Disruptor 框架源码阅读 - 如何不重复消费 - 今日头条

本文由 简悦 SimpRead 转码, 原文地址 www.toutiao.com

RingBuffer 如何保证数据不丢失由于 ringbuffer 是一个环形的队列,那么生产者和消费者在遍历这个队列的时候,如何制衡呢? 1、生产

1、生产快,消费慢,数据丢失?

生产者速度过快,导致一个对象还没消费完,就循环生产了一个新的对象要加入 ringbuffer,导致消费不完整,造成数据丢失?

我们注意到,在我们获取生产者下一个位置的时候,是通过 ringbuffer 的 next 方法,而这个 next 方式是调用了 sequencer 的 next 方法

https://p26.toutiaoimg.com/origin/pgc-image/e0d28eb6abcc4513a48ca0e4bde5eee0?from=pc

https://p26.toutiaoimg.com/origin/pgc-image/7575516721d5493a9a89222df291c7b2?from=pc

这个对象,在我们创建 disruptor 对象的时候,创建的

https://p26.toutiaoimg.com/origin/pgc-image/456f8fd2b300408ca59ef2b8f0e92e01?from=pc

https://p26.toutiaoimg.com/origin/pgc-image/93d92719dea7445380da60d6310ce29c?from=pc

https://p26.toutiaoimg.com/origin/pgc-image/84da5228f1284d899305381ffa49e907?from=pc

所以这个 ringbuffer 就是 disruptor 中的 sequencer 对象,那么在进行获取 next 的时候,这里是如何获取下一个的呢?是否会对这个生产获取下一个序列进行相应的等待策略,避免产生相应的干扰!!!

这个各位看官还需多看看里面的代码以及封装(特别是封装,真是九转十八弯),多熟悉,我这绕着绕着很容易就绕晕了,刚开始也是云里雾里。

EventProcessor 顾名思义,就是事件处理器(handle 和 process 都可以翻译为 “处理”,但是 process 侧重于机器的处理,而 handle 侧重于有人工的处理,所以使用 handle 表示用户逻辑的处理,使用 process 表示机器的处理),这个接口有两个实现类,分别是 WorkProcessor 和 BatchEventProcessor,它们对应的逻辑处理消费者分别是 EventHandler 和 WorkHandler。下面是 EventProcessor 的 UML 类图及 EventHandler 和 EventProcessor 的接口定义。

https://p26.toutiaoimg.com/origin/pgc-image/ada5e7465c9d4c4bbcbb9d925f4411d7?from=pc

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
/**
* Callback interface to be implemented for processing events as they become available in the {@link RingBuffer}
*
* @param <T> event implementation storing the data for sharing during exchange or parallel coordination of an event.
* @see BatchEventProcessor#setExceptionHandler(ExceptionHandler) if you want to handle exceptions propagated out of the handler.
* 处理事件的回调接口
*/
public interface EventHandler<T>
{
 /**
 * Called when a publisher has published an event to the {@link RingBuffer}
 *
 * @param event published to the {@link RingBuffer}
 * @param sequence of the event being processed
 * @param endOfBatch flag to indicate if this is the last event in a batch from the {@link RingBuffer}
 * @throws Exception if the EventHandler would like the exception handled further up the chain.
 */
 void onEvent(T event, long sequence, boolean endOfBatch) throws Exception;
}
/**
* EventProcessors waitFor events to become available for consumption from the {@link RingBuffer}
* <p>
* An EventProcessor will generally be associated with a Thread for execution.
* 事件执行器,等待RingBuffer有可用消费事件。一个事件处理器关联一个执行线程
*/
public interface EventProcessor extends Runnable
{
 /**
 * Get a reference to the {@link Sequence} being used by this {@link EventProcessor}.
 *
 * @return reference to the {@link Sequence} for this {@link EventProcessor}
 */
 Sequence getSequence();
 /**
 * Signal that this EventProcessor should stop when it has finished consuming at the next clean break.
 * It will call {@link SequenceBarrier#alert()} to notify the thread to check status.
 */
 void halt();
 boolean isRunning();
}

EventProcessor 接口继承了 Runnable 接口,主要有两种实现:单线程批量处理 BatchEventProcessor 和多线程处理 WorkProcessor

