KnowStreaming系列教程第三篇—

KnowStreaming系列教程第三篇——调度任务模块

news2024/11/23 15:38:05

前一篇文章KnowStreaming系列教程第二篇——项目整体架构分析_诸葛子房_的博客-CSDN博客

讲述了KS的整体项目目录，这边文章来讲述下KS在调度模块里面对于指标采集和元数据同步

一、调度模块代码主要在km-task里面

public class TaskClusterAddedListener implements ApplicationListener<ClusterPhyAddedEvent> {
    private static final ILog LOGGER = LogFactory.getLog(TaskClusterAddedListener.class);

    @Override
    public void onApplicationEvent(ClusterPhyAddedEvent event) {
        LOGGER.info("method=onApplicationEvent||clusterPhyId={}||msg=listened new cluster", event.getClusterPhyId());
        Long now = System.currentTimeMillis();

        // 交由KS自定义的线程池，异步执行任务
        FutureUtil.quickStartupFutureUtil.submitTask(() -> triggerAllTask(event.getClusterPhyId(), now));
    }

    private void triggerAllTask(Long clusterPhyId, Long startTimeUnitMs) {
        ClusterPhy tempClusterPhy = null;

        // 120秒内无加载进来，则直接返回退出
        while (System.currentTimeMillis() - startTimeUnitMs <= 120L * 1000L) {
            tempClusterPhy = LoadedClusterPhyCache.getByPhyId(clusterPhyId);
            if (tempClusterPhy != null) {
                break;
            }

            BackoffUtils.backoff(1000);
        }

        if (tempClusterPhy == null) {
            return;
        }

        // 获取到之后，再延迟5秒，保证相关的集群都被正常加载进来，这里的5秒不固定
        BackoffUtils.backoff(5000);
        final ClusterPhy clusterPhy = tempClusterPhy;

        // 集群执行集群元信息同步
        List<AbstractAsyncMetadataDispatchTask> metadataServiceList = new ArrayList<>(SpringTool.getBeansOfType(AbstractAsyncMetadataDispatchTask.class).values());
        for (AbstractAsyncMetadataDispatchTask dispatchTask: metadataServiceList) {
            try {
                dispatchTask.asyncProcessSubTask(clusterPhy, startTimeUnitMs);
            } catch (Exception e) {
                // ignore
            }
        }

        // 再延迟5秒，保证集群元信息都已被正常同步至DB，这里的5秒不固定
        BackoffUtils.backoff(5000);

        // 集群集群指标采集
        List<AbstractAsyncMetricsDispatchTask> metricsServiceList = new ArrayList<>(SpringTool.getBeansOfType(AbstractAsyncMetricsDispatchTask.class).values());
        for (AbstractAsyncMetricsDispatchTask dispatchTask: metricsServiceList) {
            try {
                dispatchTask.asyncProcessSubTask(clusterPhy, startTimeUnitMs);
            } catch (Exception e) {
                // ignore
            }
        }
    }
}

通过监听集群添加事件，触发元数据同步和指标采集调度任务

具体实现可参考：

spring 根据接口或者抽象类获取子类执行: https://blog.csdn.net/u012501054/article/details/103927674

二、调度任务分布式系统如何做到单节点运行，避免多台机器调度

AbstractDispatchTask 里面的execute 方法通过实现任务分配

public TaskResult execute(JobContext jobContext) {
        try {
            long triggerTimeUnitMs = System.currentTimeMillis();

            // 获取所有的任务
            List<E> allTaskList = this.listAllTasks();
            if (ValidateUtils.isEmptyList(allTaskList)) {
                LOGGER.debug("all-task is empty, finish process, taskName:{} jobContext:{}", taskName, jobContext);
                return TaskResult.SUCCESS;
            }

            // 计算当前机器需要执行的任务
            List<E> subTaskList = this.selectTask(allTaskList, jobContext.getAllWorkerCodes(), jobContext.getCurrentWorkerCode());

            if (ValidateUtils.isEmptyList(allTaskList)) {
                LOGGER.debug("sub-task is empty, finish process, taskName:{} jobContext:{}", taskName, jobContext);
                return TaskResult.SUCCESS;
            }

            // 进行任务处理
            TaskResult ret = this.processTask(subTaskList, triggerTimeUnitMs);

            //组装信息
            TaskResult taskResult = new TaskResult();
            taskResult.setCode(ret.getCode());
            taskResult.setMessage(ConvertUtil.list2String(subTaskList, ","));

            return taskResult;

        } catch (Exception e) {
            LOGGER.error("process task failed, taskName:{}", taskName, e);

            return new TaskResult(TaskResult.FAIL_CODE, e.toString());
        }
    }

对应代码解释如下：