《disruptor笔记》系列链接
- 快速入门
- Disruptor类分析
- 环形队列的基础操作(不用Disruptor类)
- 事件消费知识点小结
- 事件消费实战
- 常见场景
- 等待策略
- 知识点补充(终篇)
本篇概览
- 本文是《disruptor笔记》系列的第三篇,主要任务是编码实现消息生产和消费,与《disruptor笔记之一:快速入门》不同的是,本次开发不使用Disruptor类,和Ring Buffer(环形队列)相关的操作都是自己写代码实现;
- 这种脱离Disruptor类操作Ring Buffer的做法,不适合用在生产环境,但在学习Disruptor的过程中,这是种高效的学习手段,经过本篇实战后,在今后使用Disruptor时,您在开发、调试、优化等各种场景下都能更加得心应手;
- 简单的消息生产消费已不能满足咱们的学习热情,今天的实战要挑战以下三个场景:
- 100个事件,单个消费者消费;
- 100个事件,三个消费者,每个都独自消费这个100个事件;
- 100个事件,三个消费者共同消费这个100个事件;
前文回顾
为了完成本篇的实战,前文《disruptor笔记之二:Disruptor类分析》已做了充分的研究分析,建议观看,这里简单回顾以下Disruptor类的几个核心功能,这也是咱们编码时要实现的:
- 创建环形队列(RingBuffer对象)
- 创建SequenceBarrier对象,用于接收ringBuffer中的可消费事件
- 创建BatchEventProcessor,负责消费事件
- 绑定BatchEventProcessor对象的异常处理类
- 调用ringBuffer.addGatingSequences,将消费者的Sequence传给ringBuffer
- 启动独立线程,用来执行消费事件的业务逻辑
- 理论分析已经完成,接下来开始编码;
源码下载
- 本篇实战中的完整源码可在GitHub下载到,地址和链接信息如下表所示(https://github.com/zq2599/blog_demos):
名称 | 链接 | 备注 |
---|---|---|
项目主页 | https://github.com/zq2599/blog_demos | 该项目在GitHub上的主页 |
git仓库地址(https) | https://github.com/zq2599/blog_demos.git | 该项目源码的仓库地址,https协议 |
git仓库地址(ssh) | git@github.com:zq2599/blog_demos.git | 该项目源码的仓库地址,ssh协议 |
- 这个git项目中有多个文件夹,本次实战的源码在disruptor-tutorials文件夹下,如下图红框所示:
- disruptor-tutorials是个父工程,里面有多个module,本篇实战的module是low-level-operate,如下图红框所示:
开发
- 进入编码阶段,今天的任务是挑战以下三个场景:
- 100个事件,单个消费者消费;
- 100个事件,三个消费者,每个都独自消费这个100个事件;
- 100个事件,三个消费者共同消费这个100个事件;
- 咱们先把工程建好,然后编写公共代码,例如事件定义、事件工厂等,最后才是每个场景的开发;
- 在父工程disruptor-tutorials新增名为low-level-operate的module,其build.gradle如下:
plugins {
id 'org.springframework.boot'
}
dependencies {
implementation 'org.projectlombok:lombok'
implementation 'org.springframework.boot:spring-boot-starter'
implementation 'org.springframework.boot:spring-boot-starter-web'
implementation 'com.lmax:disruptor'
testImplementation('org.springframework.boot:spring-boot-starter-test')
}
- 然后是springboot启动类:
package com.bolingcavalry;
import org.springframework.boot.SpringApplication;
import org.springframework.boot.autoconfigure.SpringBootApplication;
@SpringBootApplication
public class LowLevelOperateApplication {
public static void main(String[] args) {
SpringApplication.run(LowLevelOperateApplication.class, args);
}
}
- 事件类,这是事件的定义:
package com.bolingcavalry.service;
import lombok.Data;
import lombok.NoArgsConstructor;
import lombok.ToString;
@Data
@ToString
@NoArgsConstructor
public class StringEvent {
private String value;
}
- 事件工厂,定义如何在内存中创建事件对象:
package com.bolingcavalry.service;
import com.lmax.disruptor.EventFactory;
public class StringEventFactory implements EventFactory<StringEvent> {
@Override
public StringEvent newInstance() {
return new StringEvent();
}
}
- 事件生产类,定义如何将业务逻辑的事件转为disruptor事件发布到环形队列,用于消费:
package com.bolingcavalry.service;
import com.lmax.disruptor.RingBuffer;
public class StringEventProducer {
// 存储数据的环形队列
