ZooKeeper实现生产-消费者队列
- 2019 年 10 月 8 日
- 筆記
目录
对前续代码的重构
- 队列的生产者
- 队列的消费者
- 测试日志
- 源代码
生产-消费者队列，用于多节点的分布式数据结构，生产和消费数据。生产者创建一个数据对象，并放到队列中；消费者从队列中取出一个数据对象并进行处理。在ZooKeeper中，队列可以使用一个容器节点下创建多个子节点来实现；创建子节点时，CreateMode使用 PERSISTENT_SEQUENTIAL，ZooKeeper会自动在节点名称后面添加唯一序列号。EPHEMERAL_SEQUENTIAL也有同样的特点，区别在于会话结束后是否会自动删除。
敲小黑板：*_SEQUENTIAL是ZooKeeper的一个很重要的特性，分布式锁、选举制度都依靠这个特性实现的。
1 对前续代码的重构
之前的文章，我们已经用实现了Watcher和Barrier，创建ZooKeeper连接的代码已经复制了一遍。后续还需要类似的工作，因此先对原有代码做一下重构，让代码味道干净一点。
以下是 process(WatchedEvent)的代码
final public void process(WatchedEvent event) { if (Event.EventType.None.equals(event.getType())) { // 连接状态发生变化 if (Event.KeeperState.SyncConnected.equals(event.getState())) { // 连接建立成功 connectedSemaphore.countDown(); } } else if (Event.EventType.NodeCreated.equals(event.getType())) { processNodeCreated(event); } else if (Event.EventType.NodeDeleted.equals(event.getType())) { processNodeDeleted(event); } else if (Event.EventType.NodeDataChanged.equals(event.getType())) { processNodeDataChanged(event); } else if (Event.EventType.NodeChildrenChanged.equals(event.getType())) { processNodeChildrenChanged(event); } }
以ZooKeeperBarrier为例，看看重构之后的构造函数和监听Event的代码
ZooKeeperBarrier(String address, String tableSerial, int tableCapacity, String customerName) throws IOException { super(address); this.tableSerial = createRootNode(tableSerial); this.tableCapacity = tableCapacity; this.customerName = customerName; } protected void processNodeChildrenChanged(WatchedEvent event) { log.info("{} 接收到了通知 : {}", customerName, event.getType()); // 子节点有变化 synchronized (mutex) { mutex.notify(); } }
2 队列的生产者
生产者的关键代码
String elementName = queueName + "/element"; ArrayList<ACL> ids = ZooDefs.Ids.OPEN_ACL_UNSAFE; CreateMode createMode = CreateMode.PERSISTENT_SEQUENTIAL; getZooKeeper().create(elementName, value, ids, createMode);
注意，重点是PERSISTENT_SEQUENTIAL，PERSISTENT是表示永久存储直到有命令删除，SEQUENTIAL表示自动在后面添加自增的唯一序列号。这样，尽管elementName都一样，但实际生成的zNode名字在 “element”后面会添加格式为%010d的10个数字，如0000000001。如一个完整的zNode名可能为/queue/element0000000021。
3 队列的消费者
消费者尝试从子节点列表获取zNode名最小的一个子节点，如果队列为空则等待NodeChildrenChanged事件。关键代码
/** 队列的同步信号 */ private static Integer queueMutex = Integer.valueOf(1); @Override protected void processNodeChildrenChanged(WatchedEvent event) { synchronized (queueMutex) { queueMutex.notify(); } } /** * 从队列中删除第一个对象 * * @return * @throws KeeperException * @throws InterruptedException */ int consume() throws KeeperException, InterruptedException { while (true) { synchronized (queueMutex) { List<String> list = getZooKeeper().getChildren(queueName, true); if (list.size() == 0) { queueMutex.wait(); } else { // 获取第一个子节点的名称 String firstNodeName = getFirstElementName(list); // 删除节点，并返回节点的值 return deleteNodeAndReturnValue(firstNodeName); } } } }
4 测试日志
把测试结果放源码前面，免得大家被无聊的代码晃晕。
测试代码创建了两个线程，一个线程是生产者，按随机间隔往队列中添加对象；一个线程是消费者，随机间隔尝试从队列中取出第一个，如果当时队列为空，会等到直到新的数据。
两个进程都加上随机间隔，是为了模拟生产可能比消费更快的情况。以下是测试日志，为了更突出，生产和消费的日志我增加了不同的文字样式。
