当前位置: 首页 > news >正文

Flink源码之Checkpoint执行流程

checkpoint

Checkpoint完整流程如上图所示:

  1. JobMaster的CheckpointCoordinator向所有SourceTask发送RPC触发一次CheckPoint
  2. SourceTask向下游广播CheckpointBarrier
  3. SouceTask完成状态快照后向JobMaster发送快照结果
  4. 非SouceTask在Barrier对齐后完成状态快照向JobMaster发送快照结果
  5. JobMaster保存SubTask快照结果
  6. JobMaster收到所有SubTask快照结果后保存快照信息,想SubTask通知Checkpoint完成

以下对整个流程具体说明。

CheckpointCoordinator

JobMaster将JobGraph转换为ExecutionGraph时,如果开启Checkpoint,会为ExecutionGraph生成一个CheckpointCoordinator

DefaultExecutionGraphBuilder.buildGraph//在此会将JobGraph转换为ExecutionGraphDefaultExecutionGraph::newDefaultExecutionGraph::attachJobGraph //创建ExecutionJobVertexDefaultExecutionTopology.fromExecutionGraph //创建ExecutionTopologyDefaultExecutionGraph::enableCheckpointing //创建CheckpointCoordinatorDefaultExecutionGraph::createCheckpointPlanCalculator//创建DefaultCheckpointPlanCalculatorCheckpointCoordinator::new 

CheckpointCoordinator封装了StateBackend和CheckpointStorage

StateBackend负责管理状态:

  • HashMapStateBackend //内存
  • EmbeddedRocksDBStateBackend //内存+磁盘

CheckpointStorage则是负责存储StateBackend管理的状态:

  • JobManagerCheckpointStorage //checkpoint state放入JobManager内存
  • FileSystemCheckpointStorage //配置state.checkpoints.dir时

在为StreamTask构造SubtaskCheckpointCoordinatorImpl时会调用:

CheckpointStorage::createCheckpointStorage

创建CheckpointStorageAccess用于执行Checkpoint时解析状态存储位置

  • MemoryBackendCheckpointStorageAccess //对应JobManagerCheckpointStorage
  • FsCheckpointStorageAccess //对应FileSystemCheckpointStorage

CheckpointCoordinator在执行状态快照时会调用

CheckpointStorageAccess::resolveCheckpointStorageLocation

生成CheckpointStreamFactory用于生成读写状态数据流

  • MemCheckpointStreamFactory //对应JobManagerCheckpointStorage
  • FsCheckpointStreamFactory //对应FileSystemCheckpointStorage

Checkpoint触发流程

JobMaster状态转换为running后,通过CheckpointCoordinator向SourceTask发送TriggerCheckpoint

JobMaster端触发流程

JobMaster::start  //RPCServer启动
JobMaster::onStart
JobMaster::startJobExecution
JobMaster::startJobMasterServices //获取RM地址后与RM建立连接
JobMaster::startScheduling
SchedulerBase::startScheduling
DefaultScheduler::startSchedulingInternal
SchedulerBase::transitionToRunningDefaultExecutionGraph::transitionToRunning //调用ExecutionGraph监听器通知状态变化CheckpointCoordinatorDeActivator::jobStatusChanges//触发checkpointCheckpointCoordinator::startCheckpointSchedulerCheckpointCoordinator::scheduleTriggerWithDelay //定时不断触发CheckpointCheckpointCoordinator::triggerCheckpointCheckpointCoordinator::startTriggeringCheckpointDefaultCheckpointPlanCalculator::calculateCheckpointPlan//Plan中会隔离出SourceTask作为作为Trigger Checkpoint的入口CheckpointCoordinator::createPendingCheckpointCheckpointCoordinator::triggerCheckpointRequestCheckpointCoordinator::triggerTasks Execution::triggerCheckpoint //向每个SourceTask发送TriggerCheckpoint请求Execution::triggerCheckpointHelperTaskManagerGateway::triggerCheckpoint//向TaskExecutor发RPC

