ruanwenjun commented on code in PR #15528: URL: https://github.com/apache/dolphinscheduler/pull/15528#discussion_r1474078555
########## dolphinscheduler-master/src/main/java/org/apache/dolphinscheduler/server/master/runner/taskgroup/TaskGroupCoordinator.java: ########## @@ -0,0 +1,451 @@ +/* + * Licensed to the Apache Software Foundation (ASF) under one or more + * contributor license agreements. See the NOTICE file distributed with + * this work for additional information regarding copyright ownership. + * The ASF licenses this file to You under the Apache License, Version 2.0 + * (the "License"); you may not use this file except in compliance with + * the License. You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +package org.apache.dolphinscheduler.server.master.runner.taskgroup; + +import org.apache.dolphinscheduler.common.constants.Constants; +import org.apache.dolphinscheduler.common.enums.Flag; +import org.apache.dolphinscheduler.common.enums.TaskGroupQueueStatus; +import org.apache.dolphinscheduler.common.enums.WorkflowExecutionStatus; +import org.apache.dolphinscheduler.common.lifecycle.ServerLifeCycleManager; +import org.apache.dolphinscheduler.common.thread.BaseDaemonThread; +import org.apache.dolphinscheduler.common.thread.ThreadUtils; +import org.apache.dolphinscheduler.dao.entity.ProcessInstance; +import org.apache.dolphinscheduler.dao.entity.TaskGroup; +import org.apache.dolphinscheduler.dao.entity.TaskGroupQueue; +import org.apache.dolphinscheduler.dao.entity.TaskInstance; +import org.apache.dolphinscheduler.dao.repository.ProcessInstanceDao; +import org.apache.dolphinscheduler.dao.repository.TaskGroupDao; +import org.apache.dolphinscheduler.dao.repository.TaskGroupQueueDao; +import org.apache.dolphinscheduler.dao.repository.TaskInstanceDao; +import org.apache.dolphinscheduler.extract.base.client.SingletonJdkDynamicRpcClientProxyFactory; +import org.apache.dolphinscheduler.extract.master.IWorkflowInstanceService; +import org.apache.dolphinscheduler.extract.master.transportor.TaskInstanceWakeupRequest; +import org.apache.dolphinscheduler.extract.master.transportor.TaskInstanceWakeupResponse; +import org.apache.dolphinscheduler.plugin.task.api.enums.TaskExecutionStatus; +import org.apache.dolphinscheduler.plugin.task.api.utils.LogUtils; +import org.apache.dolphinscheduler.registry.api.RegistryClient; +import org.apache.dolphinscheduler.registry.api.enums.RegistryNodeType; + +import org.apache.commons.collections4.CollectionUtils; +import org.apache.commons.lang3.time.StopWatch; + +import java.util.Date; +import java.util.List; +import java.util.Map; +import java.util.function.Function; +import java.util.stream.Collectors; + +import lombok.extern.slf4j.Slf4j; + +import org.springframework.beans.factory.annotation.Autowired; +import org.springframework.stereotype.Component; + +/** + * The TaskGroupCoordinator use to manage the task group slot. The task group slot is used to limit the number of {@link TaskInstance} that can be run at the same time. + * <p> + * The {@link TaskGroupQueue} is used to represent the task group slot. When a {@link TaskGroupQueue} which inQueue is YES means the {@link TaskGroupQueue} is using by a {@link TaskInstance}. + * <p> + * When the {@link TaskInstance} need to use task group, we should use @{@link TaskGroupCoordinator#acquireTaskGroupSlot(TaskInstance)} to acquire the task group slot, + * this method doesn't block should always acquire successfully, and you should directly stop dispatch the task instance. + * When the task group slot is available, the TaskGroupCoordinator will wake up the waiting {@link TaskInstance} to dispatch. + * <pre> + * if(needAcquireTaskGroupSlot(taskInstance)) { + * taskGroupCoordinator.acquireTaskGroupSlot(taskInstance); + * return; + * } + * </pre> + * <p> + * When the {@link TaskInstance} is finished, we should use @{@link TaskGroupCoordinator#releaseTaskGroupSlot(TaskInstance)} to release the task group