49:47.866 [INFO] ZooKeeperQueueTest.testQueue(29) 开始ZooKeeper队列测试，本次将测试 10 个数据 49:48.076 [DEBUG] ZooKeeperQueue.log(201) + Profiler [tech.codestory.zookeeper.queue.ZooKeeperQueue 连接到ZooKeeper] |-- elapsed time [开始链接] 119.863 milliseconds. |-- elapsed time [等待连接成功的Event] 40.039 milliseconds. |-- Total [tech.codestory.zookeeper.queue.ZooKeeperQueue 连接到ZooKeeper] 159.911 milliseconds. 49:48.082 [DEBUG] ZooKeeperQueue.log(201) + Profiler [tech.codestory.zookeeper.queue.ZooKeeperQueue 连接到ZooKeeper] |-- elapsed time [开始链接] 103.795 milliseconds. |-- elapsed time [等待连接成功的Event] 65.899 milliseconds. |-- Total [tech.codestory.zookeeper.queue.ZooKeeperQueue 连接到ZooKeeper] 170.263 milliseconds. 49:48.102 [INFO] ZooKeeperQueueTest.run(51) 生产对象 : 1 , 然后等待 1700 毫秒 49:48.134 [INFO] ZooKeeperQueueTest.run(80) 消费对象: 1 , 然后等待 4000 毫秒 49:49.814 [INFO] ZooKeeperQueueTest.run(51) 生产对象 : 2 , 然后等待 900 毫秒 49:50.717 [INFO] ZooKeeperQueueTest.run(51) 生产对象 : 3 , 然后等待 1300 毫秒 49:52.020 [INFO] ZooKeeperQueueTest.run(51) 生产对象 : 4 , 然后等待 3700 毫秒 49:52.139 [INFO] ZooKeeperQueueTest.run(80) 消费对象: 2 , 然后等待 2800 毫秒 49:54.947 [INFO] ZooKeeperQueueTest.run(80) 消费对象: 3 , 然后等待 4500 毫秒 49:55.724 [INFO] ZooKeeperQueueTest.run(51) 生产对象 : 5 , 然后等待 3500 毫秒 49:59.228 [INFO] ZooKeeperQueueTest.run(51) 生产对象 : 6 , 然后等待 4200 毫秒 49:59.454 [INFO] ZooKeeperQueueTest.run(80) 消费对象: 4 , 然后等待 2400 毫秒 50:01.870 [INFO] ZooKeeperQueueTest.run(80) 消费对象: 5 , 然后等待 4900 毫秒 50:03.435 [INFO] ZooKeeperQueueTest.run(51) 生产对象 : 7 , 然后等待 4500 毫秒 50:06.776 [INFO] ZooKeeperQueueTest.run(80) 消费对象: 6 , 然后等待 3600 毫秒 50:07.938 [INFO] ZooKeeperQueueTest.run(51) 生产对象 : 8 , 然后等待 1900 毫秒 50:09.846 [INFO] ZooKeeperQueueTest.run(51) 生产对象 : 9 , 然后等待 3200 毫秒 50:10.388 [INFO] ZooKeeperQueueTest.run(80) 消费对象: 7 , 然后等待 2900 毫秒 50:13.051 [INFO] ZooKeeperQueueTest.run(51) 生产对象 : 10 , 然后等待 4900 毫秒 50:13.294 [INFO] ZooKeeperQueueTest.run(80) 消费对象: 8 , 然后等待 300 毫秒 50:13.600 [INFO] ZooKeeperQueueTest.run(80) 消费对象: 9 , 然后等待 4800 毫秒 50:18.407 [INFO] ZooKeeperQueueTest.run(80) 消费对象: 10 , 然后等待 2400 毫秒
5 完整源码
5.1 ZooKeeperBase.java
package tech.codestory.zookeeper; import java.io.IOException; import java.util.concurrent.CountDownLatch; import org.apache.zookeeper.*; import org.apache.zookeeper.data.Stat; import org.slf4j.Logger; import org.slf4j.LoggerFactory; import org.slf4j.profiler.Profiler; /** * 为 ZooKeeper测试代码创建一个基类，封装建立连接的过程 * * @author code story * @date 2019/8/16 */ public class ZooKeeperBase implements Watcher { /** 日志，不使用 @Slf4j ，是要使用子类的log */ Logger log = null; /** 等待连接建立成功的信号 */ private CountDownLatch connectedSemaphore = new CountDownLatch(1); /** ZooKeeper 客户端 */ private ZooKeeper zooKeeper = null; /** 避免重复根节点 */ static Integer rootNodeInitial = Integer.valueOf(1); /** 构造函数 */ public ZooKeeperBase(String address) throws IOException { log = LoggerFactory.getLogger(getClass()); Profiler profiler = new Profiler(this.getClass().getName() + " 连接到ZooKeeper"); profiler.start("开始链接"); zooKeeper = new ZooKeeper(address, 3000, this); try { profiler.start("等待连接成功的Event"); connectedSemaphore.await(); } catch (InterruptedException e) { log.error("InterruptedException", e); } profiler.stop(); profiler.setLogger(log); profiler.log(); } /** * 创建测试需要的根节点 * * @param rootNodeName * @return */ public String createRootNode(String rootNodeName) { synchronized (rootNodeInitial) { // 创建 tableSerial 的zNode try { Stat existsStat = getZooKeeper().exists(rootNodeName, false); if (existsStat == null) { rootNodeName = getZooKeeper().create(rootNodeName, new byte[0], ZooDefs.Ids.OPEN_ACL_UNSAFE, CreateMode.PERSISTENT); } } catch (KeeperException e) { log.error("KeeperException", e); } catch (InterruptedException e) { log.error("InterruptedException", e); } } return rootNodeName; } /** 读取ZooKeeper对象，供子类调用 */ protected ZooKeeper getZooKeeper() { return zooKeeper; } @Override final public void process(WatchedEvent event) { if (Event.EventType.None.equals(event.getType())) { // 连接状态发生变化 if (Event.KeeperState.SyncConnected.equals(event.getState())) { // 连接建立成功 connectedSemaphore.countDown(); } } else if (Event.EventType.NodeCreated.equals(event.getType())) { processNodeCreated(event); } else if (Event.EventType.NodeDeleted.equals(event.getType())) { processNodeDeleted(event); } else if (Event.EventType.NodeDataChanged.equals(event.getType())) { processNodeDataChanged(event); } else if (Event.EventType.NodeChildrenChanged.equals(event.getType())) { processNodeChildrenChanged(event); } } /** * 处理事件: NodeCreated * * @param event */ protected void processNodeCreated(WatchedEvent event) {} /** * 处理事件: NodeDeleted * * @param event */ protected void processNodeDeleted(WatchedEvent event) {} /** * 处理事件: NodeDataChanged * * @param event */ protected void processNodeDataChanged(WatchedEvent event) {} /** * 处理事件: NodeChildrenChanged * * @param event */ protected void processNodeChildrenChanged(WatchedEvent event) {} }
5.2 ZooKeeperQueue.java
package tech.codestory.zookeeper.queue; import java.io.IOException; import java.nio.ByteBuffer; import java.util.ArrayList; import java.util.List; import org.apache.zookeeper.CreateMode; import org.apache.zookeeper.KeeperException; import org.apache.zookeeper.WatchedEvent; import org.apache.zookeeper.ZooDefs; import org.apache.zookeeper.data.ACL; import org.apache.zookeeper.data.Stat; import lombok.extern.slf4j.Slf4j; import tech.codestory.zookeeper.ZooKeeperBase; /** * ZooKeeper实现Queue * * @author code story * @date 2019/8/16 */ @Slf4j public class ZooKeeperQueue extends ZooKeeperBase { /** 队列名称 */ private String queueName; /** 队列的同步信号 */ private static Integer queueMutex = Integer.valueOf(1); /** * 构造函数 * * @param address * @param queueName * @throws IOException */ public ZooKeeperQueue(String address, String queueName) throws IOException { super(address); this.queueName = createRootNode(queueName); } @Override protected void processNodeChildrenChanged(WatchedEvent event) { synchronized (queueMutex) { queueMutex.notify(); } } /** * 将对象添加到队列中 * * @param i * @return */ boolean produce(int i) throws KeeperException, InterruptedException { ByteBuffer b = ByteBuffer.allocate(4); byte[] value; // Add child with value i b.putInt(i); value = b.array(); String elementName = queueName + "/element"; ArrayList<ACL> ids = ZooDefs.Ids.OPEN_ACL_UNSAFE; CreateMode createMode = CreateMode.PERSISTENT_SEQUENTIAL; getZooKeeper().create(elementName, value, ids, createMode); return true; } /** * 从队列中删除第一个对象 * * @return * @throws KeeperException * @throws InterruptedException */ int consume() throws KeeperException, InterruptedException { while (true) { synchronized (queueMutex) { List<String> list = getZooKeeper().getChildren(queueName, true); if (list.size() == 0) { queueMutex.wait(); } else { // 获取第一个子节点的名称 String firstNodeName = getFirstElementName(list); // 删除节点，并返回节点的值 return deleteNodeAndReturnValue(firstNodeName); } } } } /** * 获取第一个子节点的名称 * * @param list * @return */ private String getFirstElementName(List<String> list) { Integer min = Integer.MAX_VALUE; String minNode = null; for (String s : list) { Integer tempValue = Integer.valueOf(s.substring(7)); if (tempValue < min) { min = tempValue; minNode = s; } } return minNode; } /** * 删除节点，并返回节点的值 * * @param minNode * @return * @throws KeeperException * @throws InterruptedException */ private int deleteNodeAndReturnValue(String minNode) throws KeeperException, InterruptedException { String fullNodeName = queueName + "/" + minNode; Stat stat = new Stat(); byte[] b = getZooKeeper().getData(fullNodeName, false, stat); getZooKeeper().delete(fullNodeName, stat.getVersion()); ByteBuffer buffer = ByteBuffer.wrap(b); return buffer.getInt(); } }
5.3 ZooKeeperQueueTest.java
package tech.codestory.zookeeper.queue; import java.io.IOException; import java.security.SecureRandom; import java.util.Random; import java.util.concurrent.CountDownLatch; import org.apache.zookeeper.KeeperException; import org.testng.annotations.Test; import lombok.extern.slf4j.Slf4j; /** * ZooKeeperQueue测试 * * @author code story * @date 2019/8/16 */ @Slf4j public class ZooKeeperQueueTest { final String address = "192.168.5.128:2181"; final String queueName = "/queue"; final Random random = new SecureRandom(); // 随机生成10-20之间的个数 final int count = 10 + random.nextInt(10); /** 等待生产者和消费者线程都结束 */ private CountDownLatch connectedSemaphore = new CountDownLatch(2); @Test public void testQueue() { log.info("开始ZooKeeper队列测试，本次将测试 {} 个数据", count); new QueueProducer().start(); new QueueConsumer().start(); try { connectedSemaphore.await(); } catch (InterruptedException e) { log.error("InterruptedException", e); } } /** * 队列的生产者 */ class QueueProducer extends Thread { @Override public void run() { try { ZooKeeperQueue queue = new ZooKeeperQueue(address, queueName); for (int i = 0; i < count; i++) { int elementValue = i + 1; long waitTime = random.nextInt(50) * 100; log.info("生产对象 : {} , 然后等待 {} 毫秒", elementValue, waitTime); queue.produce(elementValue); Thread.sleep(waitTime); } } catch (IOException e) { log.error("IOException", e); } catch (InterruptedException e) { log.error("InterruptedException", e); } catch (KeeperException e) { log.error("KeeperException", e); } connectedSemaphore.countDown(); } } /** * 队列的消费者 */ class QueueConsumer extends Thread { @Override public void run() { try { ZooKeeperQueue queue = new ZooKeeperQueue(address, queueName); for (int i = 0; i < count; i++) { try { int elementValue = queue.consume(); long waitTime = random.nextInt(50) * 100; log.info("消费对象: {} , 然后等待 {} 毫秒", elementValue, waitTime); Thread.sleep(waitTime); } catch (KeeperException e) { i--; log.error("KeeperException", e); } catch (InterruptedException e) { log.error("InterruptedException", e); } } connectedSemaphore.countDown(); } catch (IOException e) { log.error("IOException", e); } } } }
