ElasticJob - 任务执行过程解析

ElasticJob是基于Quartz实现的弹性分布式任务调度框架，整个任务触发的底层是Quartz来触发。LiteJob框架触发任务执行的开始，下面来分析任务的执行过程。

public final class LiteJob implements Job {
    
    @Override
    public void execute(final JobExecutionContext context) throws JobExecutionException {
        //根据任务类型获取执行器，SimpleJobExecutor
        JobExecutorFactory.getJobExecutor(elasticJob, jobFacade).execute();
    }
}

ElasticJob根据任务的作业类型，获取不同的作业执行器，这里以SimpleJobExecutor为例。它继承了AbstractElasticJobExecutor，它是所有任务执行流程的模板类。

protected AbstractElasticJobExecutor(final JobFacade jobFacade) {
        this.jobFacade = jobFacade;
        jobRootConfig = jobFacade.loadJobRootConfiguration(true);
        jobName = jobRootConfig.getTypeConfig().getCoreConfig().getJobName();
        executorService = ExecutorServiceHandlerRegistry.getExecutorServiceHandler(jobName, (ExecutorServiceHandler) getHandler(JobProperties.JobPropertiesEnum.EXECUTOR_SERVICE_HANDLER));
        jobExceptionHandler = (JobExceptionHandler) getHandler(JobProperties.JobPropertiesEnum.JOB_EXCEPTION_HANDLER);
        itemErrorMessages = new ConcurrentHashMap<>(jobRootConfig.getTypeConfig().getCoreConfig().getShardingTotalCount(), 1);
    }

在构造方法里面，是根据JobName为每个Job实例初始化了一个任务执行线程池，就是一个普通的ThreadPoolExecutor封装。下面分析下任务执行的流程：

 public void execute() {
            try {
                //TODO 检验作业服务器和zk服务器系统时差是否可忍受，默认是不开启检查的
                jobFacade.checkJobExecutionEnvironment();
            } catch (final JobExecutionEnvironmentException cause) {
                jobErrorHandler.handleException(jobConfig.getJobName(), cause);
            }
            //TODO 获取分片的上下文信息
            ShardingContexts shardingContexts = jobFacade.getShardingContexts();
            jobFacade.postJobStatusTraceEvent(shardingContexts.getTaskId(), State.TASK_STAGING, String.format("Job '%s' execute begin.", jobConfig.getJobName()));
            //TODO 查看当前任务是否有分片正在执行，就创建节点/sharding/{item}/misfire
            if (jobFacade.misfireIfRunning(shardingContexts.getShardingItemParameters().keySet())) {
                jobFacade.postJobStatusTraceEvent(shardingContexts.getTaskId(), State.TASK_FINISHED, String.format(
                        "Previous job '%s' - shardingItems '%s' is still running, misfired job will start after previous job completed.", jobConfig.getJobName(),
                        shardingContexts.getShardingItemParameters().keySet()));
                return;
            }
            try {
                //TODO 任务开始执行，这里会去执行业务配置的监听器ElasticJobListener
                jobFacade.beforeJobExecuted(shardingContexts);
                //CHECKSTYLE:OFF
            } catch (final Throwable cause) {
                //CHECKSTYLE:ON
                jobErrorHandler.handleException(jobConfig.getJobName(), cause);
            }
            //TODO 任务的执行处理
            execute(shardingContexts, ExecutionSource.NORMAL_TRIGGER);
            //TODO 任务是否有错过执行，重新出发调度一次
            while (jobFacade.isExecuteMisfired(shardingContexts.getShardingItemParameters().keySet())) {
                jobFacade.clearMisfire(shardingContexts.getShardingItemParameters().keySet());
                execute(shardingContexts, ExecutionSource.MISFIRE);
            }
            //TODO 执行故障转移，防止未及时故障转移吧，其实有个listener在监控，为什么这里还要手动执行下 故障转移
            jobFacade.failoverIfNecessary();
            try {
                //TODO 任务执行完成，这里是去触发业务配置的监听器的执行
                jobFacade.afterJobExecuted(shardingContexts);
                //CHECKSTYLE:OFF
            } catch (final Throwable cause) {
                //CHECKSTYLE:ON
                jobErrorHandler.handleException(jobConfig.getJobName(), cause);
            }
        }
    

