保障服务的持续高可用、高性能及负载均衡

保障服务的持续高可用、高性能及负载均衡 项目主页

 

高可用: 服务多副本
高性能: 超时限制
负载均衡: 环形队列

已经实现的功能:
1. HTTP调用方式的搜索服务
2. REDIS访问

一句话说明白这个项目的原理:

后端有多个一模一样的搜索，还有多个一模一样的REDIS，中间层的业务逻辑不直接调用搜索和REDIS，而是通过这个组件来调用

 

 

在WEB项目中的使用方法

 

1. 编译依赖:

    git clone https://github.com/ysc/high-availability.git
    cd high-availability
    mvn install

2. 在 pom.xml 中指定以下依赖:

    <dependency>
        <groupId>org.apdplat.service</groupId>
        <artifactId>high-availability</artifactId>
        <version>1.0-SNAPSHOT</version>
    </dependency>

3. 在 src/main/resources 目录下新建目录 conf, 然后在 conf 目录下新建文件 conf.txt, 加入如下配置项:

    #search api depends on these servers
    search.api.server.urls=http://192.168.0.100:8080/search.jsp, http://192.168.0.101:8080/search.jsp
    #timeout config to guarantee response time
    search.api.timeout.seconds=1
    #whether output status to log in every unavailable urls valid process
    status.log.enabled=true
    #unavailable schedule initial delay seconds
    unavailable.schedule.initial.delay.seconds=60
    #unavailable schedule period seconds
    unavailable.schedule.period.seconds=60
    #redis servers
    redis.servers=192.168.0.102:6379:b01cbe1209a545a7cdb, 192.168.0.103:6379:b01cbe1209a545a7cdb

    redis.pool.blockWhenExhausted=false
    redis.pool.jmxEnabled=false
    redis.pool.lifo=true
    redis.pool.maxIdle=500
    redis.pool.maxTotal=500
    redis.pool.maxWaitMillis=500
    redis.pool.minEvictableIdleTimeMillis=1800000
    redis.pool.minIdle=0
    redis.pool.numTestsPerEvictionRun=3
    redis.pool.softMinEvictableIdleTimeMillis=1800000
    redis.pool.testOnBorrow=true
    redis.pool.testWhileIdle=true
    redis.pool.timeBetweenEvictionRunsMillis=-1
    redis.pool.readTimeoutMillis=2000

4. 在 web.xml 中配置一个监听器:

    <listener>
        <listener-class>org.apdplat.data.visualization.container.HighAvailabilityListener</listener-class>
    </listener>

5. 监听器 HighAvailabilityListener.java 类的代码如下:

    public class HighAvailabilityListener  implements ServletContextListener {
        private static final Logger LOGGER = LoggerFactory.getLogger(HighAvailabilityListener.class);

        @Override
        public void contextInitialized(ServletContextEvent sce) {
            String conf = sce.getServletContext().getRealPath("/WEB-INF/classes/conf/");
            LOGGER.info("启动搜索服务, 监控配置目录: {}", conf);
            ConfWatcher.startWatch(conf);
            sce.getServletContext().setAttribute("SearchAPI", new SearchAPIImpl());
        }

        @Override
        public void contextDestroyed(ServletContextEvent sce) {
            LOGGER.info("停止搜索服务");
            SearchAPI searchAPI = (SearchAPI)sce.getServletContext().getAttribute("SearchAPI");
            if(searchAPI != null){
                searchAPI.close();
            }
            LOGGER.info("停止监控配置目录");
            ConfWatcher.stopWatch();
        }
    }

6. 在 jsp 中调用搜索服务:

    SearchAPI searchAPI = (SearchAPI) application.getAttribute("SearchAPI");
    if(searchAPI == null) {
        out.println("搜索服务未启动");
        return;
    }
    String keyword = request.getParameter("kw") == null ? "CCTV-1" : request.getParameter("kw");
    int topN = 5;
    try{
        topN = Integer.parseInt(request.getParameter("topN"));
    }catch (Exception e){
        //
    }
    String result = null;
    long start = System.currentTimeMillis();
    result = searchAPI.search(keyword, topN);
    String cost = Utils.getTimeDes(System.currentTimeMillis()-start);
    // 如果想知道搜索服务的状态
    String status = searchAPI.getStatus().replace("\n", "<br/>").replace("\t", "&nbsp; &nbsp; &nbsp; &nbsp; ");

7. 获取REDIS连接:

    String DETECT_KEY = "redis_ha_detector";

