我用夸克网盘分享了「Spark内核设计的艺术：架构设计与实现.epub」，点击链接即可保存。打开「夸克APP」在线查看，支持多种文档格式转换。

内容简介

多位专家联袂推荐，360大数据专家撰写，基于Spark 2.1.0剖析架构与实现精髓。细化到方法级，提炼出多个流程图，立体呈现架构、环境、调度、存储、计算、部署、API七大核心设计。本书一共有10章内容，主要包括以下部分。 准备部分（第1～2章）：简单介绍了Spark的环境搭建和基本原理。本部分通过详尽的描述，有效降低了读者进入Spark世界的门槛，同时能对Spark背景知识及整体设计有宏观的认识。 基础部分（第3～5章）：介绍Spark的基础设施（包括配置、RPC、度量等）、SparkContext的初始化、Spark执行所需要的环境等内容。经过此部分的学习，将能够对RPC框架的设计、执行环境的功能有深入的理解，这也是对核心内容了解的前提。 核心部分（第6～9章）：为Spark最核心的部分，包括存储体系、调度系统、计算引擎、部署模式等。通过本部分的学习，读者将充分了解Spark的数据处理体系细节，能够对Spark核心功能进行扩展、性能优化以及对线上问题进行精准排查。 API部分（第10章）：这部分主要对Spark的新老API进行对比，对新API进行简单介绍。

作者简介

耿嘉安 10余年IT行业相关经验。先后就职于阿里巴巴、艺龙、360，专注于开源和大数据领域。在大量的工作实践中，对J2EE、JVM、Tomcat、Spring、Hadoop、Spark、MySQL、Redis都有深入研究，尤其喜欢剖析开源项目的源码实现。早期从事J2EE企业级应用开发，对Java相关技术有独到见解。著有《深入理解Spark：核心思想与源码分析》一书。

网友热评

阿里阿里巴巴: 内容还行，有一定深度，但是书的排版、编辑不太好，看起来就感觉像是在堆代码，可以精简一些 书忆江南: 这本书不是入门书，建议有了《Spark权威指南》、《Scala实用指南》、设计模式、Java等基础，对Spark常用API和表层原理有所了解后再看，适合大数据平台开发等需要定制修改最底层组件源码的人看，目前唯一一本讲spark2源码的书，基础不够确实看的会比较累 当然这本书个人觉得的一个缺点是，只是罗列了代码然后稍微讲了下，没有把spark某个操作的过程总结成有哪些步骤，比如SparkContext初始化经总结总共是哪几步，只是在讲各块代码的细节，彼此的逻辑步骤关系没有最后总结一下串联起来，像是割裂的 privet: 很多内容都是对一段代码片段进行逐一的逻辑讲解。我觉得应该更多地增加些作者的思考。不然的话我觉得还不如直接阅读源码。 一条狗: 跟着书籍看源码，还是有帮助的。作者的技术功底还是可以的。适合想看源码的同学学习