在使用 Disruptor 帮助类构建消费者时,使用 handleEventsWith 方法传入多个 EventHandler,内部使用多个 BatchEventProcessor 关联多个线程执行。这种情况类似 JMS 中的发布订阅模式,同一事件会被多个消费者并行消费。适用于同一事件触发多种操作。

而使用 Disruptor 的 handleEventsWithWorkerPool 传入多个 WorkHandler 时,内部使用多个 WorkProcessor 关联多个线程执行。这种情况类似 JMS 的点对点模式,同一事件会被一组消费者其中之一消费。适用于提升消费者并行处理能力。

我们先回顾下 Disruptor 消费者的两个特点:消费者依赖图(即下文所谓的 “消费链”)和事件多播。

假设现在有 A,B,C,D 四个消费者,它们都能组成什么样的形式呢?从众多的排列组合中,我挑了 4 组比较有代表性的消费链形式。

https://p26.toutiaoimg.com/origin/pgc-image/7af770eefbdb49b18ddc64aabd486190?from=pc

image.png

  • 第 1 组中,消费者 A 消费按成后,B、C、D 可同时消费;
  • 第 2 组中,消费者 A、B、C、D 顺序消费;
  • 第 3 组中,消费者 A、B 顺序消费后,C、D 同时消费;
  • 第 4 组中,消费者 A 在消费完成后,B 和 C 可以同时消费,但是必须在都消费完成后,D 才能消费。

标号为 1、3、4 的消费链都使用了事件多播,可见事件多播属于消费链的一种组合形式。注意,在上面 4 种组合中,每个组合的每一水平行,都属于一个消费者组。

这些还只是较为简单的消费链组成,实际中消费链可能会更复杂。

那么在 Disruptor 内部是怎么实现消费链的呢?

我们可以先思考下。如果想把独立的消费者组成消费链,那么后方的消费者(组)必然要知道在它前方的消费者(组)的处理情况,否则就做不到顺序消费。同时,消费者也要了解生产者的位置,来判断是否有可用事件。之前我们分析生产者代码的时候,已经讲过,生产者为了不覆盖没有消费完全的事件,必须知道最慢消费者的处理情况

做到了这些才会有能力去控制消费者组成消费链。下面让我们具体看 Disruptor 中的实现。

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
package liuqiang.complex.multi;
import com.lmax.disruptor.EventFactory;
import com.lmax.disruptor.RingBuffer;
import com.lmax.disruptor.YieldingWaitStrategy;
import com.lmax.disruptor.dsl.Disruptor;
import com.lmax.disruptor.dsl.EventHandlerGroup;
import com.lmax.disruptor.dsl.ProducerType;
import liuqiang.complex.common.*;
import java.util.concurrent.Executors;
public class Main3 {
 //单生产者,多消费者模式。多消费者对于消息不重复消费。例如:1线程消费了消息0,则2线程只能从0后面的消息消费,不能对消息0进行消费。
 public static void main(String[] args) throws Exception {
 EventFactory<Order> factory = new OrderFactory();
 int ringBufferSize = 1024 * 1024;
 Disruptor<Order> disruptor =
 new Disruptor<Order>(factory, ringBufferSize, Executors.defaultThreadFactory(), ProducerType.SINGLE, new YieldingWaitStrategy());
 /*
 * 该方法传入的消费者需要实现WorkHandler接口,方法的内部实现是:先创建WorkPool,然后封装WorkPool为EventHandlerPool返回。
 * 消费者1、2对于消息的消费有时有竞争,保证同一消息只能有一个消费者消费
 */
 disruptor.handleEventsWithWorkerPool(new OrderHandler1("1"), new OrderHandler1("2"));
 disruptor.start();
 RingBuffer<Order> ringBuffer = disruptor.getRingBuffer();
 Producer producer = new Producer(ringBuffer);
 //单生产者,生产3条数据
 for (int l = 0; l < 3; l++) {
 producer.onData(l + "");
 }
 //为了保证消费者线程已经启动,留足足够的时间。具体原因详见另一篇博客:disruptor的shutdown失效问题
 Thread.sleep(1000);
 disruptor.shutdown();
 }
}

