Search in sources :

Example 6 with ManagerContralFactory

use of com.bonree.brfs.schedulers.ManagerContralFactory in project BRFS by zhangnianli.

the class CopyCheckJob method operation.

@Override
public void operation(JobExecutionContext context) throws Exception {
    LOG.info("createCheck Copy Job working");
    ManagerContralFactory mcf = ManagerContralFactory.getInstance();
    MetaTaskManagerInterface release = mcf.getTm();
    StorageRegionManager snm = mcf.getSnm();
    ServiceManager sm = mcf.getSm();
    List<String> srs = TaskStateLifeContral.getSRs(snm);
    TaskStateLifeContral.watchSR(release, srs, TaskType.SYSTEM_COPY_CHECK.name());
    // 判断是否有恢复任务,有恢复任务则不进行创建
    if (WatchSomeThingJob.getState(WatchSomeThingJob.RECOVERY_STATUSE)) {
        LOG.warn("rebalance task is running !! skip check copy task");
        return;
    }
    String taskType = TaskType.SYSTEM_COPY_CHECK.name();
    List<Service> services = sm.getServiceListByGroup(Configs.getConfiguration().GetConfig(CommonConfigs.CONFIG_DATA_SERVICE_GROUP_NAME));
    if (services == null || services.isEmpty()) {
        LOG.warn("skip create {} task, because service is empty", taskType);
        return;
    }
    List<StorageRegion> snList = snm.getStorageRegionList();
    if (snList == null || snList.isEmpty()) {
        LOG.warn("storagename list is null");
        return;
    }
    // 1.获取sn创建任务的实际那
    TaskTypeModel tmodel = release.getTaskTypeInfo(taskType);
    if (tmodel == null) {
        tmodel = new TaskTypeModel();
        tmodel.setSwitchFlag(true);
        release.setTaskTypeModel(taskType, tmodel);
    }
    Map<String, Long> sourceTimes = tmodel.getSnTimes();
    LOG.debug("update init sn time :{}", sourceTimes);
    // 2.过滤不符合副本校验的sn信息
    List<StorageRegion> needSns = CopyCountCheck.filterSn(snList, services.size());
    // 3.针对第一次出现的sn补充时间
    sourceTimes = CopyCountCheck.repairTime(sourceTimes, needSns);
    Map<String, List<String>> losers = CopyCountCheck.collectLossFile(needSns, services, sourceTimes);
    LOG.info("update before sn time :{}", sourceTimes);
    Pair<TaskModel, Map<String, Long>> pair = CreateSystemTask.creatTaskWithFiles(sourceTimes, losers, needSns, TaskType.SYSTEM_COPY_CHECK, RECOVERY_NUM, 0);
    if (pair == null) {
        LOG.warn("create pair is empty !!!!");
        return;
    }
    TaskModel task = pair.getFirst();
    String taskName = null;
    if (task != null) {
        List<String> servers = CreateSystemTask.getServerIds(services);
        taskName = CreateSystemTask.updateTask(release, task, servers, TaskType.SYSTEM_COPY_CHECK);
    }
    if (!BrStringUtils.isEmpty(taskName)) {
        LOG.info("create {} {} task successfull !!!", taskType, taskName);
    }
    sourceTimes = pair.getSecond();
    // 更新sn临界信息
    tmodel = release.getTaskTypeInfo(taskType);
    if (tmodel == null) {
        tmodel = new TaskTypeModel();
        tmodel.setSwitchFlag(true);
        LOG.warn("taskType {} metadata loss create against !!", taskType);
    }
    tmodel.putAllSnTimes(sourceTimes);
    release.setTaskTypeModel(taskType, tmodel);
    LOG.debug("update sn time {}", sourceTimes);
    createTransferTasks(release);
}
Also used : Service(com.bonree.brfs.common.service.Service) TaskTypeModel(com.bonree.brfs.schedulers.task.model.TaskTypeModel) ManagerContralFactory(com.bonree.brfs.schedulers.ManagerContralFactory) StorageRegion(com.bonree.brfs.duplication.storageregion.StorageRegion) ServiceManager(com.bonree.brfs.common.service.ServiceManager) List(java.util.List) MetaTaskManagerInterface(com.bonree.brfs.schedulers.task.manager.MetaTaskManagerInterface) Map(java.util.Map) StorageRegionManager(com.bonree.brfs.duplication.storageregion.StorageRegionManager) TaskModel(com.bonree.brfs.schedulers.task.model.TaskModel)