slot. + * <pre> + * if(needToReleaseTaskGroupSlot(taskInstance)) { + * taskGroupCoordinator.releaseTaskGroupSlot(taskInstance); + * } + * </pre> + */ +@Slf4j +@Component +public class TaskGroupCoordinator extends BaseDaemonThread { + + @Autowired + private RegistryClient registryClient; + + @Autowired + private TaskGroupDao taskGroupDao; + + @Autowired + private TaskGroupQueueDao taskGroupQueueDao; + + @Autowired + private TaskInstanceDao taskInstanceDao; + + @Autowired + private ProcessInstanceDao processInstanceDao; + + public TaskGroupCoordinator() { + super("TaskGroupCoordinator"); + } + + @Override + public synchronized void start() { + log.info("TaskGroupCoordinator starting..."); + super.start(); + log.info("TaskGroupCoordinator started..."); + } + + @Override + public void run() { + while (!ServerLifeCycleManager.isStopped()) { + try { + if (!ServerLifeCycleManager.isRunning()) { + continue; + } + try { + registryClient.getLock(RegistryNodeType.MASTER_TASK_GROUP_COORDINATOR_LOCK.getRegistryPath()); + StopWatch taskGroupCoordinatorRoundTimeCost = StopWatch.createStarted(); + + amendTaskGroupUseSize(); + amendTaskGroupQueueStatus(); + dealWithForceStartTaskGroupQueue(); + dealWithWaitingTaskGroupQueue(); + + taskGroupCoordinatorRoundTimeCost.stop(); + log.info("TaskGroupCoordinator round time cost: {}/ms", + taskGroupCoordinatorRoundTimeCost.getTime()); + } finally { + registryClient.releaseLock(RegistryNodeType.MASTER_TASK_GROUP_COORDINATOR_LOCK.getRegistryPath()); + } + } catch (Throwable e) { + log.error("TaskGroupCoordinator error", e); + } finally { + // sleep 5s + ThreadUtils.sleep(Constants.SLEEP_TIME_MILLIS * 5); + } + } + } + + /** + * Make sure the TaskGroup useSize is equal to the TaskGroupQueue which status is {@link TaskGroupQueueStatus#ACQUIRE_SUCCESS} and forceStart is {@link org.apache.dolphinscheduler.common.enums.Flag#NO}. + */ + private void amendTaskGroupUseSize() { + // The TaskGroup useSize should equal to the TaskGroupQueue which inQueue is YES and forceStart is NO + List<TaskGroup> taskGroups = taskGroupDao.queryAllTaskGroups(); + if (CollectionUtils.isEmpty(taskGroups)) { + return; + } + for (TaskGroup taskGroup : taskGroups) { + List<TaskGroupQueue> taskGroupQueues = + taskGroupQueueDao.queryAcquiredTaskGroupQueueByGroupId(taskGroup.getId()); + int actualUseSize = taskGroupQueues.size(); + if (taskGroup.getUseSize() == actualUseSize) { + continue; + } + log.warn("The TaskGroup: {} useSize is {}, but the actual use size is {}, will amend it", + taskGroup.getName(), + taskGroup.getUseSize(), actualUseSize); + taskGroup.setUseSize(actualUseSize); + taskGroupDao.updateById(taskGroup); + } + } + + /** + * Make sure the TaskGroupQueue status is {@link TaskGroupQueueStatus#RELEASE} when the related {@link TaskInstance} is not exist or status is finished. + */ + private void amendTaskGroupQueueStatus() { + List<TaskGroupQueue> taskGroupQueues = taskGroupQueueDao.queryAllInQueueTaskGroupQueue(); + List<Integer> taskInstanceIds = taskGroupQueues.stream() + .map(TaskGroupQueue::getTaskId) + .collect(Collectors.toList()); + Map<Integer, TaskInstance> taskInstanceMap = taskInstanceDao.queryByIds(taskInstanceIds) + .stream() + .collect(Collectors.toMap(TaskInstance::getId, Function.identity())); + + for (TaskGroupQueue taskGroupQueue : taskGroupQueues) { + int taskId = taskGroupQueue.getTaskId(); + TaskInstance taskInstance = taskInstanceMap.get(taskId); + + if (taskInstance == null) { + log.warn("The TaskInstance: {} is not exist, will release the TaskGroupQueue: {}", taskId, + taskGroupQueue); + releaseTaskGroupQueueSlot(taskGroupQueue); + continue; + } + + if (taskInstance.getState().isFinished()) { + log.warn("The TaskInstance: {} state: {} finished, will release the TaskGroupQueue: {}", + taskInstance.getName(), taskInstance.getState(), taskGroupQueue); + releaseTaskGroupQueueSlot(taskGroupQueue); + continue; + } + } + } + + private void dealWithForceStartTaskGroupQueue() { + // Find the force