Zookeeper读写性能测试

关于Zookeeper，之前有过几篇 文章已经介绍过了，因此本文不赘述。 本次小编对单机部署的Zookeeper的读、写进行了一次简单 性能测试。 性能测试脚本由 java完成，具体请看代码清单：

package com.kiven.test; import java.util.List; import java.util.concurrent.CountDownLatch; import org.apache.log4j.PropertyConfigurator; import org.apache.zookeeper.CreateMode; import org.apache.zookeeper.KeeperException; import org.apache.zookeeper.WatchedEvent; import org.apache.zookeeper.Watcher; import org.apache.zookeeper.Watcher.Event.KeeperState; import org.apache.zookeeper.ZooDefs.Ids; import org.apache.zookeeper.ZooKeeper; public class Test{ /** *  server列表, 以逗号分割 */ protected static String hosts = "172.16.217.148:2181"; /** * 连接的超时时间, 毫秒 */ private final int SESSION_TIMEOUT = 5000; private CountDownLatch connectedSignal = new CountDownLatch(1); protected static ZooKeeper zk; private static String nodePath = "/Test/test1"; //static String data = "a very long string about data to set to zookeeper"; static int threads = 10;    //线程数 static int runs = 1000;     //迭代次数 static int start = 0;       //none static int size = 1024*4;   //写入数据的大小,单位：字节 static byte[] testdata;     //测试数据 public static void main(String[] args) throws Exception { PropertyConfigurator.configure("log4j.properties"); //生成写入的数据，大小size字节 testdata = new byte[size]; for(int i=0;i<size;i++){ testdata[i] = 'A'; } Test test = new Test(); //连接 test.connect(); WorkerStat[] statArray = new WorkerStat[threads]; Thread[] threadArray = new Thread[threads]; WorkerStat mainStat = new WorkerStat(); mainStat.runs = runs * threads; long begin = System.currentTimeMillis(); for (int i = 0; i < threads; i++) { statArray[i] = new WorkerStat(); statArray[i].start = start + i * runs; statArray[i].runs = runs; threadArray[i] = new SetterThread(statArray[i]); threadArray[i].start(); } for (int i = 0; i < threads; i++) { threadArray[i].join(); } mainStat.setterTime = System.currentTimeMillis() - begin; begin = System.currentTimeMillis(); for (int i = 0; i < threads; i++) { threadArray[i] = new GetterThread(statArray[i]); threadArray[i].start(); } for (int i = 0; i < threads; i++) { threadArray[i].join(); } mainStat.getterTime = System.currentTimeMillis() - begin; WorkerStat totalStat = new WorkerStat(); System.out.println("Test over!!"); System.out.println("Thread("+threads+")\truns\tset time(ms)\tget time(ms)"); for (int i = 0; i < threads; i++) { totalStat.runs = totalStat.runs + statArray[i].runs; totalStat.setterTime = totalStat.setterTime + statArray[i].setterTime; totalStat.getterTime = totalStat.getterTime + statArray[i].getterTime; } System.out.println("Total\t\t" + totalStat.runs + "\t"+ totalStat.setterTime + "\t\t" + totalStat.getterTime); System.out.println("Avg\t\t" + runs + "\t" + totalStat.setterTime/ threads + "\t\t" + totalStat.getterTime / threads); System.out.println("TPS\t\t\t" + 1000 * totalStat.runs/ totalStat.setterTime + "\t\t" + 1000 * totalStat.runs/ totalStat.getterTime); System.out.println("\nMain\t\t" + mainStat.runs + "\t"+ mainStat.setterTime + "\t\t" + mainStat.getterTime); System.out.println("TPS\t\t" + 1000 * mainStat.runs/ mainStat.setterTime + "\t\t" + 1000 * mainStat.runs/ mainStat.getterTime); } private static class WorkerStat { public int start, runs; public long setterTime, getterTime; public WorkerStat() { start = runs = 0; setterTime = getterTime = 0; } } private static class SetterThread extends Thread { private WorkerStat stat; SetterThread(WorkerStat stat) { this.stat = stat; } public void run() { long begin = System.currentTimeMillis(); for (int i = stat.start; i < stat.start + stat.runs; i++) { //写入节点数据 try { zk.setData(nodePath, testdata, -1); } catch (Exception e) { e.printStackTrace(); } } long end = System.currentTimeMillis(); stat.setterTime = end - begin; } } private static class GetterThread extends Thread { private WorkerStat stat; GetterThread(WorkerStat stat) { this.stat = stat; } public void run() { long begin = System.currentTimeMillis(); for (int i = stat.start; i < stat.start + stat.runs; i++) { //读取节点数据 try { zk.getData(nodePath, false, null); } catch (Exception e) { e.printStackTrace(); } } long end = System.currentTimeMillis(); stat.getterTime = end - begin; } }