Example 7 with ManagerContralFactory

use of com.bonree.brfs.schedulers.ManagerContralFactory in project BRFS by zhangnianli.

the class ManagerMetaTaskJob method operation.

@Override
public void operation(JobExecutionContext context) throws Exception {
    LOG.info("revise task work");
    JobDataMap data = context.getJobDetail().getJobDataMap();
    // 任务过期时间 ms
    String ttlTimeStr = data.getString(JobDataMapConstract.TASK_EXPIRED_TIME);
    LOG.info("task ttl time : {}", ttlTimeStr);
    long ttlTime = Long.parseLong(ttlTimeStr);
    ManagerContralFactory mcf = ManagerContralFactory.getInstance();
    MetaTaskManagerInterface release = mcf.getTm();
    // 获取可用服务
    String groupName = mcf.getGroupName();
    ServiceManager sm = mcf.getSm();
    // 2.设置可用服务
    List<String> serverIds = getServerIds(sm, groupName);
    if (serverIds == null || serverIds.isEmpty()) {
        LOG.warn("available server list is null");
        return;
    }
    for (TaskType taskType : TaskType.values()) {
        try {
            release.reviseTaskStat(taskType.name(), ttlTime, serverIds);
        } catch (Exception e) {
            LOG.warn("{}", e.getMessage());
            EmailPool emailPool = EmailPool.getInstance();
            MailWorker.Builder builder = MailWorker.newBuilder(emailPool.getProgramInfo());
            builder.setModel(this.getClass().getSimpleName() + "模块服务发生问题");
            builder.setException(e);
            builder.setMessage("管理任务数据发生错误");
            builder.setVariable(data.getWrappedMap());
            emailPool.sendEmail(builder);
        }
    }
    LOG.info("revise task success !!!");
}
Also used : JobDataMap(org.quartz.JobDataMap) ServiceManager(com.bonree.brfs.common.service.ServiceManager) TaskType(com.bonree.brfs.common.task.TaskType) MetaTaskManagerInterface(com.bonree.brfs.schedulers.task.manager.MetaTaskManagerInterface) EmailPool(com.bonree.brfs.email.EmailPool) ManagerContralFactory(com.bonree.brfs.schedulers.ManagerContralFactory) UnableToInterruptJobException(org.quartz.UnableToInterruptJobException)

Example 8 with ManagerContralFactory

use of com.bonree.brfs.schedulers.ManagerContralFactory in project BRFS by zhangnianli.

the class OperationTaskJob method operation.