• Job节点服务器和zookeeper时钟差校验，对应的是Job的maxTimeDiffSeconds最大始终差配置，默认是-1不检查

try {
            //TODO 检验作业服务器和zk服务器系统时差是否可忍受，默认是不开启检查的
            jobFacade.checkJobExecutionEnvironment();
        } catch (final JobExecutionEnvironmentException cause) {
            jobErrorHandler.handleException(jobConfig.getJobName(), cause);
        }

    public void checkMaxTimeDiffSecondsTolerable() throws JobExecutionEnvironmentException {
        int maxTimeDiffSeconds = load(true).getMaxTimeDiffSeconds();
        if (-1 == maxTimeDiffSeconds) {
            return;
        }
        long timeDiff = Math.abs(timeService.getCurrentMillis() - jobNodeStorage.getRegistryCenterTime());
        if (timeDiff > maxTimeDiffSeconds * 1000L) {
            throw new JobExecutionEnvironmentException(
                    "Time different between job server and register center exceed '%s' seconds, max time different is '%s' seconds.", timeDiff / 1000, maxTimeDiffSeconds);
        }
    }

• 获取当前节点服务器的分片信息，这里是分片的关键，在分片那节再细说，这里大概说下做的事情。如果Job配置了故障转移并且存在故障转移分片，优先执行从其它故障节点转移到当前节点服务器的分片。如果需要重新分片，则执行重新分片逻辑，获取分配到当前节点的分片信息。

public ShardingContexts getShardingContexts() {
        boolean isFailover = configService.load(true).isFailover();
        if (isFailover) {
            //TODO 获取故障转移到当前节点的分片信息
            List<Integer> failoverShardingItems = failoverService.getLocalFailoverItems();
            if (!failoverShardingItems.isEmpty()) {
                //TODO 获取故障转移分片信息
                return executionContextService.getJobShardingContext(failoverShardingItems);
            }
        }
        shardingService.shardingIfNecessary();
        List<Integer> shardingItems = shardingService.getLocalShardingItems();
        if (isFailover) {
            //TODO 删除本节点被故障转移的分片信息
            shardingItems.removeAll(failoverService.getLocalTakeOffItems());
        }
        shardingItems.removeAll(executionService.getDisabledItems(shardingItems));
        return executionContextService.getJobShardingContext(shardingItems);
    }

• 当前节点服务器该job对应的所有分片，是否有执行中的分片， 只要有一个执行中的分片，则对应的所有分片都会被设置为misfire错失执行，然后退出。 就是上次触发的执行没有完成，本次触发不会执行，会等待上次执行完成，防止任务分片在当前节点被重复执行。

  //TODO 查看当前任务是否有分片正在执行，就创建节点/sharding/{item}/misfire

  if (jobFacade.misfireIfRunning(shardingContexts.getShardingItemParameters().keySet())) {
            jobFacade.postJobStatusTraceEvent(shardingContexts.getTaskId(), State.TASK_FINISHED, String.format(
                    "Previous job '%s' - shardingItems '%s' is still running, misfired job will start after previous job completed.", jobConfig.getJobName(),
                    shardingContexts.getShardingItemParameters().keySet()));
            return;
        }

public boolean misfireIfHasRunningItems(final Collection<Integer> items) {
        if (!hasRunningItems(items)) {
            return false;
        }
        //TODO 只要其中一个分片没有执行完成，就会设置misfire
        setMisfire(items);
        return true;
    }

public boolean hasRunningItems(final Collection<Integer> items) {
        JobConfiguration jobConfig = configService.load(true);
        if (!jobConfig.isMonitorExecution()) {
            return false;
        }
        for (int each : items) {
            //TODO 只要其中一个分片没有执行完成，就会设置misfire
            if (jobNodeStorage.isJobNodeExisted(ShardingNode.getRunningNode(each))) {
                return true;
            }
        }
        return false;
    }

• 执行任务之前对应的ElasticJobListener，这个是之前在配置Job任务时对应的ElasticJobListener。

  
try {
            //TODO 任务开始执行，这里会去执行业务配置的监听器ElasticJobListener
            jobFacade.beforeJobExecuted(shardingContexts);
            //CHECKSTYLE:OFF
        } catch (final Throwable cause) {
            //CHECKSTYLE:ON
            jobErrorHandler.handleException(jobConfig.getJobName(), cause);
        }

• 具体任务分片的调度执行，在分片任务执行前，在当前节点服务器把任务在内存中设置为running状态，如果配置了幂等机制，同时也会在zookeeper中创建/sharding/{item}/running临时节点，在服务器宕机该节点就会自动删除了。然后就是把所有的分片任务提交到ThreadPoolExecutor中执行。这里使用的CountDownLatch控制并发等待所有的分片任务都完成，然后就删除任务执行钱创建的/sharding/{item}/running节点。同时如果开启了故障转移机制，就会删除之前创建的故障转移节点/sharding/{item}/failover