private final RingBuffer<StringEvent> ringBuffer;
public StringEventProducer(RingBuffer<StringEvent> ringBuffer) {
this.ringBuffer = ringBuffer;
}
public void onData(String content) {
// ringBuffer是个队列,其next方法返回的是下最后一条记录之后的位置,这是个可用位置
long sequence = ringBuffer.next();
try {
// sequence位置取出的事件是空事件
StringEvent stringEvent = ringBuffer.get(sequence);
// 空事件添加业务信息
stringEvent.setValue(content);
} finally {
// 发布
ringBuffer.publish(sequence);
}
}
}
- 事件处理类,收到事件后具体的业务处理逻辑:
package com.bolingcavalry.service;
import com.lmax.disruptor.EventHandler;
import lombok.Setter;
import lombok.extern.slf4j.Slf4j;
import java.util.function.Consumer;
@Slf4j
public class StringEventHandler implements EventHandler<StringEvent> {
public StringEventHandler(Consumer<?> consumer) {
this.consumer = consumer;
}
// 外部可以传入Consumer实现类,每处理一条消息的时候,consumer的accept方法就会被执行一次
private Consumer<?> consumer;
@Override
public void onEvent(StringEvent event, long sequence, boolean endOfBatch) throws Exception {
log.info("sequence [{}], endOfBatch [{}], event : {}", sequence, endOfBatch, event);
// 这里延时100ms,模拟消费事件的逻辑的耗时
Thread.sleep(100);
// 如果外部传入了consumer,就要执行一次accept方法
if (null!=consumer) {
consumer.accept(null);
}
}
}
- 定义一个接口,外部通过调用接口的方法来生产消息,再放几个常量在里面后面会用到:
package com.bolingcavalry.service;
public interface LowLevelOperateService {
/**
* 消费者数量
*/
int CONSUMER_NUM = 3;
/**
* 环形缓冲区大小
*/
int BUFFER_SIZE = 16;
/**
* 发布一个事件
* @param value
* @return
*/
void publish(String value);
/**
* 返回已经处理的任务总数
* @return
*/
long eventCount();
}
- 以上就是公共代码了,接下来逐个实现之前规划的三个场景;
100个事件,单个消费者消费
- 这是最简单的功能了,实现发布消息和单个消费者消费的功能,代码如下,有几处要注意的地方稍后提到:
package com.bolingcavalry.service.impl;
import com.bolingcavalry.service.*;
import com.lmax.disruptor.BatchEventProcessor;
import com.lmax.disruptor.RingBuffer;
import com.lmax.disruptor.SequenceBarrier;
import lombok.extern.slf4j.Slf4j;
import org.springframework.stereotype.Service;
import javax.annotation.PostConstruct;
import java.util.concurrent.ExecutorService;
import java.util.concurrent.Executors;
import java.util.concurrent.atomic.AtomicLong;
import java.util.function.Consumer;
@Service("oneConsumer")
@Slf4j
public class OneConsumerServiceImpl implements LowLevelOperateService {
private RingBuffer<StringEvent> ringBuffer;
private StringEventProducer producer;
/**
* 统计消息总数
*/
private final AtomicLong eventCount = new AtomicLong();
private ExecutorService executors;
@PostConstruct
private void init() {
// 准备一个匿名类,传给disruptor的事件处理类,
// 这样每次处理事件时,都会将已经处理事件的总数打印出来
Consumer<?> eventCountPrinter = new Consumer<Object>() {
@Override
public void accept(Object o) {
long count = eventCount.incrementAndGet();
log.info("receive [{}] event", count);
}
};
// 创建环形队列实例
ringBuffer = RingBuffer.createSingleProducer(new StringEventFactory(), BUFFER_SIZE);
// 准备线程池
executors = Executors.newFixedThreadPool(1);
//创建SequenceBarrier