StreamTask端执行流程

SourceTask

SourceTask由JobMaster RPC直接触发,执行时先广播CheckpointBarrier,然后对状态执行异步快照

TaskExecutor::triggerCheckpoint
Task::triggerCheckpointBarrier
AbstractInvokable::triggerCheckpointAsync
SourceStreamTask::triggerCheckpointAsync
StreamTask::triggerCheckpointAsync
StreamTask::triggerCheckpointAsyncInMailbox
StreamTask::performCheckpoint
SubtaskCheckpointCoordinatorImpl::checkpointStateOperatorChain.broadcastEvent //广播CheckpointBarrier
CheckpointStorage::createCheckpointStorage//为JobId创建CheckpointStorageAccess
SubtaskCheckpointCoordinatorImpl::takeSnapshotSync
CheckpointStorageWorkerView::resolveCheckpointStorageLocation//CheckpointStorageAccess创建 CheckpointStreamFactoryOperatorChain::snapshotState //对每个OperatorRegularOperatorChain::buildOperatorSnapshotFuturesRegularOperatorChain::checkpointStreamOperatorAbstractStreamOperator::snapshotStateStreamOperatorStateHandler::snapshotState//调用Operator/Keyed Backend的snapshotStateSnapshotContextSynchronousImpl::newAbstractUdfStreamOperator::snapshotState //调用UDF中snapshotState方法,一般用于更新OperatorStateDefaultOperatorStateBackend::snapshotSnapshotStrategyRunner::snapshotDefaultOperatorStateBackendSnapshotStrategy::syncPrepareResources//深copy operator state,便于后续进行异步快照DefaultOperatorStateBackendSnapshotStrategy::asyncSnapshot//异步快照					  	  CheckpointStateOutputStream::closeAndGetHandleOperatorStreamStateHandle::new //包装元信息及数据StreamStateHandleHeapKeyedStateBackend::snapshotSnapshotStrategyRunner::snapshotHeapSnapshotStrategy::syncPrepareResourcesHeapSnapshotStrategy::asyncSnapshot //采用COWSateTable异步快照CheckpointStateOutputStream::closeAndGetHandleKeyGroupsStateHandle::new //包装KeyGroup及数据StreamStateHandle
SubtaskCheckpointCoordinatorImpl::finishAndReportAsync //向JobMaster发送checkpoint的结果AsyncCheckpointRunnable::new AsyncCheckpointRunnable::runAsyncCheckpointRunnable::finalizeNonFinishedSnapshotsOperatorSnapshotFinalizer::new //等待TaskSnapshot状态信息序列化完成AsyncCheckpointRunnable::reportCompletedSnapshotStatesTaskStateManagerImpl::reportTaskStateSnapshotsRpcCheckpointResponder::acknowledgeCheckpoint//向JobMaster发送Ack,带上State信息
非SourceTask

在StreamTask启动后调用StreamTask::processInput不断读取数据进行处理, 非SourceTask在收到上游的CheckpointBarrier对齐后触发Checkpoint,

StreamTask::processInput
StreamOneInputProcessor::processInput
StreamTaskNetworkInput::emitNext(StreamTaskNetworkOutput)
AbstractStreamTaskNetworkInput::emitNext //循环不断从buffer中读取StreamElement
处理CheckpointedInputGate::pollNextCheckpointedInputGate::handleEventSingleCheckpointBarrierHandler::processBarrierSingleCheckpointBarrierHandler::markCheckpointAlignedAndTransformStateWaitingForFirstBarrier::barrierReceivedAbstractAlignedBarrierHandlerState::barrierReceivedSingleCheckpointBarrierHandler.ControllerImpl::allBarriersReceived//判断对齐AbstractAlignedBarrierHandlerState::triggerGlobalCheckpointSingleCheckpointBarrierHandler.ControllerImpl::triggerGlobalCheckpointSingleCheckpointBarrierHandler::triggerCheckpointCheckpointBarrierHandler::notifyCheckpoint //触发StreamTask CheckpointStreamTask::triggerCheckpointOnBarrierStreamTask::performCheckpoint //后续调用过程与SourceTask一样SubtaskCheckpointCoordinatorImpl::checkpointState   		