 //TODO 任务的执行处理
    execute(shardingContexts, ExecutionSource.NORMAL_TRIGGER);
    
    private void execute(final ShardingContexts shardingContexts, final ExecutionSource executionSource) {
            if (shardingContexts.getShardingItemParameters().isEmpty()) {
                jobFacade.postJobStatusTraceEvent(shardingContexts.getTaskId(), State.TASK_FINISHED, String.format("Sharding item for job '%s' is empty.", jobConfig.getJobName()));
                return;
            }
            //TODO 将当前任务设置为运行中Running的状态，分为内存状态和zk状态，在zk中创建一个临时节点
            jobFacade.registerJobBegin(shardingContexts);
            String taskId = shardingContexts.getTaskId();
            jobFacade.postJobStatusTraceEvent(taskId, State.TASK_RUNNING, "");
            try {
                process(shardingContexts, executionSource);
            } finally {
                // TODO Consider increasing the status of job failure, and how to handle the overall loop of job failure
                //TODO 清除job的running状态和删除故障转移节点/sharding/{item}/failover
                jobFacade.registerJobCompleted(shardingContexts);
                if (itemErrorMessages.isEmpty()) {
                    jobFacade.postJobStatusTraceEvent(taskId, State.TASK_FINISHED, "");
                } else {
                    jobFacade.postJobStatusTraceEvent(taskId, State.TASK_ERROR, itemErrorMessages.toString());
                }
            }
        }
    
    private void process(final ShardingContexts shardingContexts, final ExecutionSource executionSource) {
            Collection<Integer> items = shardingContexts.getShardingItemParameters().keySet();
            if (1 == items.size()) {
                int item = shardingContexts.getShardingItemParameters().keySet().iterator().next();
                JobExecutionEvent jobExecutionEvent = new JobExecutionEvent(IpUtils.getHostName(), IpUtils.getIp(), shardingContexts.getTaskId(), jobConfig.getJobName(), executionSource, item);
                process(shardingContexts, item, jobExecutionEvent);
                return;
            }
            CountDownLatch latch = new CountDownLatch(items.size());
            for (int each : items) {
                JobExecutionEvent jobExecutionEvent = new JobExecutionEvent(IpUtils.getHostName(), IpUtils.getIp(), shardingContexts.getTaskId(), jobConfig.getJobName(), executionSource, each);
                if (executorService.isShutdown()) {
                    return;
                }
                executorService.submit(() -> {
                    try {
                        process(shardingContexts, each, jobExecutionEvent);
                    } finally {
                        latch.countDown();
                    }
                });
            }
            try {
                latch.await();
            } catch (final InterruptedException ex) {
                Thread.currentThread().interrupt();
            }
        }

• 检查是否有任务分片错过执行，这里的判断是开启了misfire任务错过机制并且分片存在/sharding/{item}/misfire节点。如果开启了任务错过机制，并且有任务错过执行，这里会重新执行一次分片节点任务。任务执行之前清除了/sharding/{item}/misfire节点。

//TODO 任务是否有错过执行，重新出发调度一次
        while (jobFacade.isExecuteMisfired(shardingContexts.getShardingItemParameters().keySet())) {
            jobFacade.clearMisfire(shardingContexts.getShardingItemParameters().keySet());
            execute(shardingContexts, ExecutionSource.MISFIRE);
        }

• 在这里再手动执行下故障转移操作， 在这里不知道为啥还需要手动执行下故障转移，在故障转移监听器里面就能监听到，为啥这里还要在任务执行的时候再手动触发下？？？没太理解

//TODO 执行故障转移，防止未及时故障转移吧，其实有个listener在监控，为什么这里还要手动执行下 故障转移
jobFacade.failoverIfNecessary();

• 任务分片执行完之后，触发业务给任务配置的监听器的执行

 try {
                //TODO 任务执行完成，这里是去触发业务配置的监听器的执行
                jobFacade.afterJobExecuted(shardingContexts);
                //CHECKSTYLE:OFF
            } catch (final Throwable cause) {
                //CHECKSTYLE:ON
                jobErrorHandler.handleException(jobConfig.getJobName(), cause);
            }
    

作业执行流程图:

ElasticJob作业执行流程图