@Override
public void operation(JobExecutionContext context) {
    JobDataMap data = context.getJobDetail().getJobDataMap();
    String dataPath = data.getString(JobDataMapConstract.DATA_PATH);
    ManagerContralFactory mcf = ManagerContralFactory.getInstance();
    MetaTaskManagerInterface release = mcf.getTm();
    if (release == null) {
        throw new NullPointerException("MetaTaskManager is empty !!!");
    }
    List<TaskType> switchList = mcf.getTaskOn();
    if (switchList == null || switchList.isEmpty()) {
        LOG.warn("switch task is empty !!!");
        return;
    }
    SchedulerManagerInterface schd = mcf.getStm();
    if (schd == null) {
        throw new NullPointerException("SchedulerManagerInterface is empty !!!");
    }
    RunnableTaskInterface runTask = mcf.getRt();
    if (runTask == null) {
        throw new NullPointerException("RunnableTaskInterface is empty !!!");
    }
    String typeName;
    String currentTaskName;
    TaskModel task;
    TaskRunPattern runPattern;
    int poolSize;
    int sumbitSize;
    String serverId = mcf.getServerId();
    SumbitTaskInterface sumbitTask;
    // 判断是否有恢复任务,有恢复任务则不进行创建
    boolean rebalanceFlag = WatchSomeThingJob.getState(WatchSomeThingJob.RECOVERY_STATUSE);
    for (TaskType taskType : switchList) {
        sumbitTask = null;
        try {
            if (TaskType.SYSTEM_COPY_CHECK.equals(taskType)) {
                continue;
            }
            typeName = taskType.name();
            poolSize = schd.getTaskPoolSize(typeName);
            sumbitSize = schd.getSumbitedTaskCount(typeName);
            // 判断任务是否可以执行
            boolean isRun = runTask.taskRunnable(taskType.code(), poolSize, sumbitSize);
            if (!isRun) {
                LOG.warn("resource is limit !!! skip {} !!!", typeName);
                continue;
            }
            int retryCount = 3;
            if (TaskType.SYSTEM_CHECK.equals(taskType) || TaskType.SYSTEM_MERGER.equals(taskType) || TaskType.SYSTEM_DELETE.equals(taskType)) {
                retryCount = 0;
            }
            Pair<String, TaskModel> taskPair = TaskStateLifeContral.getCurrentOperationTask(release, typeName, serverId, retryCount);
            if (taskPair == null) {
                LOG.warn("taskType :{} taskName: null is vaild ,skiping !!!", typeName);
                continue;
            }
            currentTaskName = taskPair.getFirst();
            task = TaskStateLifeContral.changeRunTaskModel(taskPair.getSecond(), dataPath);
            // 获取执行策略
            runPattern = runTask.taskRunnPattern(task);
            if (runPattern == null) {
                LOG.warn("TaskRunPattern is null will do it once");
                runPattern = new TaskRunPattern();
                runPattern.setRepeateCount(1);
                runPattern.setSleepTime(1000);
            }
            // 创建任务提交信息
            if (TaskType.SYSTEM_DELETE.equals(taskType)) {
                sumbitTask = createSimpleTask(task, runPattern, currentTaskName, mcf.getServerId(), SystemDeleteJob.class.getCanonicalName(), dataPath);
            }
            if (TaskType.SYSTEM_CHECK.equals(taskType)) {
                sumbitTask = createSimpleTask(task, runPattern, currentTaskName, mcf.getServerId(), SystemCheckJob.class.getCanonicalName(), dataPath);
            }
            if (TaskType.USER_DELETE.equals(taskType)) {
                sumbitTask = createSimpleTask(task, runPattern, currentTaskName, mcf.getServerId(), UserDeleteJob.class.getCanonicalName(), dataPath);
            }
            if (rebalanceFlag && TaskType.SYSTEM_CHECK.equals(taskType)) {
                LOG.warn("rebalance task running !! Skip {} sumbit", taskType.name());
                continue;
            }
            if (sumbitTask == null) {
                LOG.warn("sumbit type:{}, taskName :{}, taskcontent is null", typeName, currentTaskName);
                continue;
            }
            boolean isSumbit = schd.addTask(typeName, sumbitTask);
            LOG.info("sumbit type:{}, taskName :{}, state:{}", typeName, currentTaskName, isSumbit);
            if (!isSumbit) {
                LOG.warn("next cycle will sumbit against type : {}, taskName : {}", typeName, currentTaskName);
                continue;
            }
            // 更新任务状态
            // 更新任务执行的位置
            data.put(typeName, currentTaskName);
        } catch (Exception e) {
            LOG.error("{}", e);
            EmailPool emailPool = EmailPool.getInstance();
            MailWorker.Builder builder = MailWorker.newBuilder(emailPool.getProgramInfo());
            builder.setModel(this.getClass().getSimpleName() + "模块服务发生问题");
            builder.setException(e);
            builder.setMessage("执行任务发生错误");
            builder.setVariable(data.getWrappedMap());
            emailPool.sendEmail(builder);
        }
    }
}
Also used : JobDataMap(org.quartz.JobDataMap) SumbitTaskInterface(com.bonree.brfs.schedulers.task.meta.SumbitTaskInterface) TaskRunPattern(com.bonree.brfs.schedulers.task.model.TaskRunPattern) EmailPool(com.bonree.brfs.email.EmailPool) ManagerContralFactory(com.bonree.brfs.schedulers.ManagerContralFactory) JsonException(com.bonree.brfs.common.utils.JsonUtils.JsonException) RunnableTaskInterface(com.bonree.brfs.schedulers.task.manager.RunnableTaskInterface) TaskType(com.bonree.brfs.common.task.TaskType) MetaTaskManagerInterface(com.bonree.brfs.schedulers.task.manager.MetaTaskManagerInterface) SchedulerManagerInterface(com.bonree.brfs.schedulers.task.manager.SchedulerManagerInterface) TaskModel(com.bonree.brfs.schedulers.task.model.TaskModel)