根据调用栈看出,非SourceStreamTask执行Checkpoint只是触发时机不同,SourceTask由JobMaster RPC定时不断触发,非SourceTask则是在上游的CheckpointBarrier对齐后触发Checkpoint,最终执行逻辑都是将当前算子的信息写入CheckpointStorage后向JobMaster发送确认信息。

StreamTask向JobMaster ACK信息中包含状态元信息及StreamStateHandle,根据状态存储位置分为:

  • ByteStreamStateHandle //对应JobManagerCheckpointStorage,将状态序列化为byte[]发送给JobMaster
  • FileStateHandle //对应FileSystemCheckpointStorage,将状态写入文件系统后将文件路径发送给JobMaster

JobMaster端完成流程

JobMaster收到StreamTask的acknowledgeCheckpoint后:

JobMaster::acknowledgeCheckpoint
SchedulerBase::acknowledgeCheckpoint
ExecutionGraphHandler::acknowledgeCheckpoint
CheckpointCoordinator::receiveAcknowledgeMessagePendingCheckpoint::acknowledgeTask //某一个Task的确认PendingCheckpoint::updateOperatorState//更新SubTask状态信息CheckpointCoordinator::completePendingCheckpoint//所有Task Ack后PendingCheckpoint::finalizeCheckpointCheckpoints.storeCheckpointMetadata//保存CheckpointMetadataCompletedCheckpoint::newCheckpointCoordinator::sendAcknowledgeMessages//向Task通知Checkpoint完成消息ExecutionVertex::notifyCheckpointCompleteTaskManagerGateway.notifyCheckpointComplete

JobMaster收到所有StreamTask的Checkpoint状态信息后,标志一次Checkpoint完成,这时会通知StreamTask CheckPoint完成消息,便于SubTask监听Checkpoint完成后做后续动作。

http://www.lryc.cn/news/135853.html

相关文章:

  • 【工具使用】Git的使用
  • 无涯教程-PHP Installation on Windows NT/2000/XP with IIS函数
  • EureKa快速入门
  • Sectigo EV代码签名申请步骤
  • 生信学院|08月25日《SOLIDWORKS PDM帮助企业对设计数据版本的管理应用》
  • vue页面转pdf后分页时文字被横向割裂
  • 数据结构——队列(C语言)
  • WGS84地球坐标系,GCJ02火星坐标系,BD09百度坐标系简介与转换 资料收集
  • 【面试题】前端面试复习6---性能优化
  • 隧道HTTP具备的条件
  • 部署FTP服务(二)
  • 缓存的变更(JVM本地缓存->Redis分布式缓存)
  • springMVC Unix 文件参数变更漏洞修复
  • 【LeetCode】494.目标和
  • KaiwuDB 荣获哈佛商业评论 2023“高能韧性团队奖”
  • 删除ubuntu开始菜单中的图标
  • 信息系统项目管理基础知识学习笔记 - IT 治理基础 - IT治理的驱动因素
  • 8月21-22日上课内容 第一章 MySQL数据库初始
  • 等级查询发布助手
  • 手搭手入门MyBatis-Plus
  • AI 绘画Stable Diffusion 研究(十一)sd图生图功能详解-美女换装
  • Servlet+JDBC实战开发书店项目讲解第14讲:订单管理功能
  • 基于Linux操作系统中的shell脚本
  • 8.22笔记
  • 【以太网通信】RS232 串口转以太网
  • 分享两道Java面试的算法上机题目(后续会持续补充更多)
  • 如何使用CSS实现一个平滑过渡效果?
  • 网络常见设备
  • 数据结构与算法:通往编程高地的必修课(文末送书)
  • python小脚本——批量将PDF文件转换成图片