调用handleEventsWithWorkerPool 形成 WorkerPool,并进一步封装成 EventHandlerGroup。对于同一条消息,两消费者不重复消费。

可能输出结果如下:

OrderHandler1 1,消费信息:0

OrderHandler1 2,消费信息:1

OrderHandler1 1,消费信息:2

我们重点看一下 SequenceBarrier,可直译为 “序列屏障”。SequenceBarrier 的主要作用是协调获取消费者可处理到的最大序号,内部持有着生产者和其依赖的消费者序列。它的接口定义如下。

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
public interface SequenceBarrier
{
 /**
 * Wait for the given sequence to be available for consumption.<br>
 * 等待指定序列可用
 * @param sequence to wait for
 * @return the sequence up to which is available
 * @throws AlertException if a status change has occurred for the Disruptor
 * @throws InterruptedException if the thread needs awaking on a condition variable.
 * @throws TimeoutException
 *
 */
 long waitFor(long sequence) throws AlertException, InterruptedException, TimeoutException;
 /**
 * Get the current cursor value that can be read.<br>
 * 获取当前可读游标值
 *
 * @return value of the cursor for entries that have been published.
 *
 */
 long getCursor();
 /**
 * The current alert status for the barrier.<br>
 * 当前的alert状态
 *
 * @return true if in alert otherwise false.
 */
 boolean isAlerted();
 /**
 * Alert the {@link EventProcessor}s of a status change and stay in this status until cleared.<br>
 *
 * 通知消费者状态变化。当调用EventProcessor#halt()将调用此方法。
 */
 void alert();
 /**
 * Clear the current alert status.<br>
 * 清楚alert状态
 */
 void clearAlert();
 /**
 * Check if an alert has been raised and throw an {@link AlertException} if it has.
 * 检查是否发生alert,发生将抛出异常
 * @throws AlertException if alert has been raised.
 */
 void checkAlert() throws AlertException;
}
SequenceBarrier实例引用被EventProcessor持有,用于等待并获取可用的消费事件,主要体现在waitFor这个方法。

要实现这个功能,需要 3 点条件:

  1. 知道生产者的位置。
  2. 因为 Disruptor 支持消费者链,在不同的消费者组之间,要保证后边的消 费者组只有在前消费者组中的消费者都处理完毕后,才能进行处理。
  3. 暂时没有事件可消费,在等待可用消费时,还需要使用某种等待策略进行等待。

看下 SequenceBarrier 实现类 ProcessingSequenceBarrier 的代码是如何实现 waitFor 方法。

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
final class ProcessingSequenceBarrier implements SequenceBarrier
{
 private final WaitStrategy waitStrategy; // 等待可用消费时,指定的等待策略
 private final Sequence dependentSequence; // 依赖的上组消费者的序号,如果当前为第一组则为cursorSequence(即生产者发布游标序列),否则使用FixedSequenceGroup封装上组消费者序列
 private volatile boolean alerted = false; // 当触发halt时,将标记alerted为true
 private final Sequence cursorSequence; // AbstractSequencer中的cursor引用,记录当前发布者发布的最新位置
 private final Sequencer sequencer; // MultiProducerSequencer 或 SingleProducerSequencer
 public 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) // 依赖的上一组序列长度,第一次是0
 {
 dependentSequence = cursorSequence;
 }
 else // 将上一组序列数组复制成新数组保存,引用不变
 {
 dependentSequence = new FixedSequenceGroup(dependentSequences);
 }
 }
 @Override
 public long waitFor(final long sequence)
 throws AlertException, InterruptedException, TimeoutException
 {
 // 检查是否停止服务
 checkAlert();
 // 获取最大可用序号 sequence为给定序号,一般为当前序号+1,cursorSequence记录生产者最新位置,
 long availableSequence = waitStrategy.waitFor(sequence, cursorSequence, dependentSequence, this);
 if (availableSequence < sequence)
 {
 return availableSequence;
 }
 // 返回已发布最高的序列值,将对每个序号进行校验
 return sequencer.getHighestPublishedSequence(sequence, availableSequence);
 }
 // ... 
}