//=============================================================================== /** * 连接zookeeper server */ public void connect() throws Exception { zk = new ZooKeeper(hosts, SESSION_TIMEOUT, new ConnWatcher()); // 等待连接完成 connectedSignal.await(); } /** * * @author Kiven * */ public class ConnWatcher implements Watcher{ public void process(WatchedEvent event) { // 连接建立, 回调process接口时, 其event.getState()为KeeperState.SyncConnected if (event.getState() == KeeperState.SyncConnected) { // 放开闸门, wait在connect方法上的线程将被唤醒 connectedSignal.countDown(); } } } /** 以下为各个参数的详细说明: path. znode的路径. data. 与znode关联的数据. acl. 指定权限信息, 如果不想指定权限, 可以传入Ids.OPEN_ACL_UNSAFE. 指定znode类型. CreateMode是一个枚举类, 从中选择一个成员传入即可. */ /** * 创建持久化节点 */ public void create(String Path, byte[] data) throws Exception { zk.create(Path, data, Ids.OPEN_ACL_UNSAFE, CreateMode.PERSISTENT); System.out.println("创建节点:"+Path); System.out.println("================="); } /** * *获取节点信息 *@author kiven *@createDate 2013-01-16 15:17:22 *@param path *@throws KeeperException *@throws InterruptedException */ public void getChild(String path) throws KeeperException, InterruptedException{ try{ List list=zk.getChildren(path, false); if(list.isEmpty()){ System.out.println(path+"中没有节点"); }else{ System.out.println(path+"中存在节点"); for(String child:list){ System.out.println("节点为："+child); } } }catch (KeeperException.NoNodeException e) { e.printStackTrace(); } } /** * 设置节点数据 * @throws Exception */ public void setData(String path,String data) throws Exception{ zk.setData(path, data.getBytes(), -1); System.out.println("set Data:"+"testSetData"); } /** * 读取节点数据 * @throws Exception */ public void getData() throws Exception{ System.out.println("get Data:"); zk.getData(nodePath, false, null); } /** * 删除节点 * @param path * @throws Exception */ public void delete(String path) throws Exception{ System.out.println("删除节点:"+path); //如果版本号与znode的版本号不一致，将无法删除，是一种乐观加锁机制；如果将版本号设置为-1，不会去检测版本，直接删除； zk.delete(path, -1); } /** * 关闭连接 */ public void close() { try { zk.close(); } catch (InterruptedException e) { e.printStackTrace(); } } }

　　相关功能代码中都已经注释清楚。我们看一下测试下来的结果： 写入数据： 

　　读取数据：

　　由上两张图我们可以得知：

　　1、写入（set）数据时，单个线程执行是最快的，而读取（get）数据则是4或5个线程并发是最快的。

　　2、每个节点存储的数据不能超过1M（在Zookeeper默认配置的情况下，如果写入的数据为1M会报错，无法写入数据，所以所有应用写入的数据必须<1M）

　　3、单个写入数据的体积越大，处理速度越慢，响应时间越长。

　　4、会管应用中写入的数据最大的是16方画面合成的时候，此时写入的数据可能大于4K，另外16方画面合成时候zk节点存放的数据（json）是包含大量空格和换行的（在zk节点中，空格和换行也是占大小的），因此建议压缩json数据。

　　5、测试过程中产生的log文件对磁盘的消耗和占用较大，建议定时删除历史log和shapshot

最新内容请见作者的GitHub页：http://qaseven.github.io/