SequenceBarrier sequenceBarrier = ringBuffer.newBarrier();
// 创建事件处理的工作类,里面执行StringEventHandler处理事件
BatchEventProcessor<StringEvent> batchEventProcessor = new BatchEventProcessor<>(
ringBuffer,
sequenceBarrier,
new StringEventHandler(eventCountPrinter));
// 将消费者的sequence传给环形队列
ringBuffer.addGatingSequences(batchEventProcessor.getSequence());
// 在一个独立线程中取事件并消费
executors.submit(batchEventProcessor);
// 生产者
producer = new StringEventProducer(ringBuffer);
}
@Override
public void publish(String value) {
producer.onData(value);
}
@Override
public long eventCount() {
return eventCount.get();
}
}
- 上述代码有以下几处需要注意:
- 自己创建环形队列RingBuffer实例
- 自己准备线程池,里面的线程用来获取和消费消息
- 自己动手创建BatchEventProcessor实例,并把事件处理类传入
- 通过ringBuffer创建sequenceBarrier,传给BatchEventProcessor实例使用
- 将BatchEventProcessor的sequence传给ringBuffer,确保ringBuffer的生产和消费不会出现混乱
- 启动线程池,意味着BatchEventProcessor实例在一个独立线程中不断的从ringBuffer中获取事件并消费;
- 为了验证上述代码能否正常工作,我这里写了个单元测试类,如下所示,逻辑很简单,调用OneConsumerServiceImpl.publish方法一百次,产生一百个事件,再检查OneConsumerServiceImpl记录的消费事件总数是不是等于一百:
package com.bolingcavalry.service.impl;
import com.bolingcavalry.service.LowLevelOperateService;
import lombok.extern.slf4j.Slf4j;
import org.junit.Test;
import org.junit.runner.RunWith;
import org.springframework.beans.factory.annotation.Autowired;
import org.springframework.beans.factory.annotation.Qualifier;
import org.springframework.boot.test.context.SpringBootTest;
import org.springframework.test.context.junit4.SpringRunner;
import static org.junit.Assert.assertEquals;
@RunWith(SpringRunner.class)
@SpringBootTest
@Slf4j
public class LowLeverOperateServiceImplTest {
@Autowired
@Qualifier("oneConsumer")
LowLevelOperateService oneConsumer;
private static final int EVENT_COUNT = 100;
private void testLowLevelOperateService(LowLevelOperateService service, int eventCount, int expectEventCount) throws InterruptedException {
for(int i=0;i<eventCount;i ) {
log.info("publich {}", i);
service.publish(String.valueOf(i));
}
// 异步消费,因此需要延时等待
Thread.sleep(10000);
// 消费的事件总数应该等于发布的事件数
assertEquals(expectEventCount, service.eventCount());
}
@Test
public void testOneConsumer() throws InterruptedException {
log.info("start testOneConsumerService");
testLowLevelOperateService(oneConsumer, EVENT_COUNT, EVENT_COUNT);
}
- 注意,如果您是直接在IDEA上点击图标来执行单元测试,记得勾选下图红框中选项,否则可能出现编译失败:
- 执行上述单元测试类,结果如下图所示,消息的生产和消费都符合预期,并且消费逻辑是在独立线程中执行的:
- 继续挑战下一个场景;
100个事件,三个消费者,每个都独自消费这个100个事件
- 这个场景在kafka中也有,就是三个消费者的group不同,这样每一条消息,这两个消费者各自消费一次;
- 因此,100个事件,3个消费者每人都会独立消费这100个事件,一共消费300次;
- 代码如下,有几处要注意的地方稍后提到:
package com.bolingcavalry.service.impl;
import com.bolingcavalry.service.*;
import com.lmax.disruptor.BatchEventProcessor;
import com.lmax.disruptor.RingBuffer;
import com.lmax.disruptor.SequenceBarrier;
import lombok.extern.slf4j.Slf4j;
import org.springframework.stereotype.Service;