Example 9 with ManagerContralFactory

use of com.bonree.brfs.schedulers.ManagerContralFactory in project BRFS by zhangnianli.

the class WatchSomeThingJob method operation.

@Override
public void operation(JobExecutionContext context) throws Exception {
    JobDataMap data = context.getJobDetail().getJobDataMap();
    ManagerContralFactory mcf = ManagerContralFactory.getInstance();
    String zkHost = data.getString(JobDataMapConstract.ZOOKEEPER_ADDRESS);
    // 获取client
    CuratorClient curatorClient = null;
    try {
        curatorClient = mcf.getClient();
        String basePath = mcf.getZkPath().getBaseRebalancePath();
        String tasksPath = basePath + Constants.SEPARATOR + Constants.TASKS_NODE;
        boolean isIt = isRecovery(curatorClient, tasksPath);
        // 更新map的值
        StateMap.put(RECOVERY_STATUSE, isIt);
        // 发生副本迁移就删除数据
        if (isIt) {
            WatchDog.abandonFoods();
        }
    } catch (Exception e) {
        LOG.error("{}", e);
        EmailPool emailPool = EmailPool.getInstance();
        MailWorker.Builder builder = MailWorker.newBuilder(emailPool.getProgramInfo());
        builder.setModel(this.getClass().getSimpleName() + "模块服务发生问题");
        builder.setException(e);
        builder.setMessage(mcf.getGroupName() + "(" + mcf.getServerId() + ")服务 看门狗发生错误");
        builder.setVariable(data.getWrappedMap());
        emailPool.sendEmail(builder);
    }
}
Also used : JobDataMap(org.quartz.JobDataMap) CuratorClient(com.bonree.brfs.common.zookeeper.curator.CuratorClient) EmailPool(com.bonree.brfs.email.EmailPool) ManagerContralFactory(com.bonree.brfs.schedulers.ManagerContralFactory) UnableToInterruptJobException(org.quartz.UnableToInterruptJobException) UnsupportedEncodingException(java.io.UnsupportedEncodingException)

Example 10 with ManagerContralFactory

use of com.bonree.brfs.schedulers.ManagerContralFactory in project BRFS by zhangnianli.

the class CheckCycleJob method operation.

