize, 
  final ThreadFactory threadFactory, 
  final ProducerType producerType,
  final WaitStrategy waitStrategy) {
    this(RingBuffer.create(producerType, eventFactory, ringBufferSize, waitStrategy), 
        new BasicExecutor(threadFactory));
}

在看 RingBuffer.create, 最终通过 fill 方法将 eventFactory.newInstance() 作为默认值，塞到 ringBuffer 里面

public static <E> RingBuffer<E> create(ProducerType producerType, 
  EventFactory<E> factory, int bufferSize, WaitStrategy waitStrategy) {
    switch (producerType) {
        case SINGLE:
            return createSingleProducer(factory, bufferSize, waitStrategy);
        case MULTI:
            return createMultiProducer(factory, bufferSize, waitStrategy);
        default:
            throw new IllegalStateException(producerType.toString());
    }
}

public static <E> RingBuffer<E> createSingleProducer(EventFactory<E> factory, int bufferSize, 
    WaitStrategy waitStrategy) {
    SingleProducerSequencer sequencer = new SingleProducerSequencer(bufferSize, waitStrategy);

    return new RingBuffer<E>(factory, sequencer);
}

RingBufferFields(EventFactory<E> eventFactory, Sequencer sequencer) {
    this.sequencer = sequencer;
    this.bufferSize = sequencer.getBufferSize();

    if (bufferSize < 1) {
        throw new IllegalArgumentException("bufferSize must not be less than 1");
    }
    if (Integer.bitCount(bufferSize) != 1) {
        throw new IllegalArgumentException("bufferSize must be a power of 2");
    }

    this.indexMask = bufferSize - 1;
    this.entries = new Object[sequencer.getBufferSize() + 2 * BUFFER_PAD];
    fill(eventFactory);
}

private void fill(EventFactory<E> eventFactory) {
    for (int i = 0; i < bufferSize; i++) {
        entries[BUFFER_PAD + i] = eventFactory.newInstance();
    }
}

消费事件消息

首先看 disruptor.start(): 消费事件消息入口

private final ConsumerRepository<T> consumerRepository = new ConsumerRepository<>();

public RingBuffer<T> start() {
    checkOnlyStartedOnce();
    for (final ConsumerInfo consumerInfo : consumerRepository) {
        consumerInfo.start(executor);
    }

    return ringBuffer;
}

consumerRepository 类型由 disruptor.handleEventsWith(TestHandler) 初始化, 并构造事件消息处理链

public final EventHandlerGroup<T> handleEventsWith(final EventHandler<? super T>... handlers) {
    return createEventProcessors(new Sequence[0], handlers);
}

EventHandlerGroup<T> createEventProcessors(final Sequence[] barrierSequences, final EventHandler<? super T>[] eventHandlers) {
    checkNotStarted();

    final Sequence[] processorSequences = new Sequence[eventHandlers.length];
    final SequenceBarrier barrier = ringBuffer.newBarrier(barrierSequences);

    for (int i = 0, eventHandlersLength = eventHandlers.length; i < eventHandlersLength; i++) {
        final EventHandler<? super T> eventHandler = eventHandlers[i];

        final BatchEventProcessor<T> batchEventProcessor = new BatchEventProcessor<>(ringBuffer, barrier, eventHandler);

        if (exceptionHandler != null) {
            batchEventProcessor.setExceptionHandler(exceptionHandler);
        }

        consumerRepository.add(batchEventProcessor, eventHandler, barrier);
        processorSequences[i] = batchEventProcessor.getSequence();
    }

    updateGatingSequencesForNextInChain(barrierSequences, processorSequences);

    return new EventHandlerGroup<>(this, consumerRepository, processorSequences);
}

回头看 disruptor.start() 中的 consumerInfo.start(executor)
executor = new BasicExecutor(threadFactory)，BasicExecutor 在每次 execute 任务时，每个 consumer 被分配一个 new thread

public Disruptor(
        final EventFactory<T> eventFactory,
        final int ringBufferSize,
        final ThreadFactory threadFactory,
        final ProducerType producerType,
        final WaitStrategy waitStrategy) {
    this(
        RingBuffer.create(producerType, eventFactory, ringBufferSize, waitStrategy),
        new BasicExecutor(threadFactory));
}

private Disruptor(final RingBuffer<T> ringBuffer, final Executor executor) {
    this.ringBuffer = ringBuffer;
    this.executor = executor;
}

