zookeeper 本地多线程模拟分布式事务控制

基本思路是：通过每个连接创建 临时节点（避免宕机后节点不释放）后规定节点最小的拥有获取锁的权利，那么其他的就拿不到了，但是每个节点都对前一个节点建立delete的watch机制。那么每次前一个节点释放锁（delete）触发watch 后一个节点就能获取锁 本地启动zkServer：

package xyz.luofu.www;

import org.apache.zookeeper.*;
import org.junit.Test;

import java.io.IOException;
import java.util.Collections;
import java.util.List;
import java.util.concurrent.CountDownLatch;

public class TestZookeePerLock {

    private String seqNodeNamePar = "/Order"; //这个节点先建好 create /Order
    private ThreadLocal<String> currentNodeNameThr = new ThreadLocal<>();
    private ThreadLocal<ZooKeeper> zk = new ThreadLocal<>();
    public boolean tryLock() {
        try {
            String seqNodeName = "/sq";
            zk.set(new ZooKeeper("localhost:2181", 3000, new Watcher() {
                @Override
                public void process(WatchedEvent event) {

                }
            }));
            //Ids.OPEN_ACL_UNSAFE，这就表明之后对这个节点的任何操作都不受权限控制
            currentNodeNameThr.set(zk.get().create(seqNodeNamePar + seqNodeName, new byte[0], ZooDefs.Ids.OPEN_ACL_UNSAFE, CreateMode.EPHEMERAL_SEQUENTIAL));
            List<String> children = zk.get().getChildren(seqNodeNamePar, false);
            Collections.sort(children);

            String currentNodeNameStr = currentNodeNameThr.get().substring(currentNodeNameThr.get().lastIndexOf("/") + 1);
            int index = children.indexOf(currentNodeNameStr);
            String preNodeName = "";
            if(index != 0){
                preNodeName = children.get(index - 1);
            }
            String first = children.get(0);
            if (currentNodeNameStr.equals(first)) {
                System.out.println(Thread.currentThread().getName()+"获得锁成功");
                return true;
            } else {
                final CountDownLatch countDownLatch = new CountDownLatch(1);
                zk.get().exists(seqNodeNamePar + "/" + preNodeName, new Watcher() {
                    @Override
                    public void process(WatchedEvent event) {
                        if (Event.EventType.NodeDeleted == event.getType()) {
                            countDownLatch.countDown();
                            System.out.println(Thread.currentThread().getName()+"重新获得锁成功");
                        }
                    }
                });
                System.out.println(Thread.currentThread().getName()+"等待锁");
                countDownLatch.await();
                return true;
            }
        } catch (Exception e) {
            e.printStackTrace();
            return false;
        }
    }
    public void unlock(){
        try {
            Thread.sleep(100);//避免还没建立监听这个节点就已delete。那么后面建立的delete watch 就触发不了了。
            zk.get().delete(currentNodeNameThr.get(),-1);
        } catch (InterruptedException e) {
            e.printStackTrace();
        } catch (KeeperException e) {
            e.printStackTrace();
        }
        System.out.println(Thread.currentThread().getName()+"释放锁成功");
    }

    class TestZoo implements Runnable{

        @Override
        public void run() {
            new Order().create();
            tryLock();
            boolean flag = new Store().descStore();
            unlock();
            if(flag){
                new Pay().doPay();
                System.out.println(Thread.currentThread().getName()+"支付成功");
            }
        }
    }

    @Test
    public void testZo() throws InterruptedException {
        new Thread(new TestZoo(),"线程一").start();
        new Thread(new TestZoo(),"线程二").start();
        int count = Thread.activeCount();
        while(count > 2){
            Thread.yield();
        }
        System.out.println("主线程结束");
    }
}

public class Order {
    public void create(){
        System.out.println(Thread.currentThread().getName()+"创建订单");
    }
}

public class Pay {
    public void doPay(){
    }
}

public class Store {
    public int count =1;
    public boolean descStore(){
        if(count > 0){
            count--;
            System.out.println(Thread.currentThread().getName()+"减库存成功 count:"+count);
            return true;
        }
        System.out.println(Thread.currentThread().getName()+"减库存失败");
        return false;
    }
}

运行截图：

对于分布式配置中心，也是结合watch机制，加cache。每次新增zk.create并且加入cache其他操作类似。初始化时加载配置节点/config下的所有节点进入cache，并且设置对数据操作的watch相应的操作缓存。然后其他服务只需从cache中读取数据。

zookeeper 本地多线程模拟分布式事务控制

原文：https://www.cnblogs.com/leifonlyone/p/12851534.html