import javax.annotation.PostConstruct;
import java.util.concurrent.Executor;
import java.util.concurrent.ExecutorService;
import java.util.concurrent.Executors;
import java.util.concurrent.atomic.AtomicLong;
import java.util.function.Consumer;
@Service("multiConsumer")
@Slf4j
public class MultiConsumerServiceImpl implements LowLevelOperateService {
private RingBuffer<StringEvent> ringBuffer;
private StringEventProducer producer;
/**
* 统计消息总数
*/
private final AtomicLong eventCount = new AtomicLong();
/**
* 生产一个BatchEventProcessor实例,并且启动独立线程开始获取和消费消息
* @param executorService
*/
private void addProcessor(ExecutorService executorService) {
// 准备一个匿名类,传给disruptor的事件处理类,
// 这样每次处理事件时,都会将已经处理事件的总数打印出来
Consumer<?> eventCountPrinter = new Consumer<Object>() {
@Override
public void accept(Object o) {
long count = eventCount.incrementAndGet();
log.info("receive [{}] event", count);
}
};
BatchEventProcessor<StringEvent> batchEventProcessor = new BatchEventProcessor<>(
ringBuffer,
ringBuffer.newBarrier(),
new StringEventHandler(eventCountPrinter));
// 将当前消费者的sequence实例传给ringBuffer
ringBuffer.addGatingSequences(batchEventProcessor.getSequence());
// 启动独立线程获取和消费事件
executorService.submit(batchEventProcessor);
}
@PostConstruct
private void init() {
ringBuffer = RingBuffer.createSingleProducer(new StringEventFactory(), BUFFER_SIZE);
ExecutorService executorService = Executors.newFixedThreadPool(CONSUMER_NUM);
// 创建多个消费者,并在独立线程中获取和消费事件
for (int i=0;i<CONSUMER_NUM;i ) {
addProcessor(executorService);
}
// 生产者
producer = new StringEventProducer(ringBuffer);
}
@Override
public void publish(String value) {
producer.onData(value);
}
@Override
public long eventCount() {
return eventCount.get();
}
}
- 上述代码和前面的OneConsumerServiceImpl相比差别不大,主要是创建了多个BatchEventProcessor实例,然后分别在线程池中提交;
- 验证方法依旧是单元测试,在刚才的LowLeverOperateServiceImplTest.java中增加代码即可,注意testLowLevelOperateService的第三个参数是EVENT_COUNT * LowLevelOperateService.CONSUMER_NUM,表示预期的被消费消息数为300:
@Autowired
@Qualifier("multiConsumer")
LowLevelOperateService multiConsumer;
@Test
public void testMultiConsumer() throws InterruptedException {
log.info("start testMultiConsumer");
testLowLevelOperateService(multiConsumer, EVENT_COUNT, EVENT_COUNT * LowLevelOperateService.CONSUMER_NUM);
}
- 执行单元测试,如下图所示,一共消费了300个事件,并且三个消费者在不动线程:
100个事件,三个消费者共同消费这个100个事件
- 本篇的最后一个实战是发布100个事件,然后让三个消费者共同消费100个(例如A消费33个,B消费33个,C消费34个);
- 前面用到的BatchEventProcessor是用来独立消费的,不适合多个消费者共同消费,这种多个消费共同消费的场景需要借助WorkerPool来完成,这个名字还是很形象的:一个池子里面有很多个工作者,把任务放入这个池子,工作者们每人处理一部分,大家合力将任务完成;
- 传入WorkerPool的消费者需要实现WorkHandler接口,于是新增一个实现类:
package com.bolingcavalry.service;
import com.lmax.disruptor.WorkHandler;
import lombok.extern.slf4j.Slf4j;
import java.util.function.Consumer;
@Slf4j
public class StringWorkHandler implements WorkHandler<StringEvent> {
public StringWorkHandler(Consumer<?> consumer) {
this.consumer = consumer;