start task group queue(Which is inQueue and forceStart is YES) + // Notify the related waiting task instance + // Set the taskGroupQueue status to RELEASE and remove it from queue + List<TaskGroupQueue> taskGroupQueues = taskGroupQueueDao.queryAllInQueueTaskGroupQueue() + .stream() + .filter(taskGroupQueue -> Flag.YES.getCode() == taskGroupQueue.getForceStart()) + .collect(Collectors.toList()); + for (TaskGroupQueue taskGroupQueue : taskGroupQueues) { + try { + LogUtils.setTaskInstanceIdMDC(taskGroupQueue.getTaskId()); + // notify the waiting task instance + // We notify first, it notify failed, the taskGroupQueue will be in queue, and then we will retry it + // next time. + notifyWaitingTaskInstance(taskGroupQueue); + log.info("Notify the ForceStart waiting TaskInstance: {} for taskGroupQueue: {} success", + taskGroupQueue.getTaskName(), + taskGroupQueue.getId()); + + taskGroupQueue.setInQueue(Flag.NO.getCode()); + taskGroupQueue.setStatus(TaskGroupQueueStatus.RELEASE); + taskGroupQueue.setUpdateTime(new Date()); + taskGroupQueueDao.updateById(taskGroupQueue); + log.info("Release the force start TaskGroupQueue {}", taskGroupQueue); + } catch (UnsupportedOperationException unsupportedOperationException) { + releaseTaskGroupQueueSlot(taskGroupQueue); + log.info( + "Notify the ForceStart TaskInstance: {} for taskGroupQueue: {} failed, will release the taskGroupQueue", + taskGroupQueue.getTaskName(), taskGroupQueue.getId(), unsupportedOperationException); + } catch (Throwable throwable) { + log.info("Notify the force start TaskGroupQueue {} failed", taskGroupQueue, throwable); + } finally { + LogUtils.removeTaskInstanceIdMDC(); + } + } + } + + private void dealWithWaitingTaskGroupQueue() { + // Find the TaskGroup which usage < maxSize. + // Find the highest priority inQueue task group queue(Which is inQueue and status is Waiting and force start is + // NO) belong to the + // task group. + List<TaskGroup> taskGroups = taskGroupDao.queryAvailableTaskGroups(); + if (CollectionUtils.isEmpty(taskGroups)) { + log.debug("There is no available task group"); + return; + } + for (TaskGroup taskGroup : taskGroups) { + int availableSize = taskGroup.getGroupSize() - taskGroup.getUseSize(); + if (availableSize <= 0) { + log.info("TaskGroup {} is full, available size is {}", taskGroup, availableSize); + continue; + } + List<TaskGroupQueue> taskGroupQueues = + taskGroupQueueDao.queryAllInQueueTaskGroupQueueByGroupId(taskGroup.getId()) + .stream() + .filter(taskGroupQueue -> Flag.NO.getCode() == taskGroupQueue.getForceStart()) + .filter(taskGroupQueue -> TaskGroupQueueStatus.WAIT_QUEUE == taskGroupQueue.getStatus()) + .limit(availableSize) + .collect(Collectors.toList()); + if (CollectionUtils.isEmpty(taskGroupQueues)) { + log.debug("There is no waiting task group queue for task group {}", taskGroup.getName()); + continue; + } + for (TaskGroupQueue taskGroupQueue : taskGroupQueues) { + try { + LogUtils.setTaskInstanceIdMDC(taskGroupQueue.getTaskId()); + // Reduce the taskGroupSize + boolean acquireResult = taskGroupDao.acquireTaskGroupSlot(taskGroup.getId()); + if (!acquireResult) { + log.error("Failed to acquire task group slot for task group {}", taskGroup); + continue; + } + // Notify the waiting task instance + // We notify first, it notify failed, the taskGroupQueue will be in queue, and then we will retry it + // next time. + notifyWaitingTaskInstance(taskGroupQueue); + + // Set the taskGroupQueue status to RUNNING and remove from queue + taskGroupQueue.setInQueue(Flag.YES.getCode()); + taskGroupQueue.setStatus(TaskGroupQueueStatus.ACQUIRE_SUCCESS); + taskGroupQueue.setUpdateTime(new Date()); + taskGroupQueueDao.updateById(taskGroupQueue); Review Comment: Good suggestion, I will use another PR to move these kind of code in the whole project into function. -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: [email protected] For queries about this service, please contact Infrastructure at: [email protected]