@Override
public void operation(JobExecutionContext context) throws Exception {
    LOG.info("cycle check job work !!!");
    JobDataMap data = context.getJobDetail().getJobDataMap();
    int day = data.getInt(JobDataMapConstract.CHECK_TIME_RANGE);
    if (day <= 0) {
        LOG.warn("skip cycle job!! because check time range is 0");
        return;
    }
    ManagerContralFactory mcf = ManagerContralFactory.getInstance();
    MetaTaskManagerInterface release = mcf.getTm();
    StorageRegionManager snm = mcf.getSnm();
    ServiceManager sm = mcf.getSm();
    if (WatchSomeThingJob.getState(WatchSomeThingJob.RECOVERY_STATUSE)) {
        LOG.warn("rebalance task is running !! skip check copy task ,wait next time to check");
        return;
    }
    List services = sm.getServiceListByGroup(mcf.getGroupName());
    if ((services == null) || (services.isEmpty())) {
        LOG.warn("SKIP create {} task, because service is empty", TaskType.SYSTEM_COPY_CHECK);
        return;
    }
    List<StorageRegion> snList = snm.getStorageRegionList();
    if ((snList == null) || (snList.isEmpty())) {
        LOG.warn("SKIP storagename list is null");
        return;
    }
    long currentTime = System.currentTimeMillis();
    long lGraDay = currentTime - currentTime % 86400000L;
    long sGraDay = lGraDay - day * 86400000L;
    List<StorageRegion> needSns = CopyCountCheck.filterSn(snList, services.size());
    if (needSns == null || needSns.isEmpty()) {
        LOG.warn("no storagename need check copy count ! ");
        return;
    }
    Map<String, List<Long>> snTimes = collectionTimes(needSns, sGraDay, lGraDay);
    if (snTimes == null || snTimes.isEmpty()) {
        LOG.warn("{} - {} time, no data to check copy count", TimeUtils.formatTimeStamp(sGraDay), TimeUtils.formatTimeStamp(lGraDay));
        return;
    }
    List<Map<String, Long>> tTimes = converTimes(snTimes);
    for (Map<String, Long> sourceTimes : tTimes) {
        if (sourceTimes == null || sourceTimes.isEmpty()) {
            continue;
        }
        createSingleTask(release, needSns, services, TaskType.SYSTEM_COPY_CHECK, sourceTimes);
    }
}
Also used : JobDataMap(org.quartz.JobDataMap) ManagerContralFactory(com.bonree.brfs.schedulers.ManagerContralFactory) StorageRegion(com.bonree.brfs.duplication.storageregion.StorageRegion) ServiceManager(com.bonree.brfs.common.service.ServiceManager) ArrayList(java.util.ArrayList) List(java.util.List) MetaTaskManagerInterface(com.bonree.brfs.schedulers.task.manager.MetaTaskManagerInterface) HashMap(java.util.HashMap) Map(java.util.Map) JobDataMap(org.quartz.JobDataMap) StorageRegionManager(com.bonree.brfs.duplication.storageregion.StorageRegionManager)

Aggregations

ManagerContralFactory (com.bonree.brfs.schedulers.ManagerContralFactory)18 MetaTaskManagerInterface (com.bonree.brfs.schedulers.task.manager.MetaTaskManagerInterface)9 JobDataMap (org.quartz.JobDataMap)9 StorageRegion (com.bonree.brfs.duplication.storageregion.StorageRegion)8 StorageRegionManager (com.bonree.brfs.duplication.storageregion.StorageRegionManager)7 EmailPool (com.bonree.brfs.email.EmailPool)7 TaskModel (com.bonree.brfs.schedulers.task.model.TaskModel)6 ServiceManager (com.bonree.brfs.common.service.ServiceManager)5 TaskType (com.bonree.brfs.common.task.TaskType)4 CuratorClient (com.bonree.brfs.common.zookeeper.curator.CuratorClient)4 AtomTaskModel (com.bonree.brfs.schedulers.task.model.AtomTaskModel)4 UnableToInterruptJobException (org.quartz.UnableToInterruptJobException)4 TaskResultModel (com.bonree.brfs.schedulers.task.model.TaskResultModel)3 ServerIDManager (com.bonree.brfs.server.identification.ServerIDManager)3 Map (java.util.Map)3 Service (com.bonree.brfs.common.service.Service)2 TaskState (com.bonree.brfs.common.task.TaskState)2 SecondIDParser (com.bonree.brfs.rebalance.route.SecondIDParser)2 RunnableTaskInterface (com.bonree.brfs.schedulers.task.manager.RunnableTaskInterface)2 TaskServerNodeModel (com.bonree.brfs.schedulers.task.model.TaskServerNodeModel)2