    Jedis jedis = null;
    try{
        jedis = JedisAPI.getInstance().getJedis();
        // 如果所有REDIS服务都不可用, 则返回null
        if(jedis != null) {
            jedis.set(DETECT_KEY, String.valueOf(System.currentTimeMillis()));
            System.out.println(jedis.get(DETECT_KEY));
            System.out.println("REDIS服务状态:\n" + JedisAPI.getInstance().getStatus());
        }
    }finally {
        if(jedis != null){
            // 返回连接用完后必须要关闭, 调用close方法
            jedis.close();
        }
    }    

 

 

在非WEB项目中的使用方法

 

1. 编译依赖:

    git clone https://github.com/ysc/high-availability.git
    cd high-availability
    mvn install

2. 在 pom.xml 中指定以下依赖:

    <dependency>
        <groupId>org.apdplat.service</groupId>
        <artifactId>high-availability</artifactId>
        <version>1.0-SNAPSHOT</version>
    </dependency>

3. 新建目录 conf, 然后在 conf 目录下新建文件 conf.txt, 加入如下配置项:

    #search api depends on these servers
    search.api.server.urls=http://192.168.0.100:8080/search.jsp, http://192.168.0.101:8080/search.jsp
    #timeout config to guarantee response time
    search.api.timeout.seconds=1
    #whether output status to log in every unavailable urls valid process
    status.log.enabled=true
    #unavailable schedule initial delay seconds
    unavailable.schedule.initial.delay.seconds=60
    #unavailable schedule period seconds
    unavailable.schedule.period.seconds=60
    #redis servers
    redis.servers=192.168.0.102:6379:b01cbe1209a545a7cdb, 192.168.0.103:6379:b01cbe1209a545a7cdb

    redis.pool.blockWhenExhausted=false
    redis.pool.jmxEnabled=false
    redis.pool.lifo=true
    redis.pool.maxIdle=500
    redis.pool.maxTotal=500
    redis.pool.maxWaitMillis=500
    redis.pool.minEvictableIdleTimeMillis=1800000
    redis.pool.minIdle=0
    redis.pool.numTestsPerEvictionRun=3
    redis.pool.softMinEvictableIdleTimeMillis=1800000
    redis.pool.testOnBorrow=true
    redis.pool.testWhileIdle=true
    redis.pool.timeBetweenEvictionRunsMillis=-1
    redis.pool.readTimeoutMillis=2000

4. 将conf目录加入classpath:

    java -cp conf:xxx-1.0-SNAPSHOT-jar-with-dependencies.jar 

5. 启动搜索服务并监控配置目录的代码如下:

    Path conf = Paths.get(ConfWatcher.class.getResource("/conf/").getPath());
    LOGGER.info("启动搜索服务, 监控配置目录: {}", conf);
    ConfWatcher.startWatch(conf);
    SearchAPI searchAPI = new SearchAPIImpl();

6. 调用搜索服务:

    String keyword = "CCTV-1";
    int topN = 5;

    String result = null;
    long start = System.currentTimeMillis();
    result = searchAPI.search(keyword, topN);
    String cost = Utils.getTimeDes(System.currentTimeMillis()-start);

    // 如果想知道搜索服务的状态
    String status = searchAPI.getStatus();

7. 获取REDIS连接:

    String DETECT_KEY = "redis_ha_detector";

    Jedis jedis = null;
    try{
        jedis = JedisAPI.getInstance().getJedis();
        // 如果所有REDIS服务都不可用, 则返回null
        if(jedis != null) {
            jedis.set(DETECT_KEY, String.valueOf(System.currentTimeMillis()));
            System.out.println(jedis.get(DETECT_KEY));
            System.out.println("REDIS服务状态:\n" + JedisAPI.getInstance().getStatus());
        }
    }finally {
        if(jedis != null){
            // 返回连接用完后必须要关闭, 调用close方法
            jedis.close();
        }
    }

 

 

目录结构

 

.
├── README.md
├── pom.xml
└── src
    └── main
        ├── java
        │   └── org
        │       └── apdplat
        │           └── service
        │               ├── api
        │               │   ├── JedisAPI.java
        │               │   └── SearchAPI.java
        │               ├── configration
        │               │   ├── ConfManager.java
        │               │   ├── ConfTools.java
        │               │   └── ConfWatcher.java
        │               ├── impl
        │               │   └── SearchAPIImpl.java
        │               └── utils
        │                   └── TimeUtils.java
        └── resources
            └── conf
                ├── conf.production.txt
                └── conf.txt

12 directories, 11 files

 

 

 

 

评论

2 楼 yangshangchuan 2017-01-27  

herman_liu76 写道

核心的东西没有呀？
集群选择策略，环型队列的结构与算法，为何不用一致性HASH？

这里的场景不需要使用Consistent hashing.

1 楼 herman_liu76 2017-01-25  

核心的东西没有呀？
集群选择策略，环型队列的结构与算法，为何不用一致性HASH？