聊聊OtterController

序

本文主要研究一下OtterController

OtterController

otter/node/etl/src/main/java/com/alibaba/otter/node/etl/OtterController.java

public class OtterController implements NodeTaskListener, OtterControllerMBean {

    private static final Logger                   logger      = LoggerFactory.getLogger(OtterController.class);

    // 第一层为pipelineId，第二层为S.E.T.L模块
    private Map<Long, Map<StageType, GlobalTask>> controllers = OtterMigrateMap.makeComputingMap(new Function<Long, Map<StageType, GlobalTask>>() {

                                                                  public Map<StageType, GlobalTask> apply(Long pipelineId) {
                                                                      return new MapMaker().makeMap();
                                                                  }
                                                              });
    private ConfigClientService                   configClientService;
    private ArbitrateManageService                arbitrateManageService;
    private NodeTaskService                       nodeTaskService;
    // 各种资源管理
    private DataSourceService                     dataSourceService;                                           // 连接池资源
    private DbDialectFactory                      dbDialectFactory;                                            // 数据库信息资源
    private ArbitrateEventService                 arbitrateEventService;                                       // 仲裁器资源
    private ExecutorService                       executorService;

    private StageAggregationCollector             stageAggregationCollector;

    public void start() throws Throwable {
        // 初始化节点
        initNid();
        nodeTaskService.addListener(this); // 将自己添加为NodeTask响应者
    }

    public void stop() throws Throwable {
        for (Map<StageType, GlobalTask> tasks : controllers.values()) {
            for (GlobalTask task : tasks.values()) {
                try {
                    task.shutdown();
                } catch (Exception e) {
                    logger.error("##shutdown task error!", e);
                }
            }
        }

        try {
            Long nid = configClientService.currentNode().getId();
            arbitrateManageService.nodeEvent().destory(Long.valueOf(nid));
        } catch (Exception e) {
            logger.error("##destory node error!", e);
        }

        try {
            arbitrateEventService.toolEvent().release();
        } catch (Exception e) {
            logger.error("##destory arbitrate error!", e);
        }

        try {
            nodeTaskService.stopNode(); // 通知manager停止当前node
        } catch (Exception e) {
            logger.error("##stop node error!", e);
        }

        try {
            OtterContextLocator.close();
        } catch (Exception e) {
            logger.error("##cloes spring error!", e);
        }

        ZooKeeperClient.destory();// 关闭zookeeper
    }

    //......

}

• OtterController实现了NodeTaskListener接口，提供了start、stop、process方法；其start方法主要是执行initNid及nodeTaskService.addListener(this)；其stop方法则遍历controllers的GlobalTask，挨个执行其shutdown方法，然后执行arbitrateManageService.nodeEvent().destory()、arbitrateEventService.toolEvent().release()、nodeTaskService.stopNode()、OtterContextLocator.close()及ZooKeeperClient.destory()；process方法主要是遍历nodeTasks，挨个执行stopPipeline或者startPipeline

initNid

otter/node/etl/src/main/java/com/alibaba/otter/node/etl/OtterController.java

    private void initNid() {
        // 获取一下nid变量
        String nid = System.getProperty(OtterConstants.NID_NAME);
        if (StringUtils.isEmpty(nid)) {
            throw new ConfigException("nid is not set!");
        }
        logger.info("INFO ## the nodeId = {}", nid);
        checkNidVaild(nid);
        arbitrateManageService.nodeEvent().init(Long.valueOf(nid));
        // 添加session expired处理
        NodeSessionExpired sessionExpired = new NodeSessionExpired();
        sessionExpired.setNodeEvent(arbitrateManageService.nodeEvent());
        ZooKeeperClient.registerNotification(sessionExpired);
    }