}
// 外部可以传入Consumer实现类,每处理一条消息的时候,consumer的accept方法就会被执行一次
private Consumer<?> consumer;
@Override
public void onEvent(StringEvent event) throws Exception {
log.info("work handler event : {}", event);
// 这里延时100ms,模拟消费事件的逻辑的耗时
Thread.sleep(100);
// 如果外部传入了consumer,就要执行一次accept方法
if (null!=consumer) {
consumer.accept(null);
}
}
}
- 新增服务类,实现共同消费的逻辑,有几处要注意的地方稍后会提到:
package com.bolingcavalry.service.impl;
import com.bolingcavalry.service.*;
import com.lmax.disruptor.*;
import lombok.extern.slf4j.Slf4j;
import org.springframework.stereotype.Service;
import javax.annotation.PostConstruct;
import java.util.concurrent.ExecutorService;
import java.util.concurrent.Executors;
import java.util.concurrent.atomic.AtomicLong;
import java.util.function.Consumer;
@Service("workerPoolConsumer")
@Slf4j
public class WorkerPoolConsumerServiceImpl implements LowLevelOperateService {
private RingBuffer<StringEvent> ringBuffer;
private StringEventProducer producer;
/**
* 统计消息总数
*/
private final AtomicLong eventCount = new AtomicLong();
@PostConstruct
private void init() {
ringBuffer = RingBuffer.createSingleProducer(new StringEventFactory(), BUFFER_SIZE);
ExecutorService executorService = Executors.newFixedThreadPool(CONSUMER_NUM);
StringWorkHandler[] handlers = new StringWorkHandler[CONSUMER_NUM];
// 创建多个StringWorkHandler实例,放入一个数组中
for (int i=0;i < CONSUMER_NUM;i ) {
handlers[i] = new StringWorkHandler(o -> {
long count = eventCount.incrementAndGet();
log.info("receive [{}] event", count);
});
}
// 创建WorkerPool实例,将StringWorkHandler实例的数组传进去,代表共同消费者的数量
WorkerPool<StringEvent> workerPool = new WorkerPool<>(ringBuffer, ringBuffer.newBarrier(), new IgnoreExceptionHandler(), handlers);
// 这一句很重要,去掉就会出现重复消费同一个事件的问题
ringBuffer.addGatingSequences(workerPool.getWorkerSequences());
workerPool.start(executorService);
// 生产者
producer = new StringEventProducer(ringBuffer);
}
@Override
public void publish(String value) {
producer.onData(value);
}
@Override
public long eventCount() {
return eventCount.get();
}
}
- 上述代码中,要注意的有以下两处:
- StringWorkHandler数组传入给WorkerPool后,每个StringWorkHandler实例都放入一个新的WorkProcessor实例,WorkProcessor实现了Runnable接口,在执行workerPool.start时,会将WorkProcessor提交到线程池中;
- 和前面的独立消费相比,共同消费最大的特点在于只调用了一次ringBuffer.addGatingSequences方法,也就是说三个消费者共用一个sequence实例;
- 验证方法依旧是单元测试,在刚才的LowLeverOperateServiceImplTest.java中增加代码即可,注意testWorkerPoolConsumer的第三个参数是EVENT_COUNT,表示预期的被消费消息数为100:
@Autowired
@Qualifier("workerPoolConsumer")
LowLevelOperateService workerPoolConsumer;
@Test
public void testWorkerPoolConsumer() throws InterruptedException {
log.info("start testWorkerPoolConsumer");
testLowLevelOperateService(workerPoolConsumer, EVENT_COUNT, EVENT_COUNT);
}
- 执行单元测试如下图所示,三个消费者一共消费100个事件,且三个消费者在不同线程:
- 至此,咱们在不用Disruptor类的前提下完成了三种常见场景的消息生产消费,相信您对Disruptor的底层实现也有了深刻认识,今后不论是使用还是优化Disruptor,一定可以更加得心应手;