@Override
public void start(final java.util.concurrent.Executor executor){
    //EventProcessor extends Runnable
    //executor = BasicExecutor 
    executor.execute(eventprocessor);
}

public final class BatchEventProcessor<T> implements EventProcessor {
  @Override
  public void run() {
      if (running.compareAndSet(IDLE, RUNNING)) {
          sequenceBarrier.clearAlert();

          notifyStart();
          try {
              if (running.get() == RUNNING) {
                  processEvents();
              }
          } finally {
              notifyShutdown();
              running.set(IDLE);
          }
      } else {
          if (running.get() == RUNNING) {
              throw new IllegalStateException("Thread is already running");
          } else {
              earlyExit();
          }
      }
  }
}

private void processEvents() {
    T event = null;
    long nextSequence = sequence.get() + 1L;

    while (true) {
        try {
            final long availableSequence = sequenceBarrier.waitFor(nextSequence);
            if (batchStartAware != null) {
                batchStartAware.onBatchStart(availableSequence - nextSequence + 1);
            }

            while (nextSequence <= availableSequence) {
                event = dataProvider.get(nextSequence);
                eventHandler.onEvent(event, nextSequence, nextSequence == availableSequence);
                nextSequence++;
            }

            sequence.set(availableSequence);
        } catch (final TimeoutException e) {
            notifyTimeout(sequence.get());
        } catch (final AlertException ex) {
            if (running.get() != RUNNING) {
                break;
            }
        } catch (final Throwable ex) {
            exceptionHandler.handleEventException(ex, nextSequence, event);
            sequence.set(nextSequence);
            nextSequence++;
        }
    }
}

executor.execute 也就是 BasicExecutor.execute(eventHandler) 会异步的执行 eventHandler, 也就是调用 BatchEventProcessor.run 方法

问题来了，既然是异步执行，多个 eventHandler 是怎么按照顺序去处理事件消息的？

我们看 processEvents 方法执行逻辑

先获取 BatchEventProcessor.sequence 并 +1
通过 sequenceBarrier.waitFor 也就是 WaitStrategy.waitFor 获取到可用的 availableSequence

先看下 BlockingWaitStrategy.waitFor 的实现

 public long waitFor(long sequence, Sequence cursorSequence, Sequence dependentSequence, 
    SequenceBarrier barrier)
    throws AlertException, InterruptedException {
    long availableSequence;
    if (cursorSequence.get() < sequence) {
        lock.lock();
        try {
            while (cursorSequence.get() < sequence) {
                barrier.checkAlert();
                processorNotifyCondition.await();
            }
        }
        finally {
            lock.unlock();
        }
    }

    while ((availableSequence = dependentSequence.get()) < sequence) {
        barrier.checkAlert();
    }

    return availableSequence;
}

如果 cursorSequence(ringbuffer 的索引) < sequence(batchEventProcessor 的索引) 则batchEventProcessor挂起等待
否则就用 dependentSequence 作为 availableSequence 返回
然后 batchEventProcessor 会将 availableSequence 索引之前的数据一次性处理完，并更新自身的 sequence 索引值

dependentSequence 由 ProcessingSequenceBarrier 构造方法初始化

final class ProcessingSequenceBarrier implements SequenceBarrier {
    private final WaitStrategy waitStrategy;
    private final Sequence dependentSequence;
    private volatile boolean alerted = false;
    private final Sequence cursorSequence;
    private final Sequencer sequencer;

    ProcessingSequenceBarrier(final Sequencer sequencer, final WaitStrategy waitStrategy,
        final Sequence cursorSequence, final Sequence[] dependentSequences) {
        this.sequencer = sequencer;
        this.waitStrategy = waitStrategy;
        this.cursorSequence = cursorSequence;
        if (0 == dependentSequences.length) {
            dependentSequence = cursorSequence;
        } else {
            dependentSequence = new FixedSequenceGroup(dependentSequences);
        }
    }
}

在 Disruptor.createEventProcessors 中的, 进行了初始化 ProcessingSequenceBarrier
final SequenceBarrier barrier = ringBuffer.newBarrier(barrierSequences)
createEventProcessors 仅会被 Disruptor.handleEventsWith 和 EventHandlerGroup.handleEventsWith

public class Disruptor<T> {
    public final EventHandlerGroup<T> handleEventsWith(final EventHandler<? super T>... handlers) {
        return createEventProcessors(new Sequence[0], handlers);
    }

    EventHandlerGroup<T> createEventProcessors(final Sequence[] barrierSequences,
        final EventHandler<? super T>[] eventHandlers) {
        checkNotStarted();

        final Sequence[] processorSequences = new Sequence[eventHandlers.length];
        final SequenceBarrier barrier = ringBuffer.newBarrier(barrierSequences);

        for (int i = 0, eventHandlersLength = eventHandlers.length; i < eventHandlersLength; i++) {
            final EventHandler<? super T> eventHandler = eventHandlers[i];

            final BatchEventProcessor<T> batchEventProcessor = 
                new BatchEventProcessor<>(ringBuffer, barrier, eventHandler);

            if (exceptionHandler != null) {
                batchEventProcessor.setExceptionHandler(exceptionHandler);
            }

            consumerRepository.add(batchEventProcessor, eventHandler, barrier);
            processorSequences[i] = batchEventProcessor.getSequence();
        }

        updateGatingSequencesForNextInChain(barrierSequences, processorSequences);

        return new EventHandlerGroup<>(this, consumerRepository, processorSequences);
    }
}

public class EventHandlerGroup<T> {
    private final Disruptor<T> disruptor;
    private final ConsumerRepository<T> consumerRepository;
    private final Sequence[] sequences;