• initNid方法主要执行arbitrateManageService.nodeEvent().init(Long.valueOf(nid))及ZooKeeperClient.registerNotification(sessionExpired)

startPipeline

otter/node/etl/src/main/java/com/alibaba/otter/node/etl/OtterController.java

    public void startPipeline(NodeTask nodeTask) {
        Long pipelineId = nodeTask.getPipeline().getId();
        releasePipeline(pipelineId);
        Map<StageType, GlobalTask> tasks = controllers.get(pipelineId);
        // 处理具体的任务命令
        List<StageType> stage = nodeTask.getStage();
        List<TaskEvent> event = nodeTask.getEvent();
        for (int i = 0; i < stage.size(); i++) {
            StageType stageType = stage.get(i);
            TaskEvent taskEvent = event.get(i);
            if (taskEvent.isCreate()) {
                startTask(nodeTask.getPipeline(), tasks, stageType);
            } else {
                stopTask(tasks, stageType);
            }
        }
    }

    private void startTask(Pipeline pipeline, Map<StageType, GlobalTask> tasks, StageType taskType) {
        if (tasks.get(taskType) != null && tasks.get(taskType).isAlive()) {
            logger.warn("WARN ## this task = {} has started", taskType);
        }

        GlobalTask task = null;
        if (taskType.isSelect()) {
            task = new SelectTask(pipeline.getId());
        } else if (taskType.isExtract()) {
            task = new ExtractTask(pipeline.getId());
        } else if (taskType.isTransform()) {
            task = new TransformTask(pipeline.getId());
        } else if (taskType.isLoad()) {
            task = new LoadTask(pipeline.getId());
        }

        if (task != null) {
            OtterContextLocator.autowire(task); // 注入一下spring资源
            task.start();
            tasks.put(taskType, task);
            logger.info("INFO ## start this task = {} success", taskType.toString());
        }
    }

    private void stopTask(Map<StageType, GlobalTask> tasks, StageType taskType) {
        GlobalTask task = tasks.remove(taskType);
        if (task != null) {
            task.shutdown();
            logger.info("INFO ## taskName = {} has shutdown", taskType);
        } else {
            logger.info("INFo ## taskName = {} is not started", taskType);
        }

    }

• startPipeline方法先执行releasePipeline，然后遍历指定pipeline的tasks，挨个执行startTask或者stopTask；startTask方法根据taskType类创建不同类型的GlobalTask，然后执行task.start()；stopTask则从tasks移除指定taskType的GlobalTask，然后执行task的shutdown方法

stopPipeline

otter/node/etl/src/main/java/com/alibaba/otter/node/etl/OtterController.java

    private void stopPipeline(Long pipelineId, Map<StageType, GlobalTask> tasks) {
        for (GlobalTask task : tasks.values()) {
            try {
                task.shutdown();
            } catch (Exception e) {
                logger.error("## stop s/e/t/l task error!", e);
            } finally {
                tasks.remove(task);
            }
        }
        // close other resources.
        try {
            Thread.sleep(1 * 1000); // sleep 5s，等待S.E.T.L释放线程
        } catch (InterruptedException e) {
            logger.error("ERROR ## ", e);
        }

        // 释放资源
        releasePipeline(pipelineId);
        arbitrateEventService.toolEvent().release(pipelineId);
    }

    private void releasePipeline(Long pipelineId) {
        dataSourceService.destroy(pipelineId);
        dbDialectFactory.destory(pipelineId);
    }

• stopPipeline方法则遍历tasks挨个执行task.shutdown()及tasks.remove(task)，最后执行releasePipeline及arbitrateEventService.toolEvent().release(pipelineId)

小结

OtterController实现了NodeTaskListener接口，提供了start、stop、process方法；其start方法主要是执行initNid及nodeTaskService.addListener(this)；其stop方法则遍历controllers的GlobalTask，挨个执行其shutdown方法，然后执行arbitrateManageService.nodeEvent().destory()、arbitrateEventService.toolEvent().release()、nodeTaskService.stopNode()、OtterContextLocator.close()及ZooKeeperClient.destory()；process方法主要是遍历nodeTasks，挨个执行stopPipeline或者startPipeline

doc

• OtterController

来源：oschina

链接：https://my.oschina.net/go4it/blog/4306870