    EventHandlerGroup(final Disruptor<T> disruptor, final ConsumerRepository<T> consumerRepository,
        final Sequence[] sequences) {
        this.disruptor = disruptor;
        this.consumerRepository = consumerRepository;
        this.sequences = Arrays.copyOf(sequences, sequences.length);
    }

    public final EventHandlerGroup<T> handleEventsWith(final EventHandler<? super T>... handlers) {
        return disruptor.createEventProcessors(sequences, handlers);
    }

    public final EventHandlerGroup<T> then(final EventHandler<? super T>... handlers) {
        return handleEventsWith(handlers);
    }
}

EventHandlerGroup 会拷贝一份 batchEventProcessor 中的 sequence
demo 例子中 disruptor.handleEventsWith(TestHandler).then(ThenHandler)
通过 then 方法将 TestHandler 中的 sequence 传递给 ThenHandler
这样 ThenHandler 就依赖了 TestHandler, ThenHandler 就会在 TestHandler 后执行

生产事件消息

接着看 disruptor.publishEvent(translator, i) 就是往 ringBuffer 里面放数据,

public <A> void publishEvent(EventTranslatorOneArg<E, A> translator, A arg0) {
    final long sequence = sequencer.next();
    translateAndPublish(translator, sequence, arg0);
}

private <A> void translateAndPublish(EventTranslatorOneArg<E, A> translator, long sequence, A arg0) {
    try {
        translator.translateTo(get(sequence), sequence, arg0);
    } finally {
        sequencer.publish(sequence);
    }
}

public E get(long sequence) {
    return elementAt(sequence);
}

get(sequence) 根据 sequence [ringbuffer 索引] 获取 ringbuffer 数组里的对象
translator 将其处理替换完后，ringbuffer 数组的的值将是新的值，publish 将会更新索引的标记位
waitStrategy.signalAllWhenBlocking() 会通知阻塞等待的消费者去继续消费消息

protected final Sequence cursor = new Sequence(Sequencer.INITIAL_CURSOR_VALUE);
@Override
public void publish(long sequence) {
    cursor.set(sequence);
    waitStrategy.signalAllWhenBlocking();
}

总结

流程理清楚了，我们看看知识点

ringbuffer
- 内存使用率很高，不会造成内存碎片，几乎没有浪费。业务处理的同一时间，访问的内存数据段集中。
  可以更好的适应不同系统，取得较高的性能。内存的物理布局简单单一，不太容易发生内存越界、悬空指针等 bug，出了问题也容易在内存级别分析调试。
  做出来的系统容易保持健壮。
cpu cache
- CPU 访问内存时会等待，导致计算资源大量闲置，降低 CPU 整体吞吐量。
  由于内存数据访问的热点集中性，在 CPU 和内存之间用较为快速而成本较高（相对于内存）的介质做一层缓存，就显得性价比极高了

常见问题

distruptor 框架导致 CPU 高
- 当消费者的消费速度大于生产者时，消费者有不同的 WaitStrategy 进行等待, 构建 disruptor 的时候，可以指定等待策略
  1
  new Disruptor[Event](factory, 4, threadFactory, ProducerType.SINGLE, new BlockingWaitStrategy())
  下面是 WaitStrategy 的一些实现类的简介 对应类上的comment
  - BusySpinWaitStrategy
    - 将线程绑定在 CPU 核上，减少线程切换，轮训获取 ringbuffer 上的数据
  - PhasedBackoffWaitStrategy
    - 通过先自旋，后 yield 让出 CPU，最后通过自定义的 fallbackStrategy 来实现当代
  - BlockingWaitStrategy
    - 通过 lock 来实现等待，ringbuffer中有消息后会通过 signalAll 来唤起锁进行消费
  - TimeoutBlockingWaitStrategy
    - 通过带有超时时间的 lock 来实现等待，过了超时时间会 throw exception
  - SleepingWaitStrategy
    - 通过先自旋，后 yield 让出 CPU，最后 lock 的方式来进行等待，这个锁可以减少锁的唤起带来的损耗
  - YieldingWaitStrategy
    - 通过先自旋，后 yield 让出 CPU 的方式来进行等待

参考资料

https://github.com/LMAX-Exchange/disruptor/wiki/Introduction