图书目录

目录Contents 本书赞誉 前言 第1章　环境准备········································1 1.1　运行环境准备···········································2 1.1.1　安装JDK·········································2 1.1.2　安装Scala········································2 1.1.3　安装Spark·······································3 1.2　Spark初体验···································4 1.2.1　运行spark-shell·······························4 1.2.2　执行wordcount······························5 1.2.3　剖析spark-shell·······························9 1.3　阅读环境准备·········································14 1.3.1　安装SBT·······································15 1.3.2　安装Git·········································15 1.3.3　安装EclipseScalaIDE插件········15 1.4　Spark源码编译与调试·························17 1.5　小结···························23 第2章　设计理念与基本架构···············24 2.1　初识Spark··································25 2.1.1　HadoopMRv1的局限···················25 2.1.2　Spark的特点·································26 2.1.3Spark使用场景·····························28 2.2　Spark基础知识······································29 2.3　Spark基本设计思想·····························31 2.3.1　Spark模块设计·····························32 2.3.2　Spark模型设计·····························34 2.4　Spark基本架构···································36 2.5　小结·································38 第3章　Spark基础设施·························39 3.1　Spark配置········································40 3.1.1　系统属性中的配置·······················40 3.1.2　使用SparkConf配置的API·········41 3.1.3　克隆SparkConf配置····················42 3.2　Spark内置RPC框架····························42 3.2.1　RPC配置TransportConf··············45 3.2.2　RPC客户端工厂Transport-ClientFactory·······················47 3.2.3　RPC服务端TransportServer········53 3.2.4　管道初始化···································56 3.2.5　TransportChannelHandler详解·····57 3.2.6　服务端RpcHandler详解··············63 3.2.7　服务端引导程序Transport-ServerBootstrap·····················68 3.2.8　客户端TransportClient详解········71 3.3　事件总线····································78 3.3.1　ListenerBus的继承体系···············79 3.3.2　SparkListenerBus详解··················80 3.3.3　LiveListenerBus详解····················83 3.4　度量系统···········································87 3.4.1　Source继承体系···························87 3.4.2　Sink继承体系·······························89 3.5　小结·········································92 第4章　SparkContext的初始化·········93 4.1　SparkContext概述·································94 4.2　创建Spark环境·····································97 4.3　SparkUI的实现····································100 4.3.1　SparkUI概述·······························100 4.3.2　WebUI框架体系·························102 4.3.3　创建SparkUI·······························107 4.4　创建心跳接收器··································111 4.5　创建和启动调度系统··························112 4.6　初始化块管理器BlockManager·······114 4.7　启动度量系统·······························114 4.8　创建事件日志监听器··························115 4.9　创建和启动ExecutorAllocation-Manager··························116 4.10　ContextCleaner的创建与启动········120 4.10.1　创建ContextCleaner·················120 4.10.2　启动ContextCleaner·················120 4.11　额外的SparkListener与启动事件总线··························122 4.12　Spark环境更新··································123 4.13　SparkContext初始化的收尾···········127 4.14　SparkContext提供的常用方法·······128 4.15　SparkContext的伴生对象················130 4.16　小结····································131 第5章　Spark执行环境························132 5.1　SparkEnv概述·································133 5.2　安全管理器SecurityManager············133 5.3　RPC环境·········································135 5.3.1　RPC端点RpcEndpoint···············136 5.3.2　RPC端点引用RpcEndpointRef···139 5.3.3　创建传输上下文TransportConf···142 5.3.4　消息调度器Dispatcher···············142 5.3.5　创建传输上下文Transport-Context·························154 5.3.6　创建传输客户端工厂Transport-ClientFactory····················159 5.3.7　创建TransportServer···················160 5.3.8　客户端请求发送·························162 5.3.9　NettyRpcEnv中的常用方法·······173 5.4　序列化管理器SerializerManager·····175 5.5　广播管理器BroadcastManager·········178 5.6　map任务输出跟踪器··························185 5.6.1　MapOutputTracker的实现··········187 5.6.2　MapOutputTrackerMaster的实现原理·······················191 5.7　构建存储体系·······································199 5.8　创建度量系统·······································201 5.8.1　MetricsCon?g详解·····················203 5.8.2　MetricsSystem中的常用方法····207 5.8.3　启动MetricsSystem····················209 5.9　输出提交协调器··································211 5.9.1　OutputCommitCoordinator-Endpoint的实现··················211 5.9.2　OutputCommitCoordinator的实现··························212 5.9.3　OutputCommitCoordinator的工作原理························216 5.10　创建SparkEnv····································217 5.11　小结·····································217 第6章　存储体系·····································219 6.1　存储体系概述·······································220 6.1.1　存储体系架构·····························220 6.1.2　基本概念·····································222 6.2　Block信息管理器································227 6.2.1　Block锁的基本概念···················227 6.2.2　Block锁的实现···························229 6.3　磁盘Block管理器······························234 6.3.1　本地目录结构·····························234 6.3.2　DiskBlockManager提供的方法···························236 6.4　磁盘存储DiskStore·····························239 6.5　内存管理器·····································242 6.5.1　内存池模型·································243 6.5.2　StorageMemoryPool详解···········244 6.5.3　MemoryManager模型················247 6.5.4　Uni?edMemoryManager详解····250 6.6　内存存储MemoryStore······················252 6.6.1　MemoryStore的内存模型··········253 6.6.2　MemoryStore提供的方法··········255 6.7　块管理器BlockManager····················265 6.7.1　BlockManager的初始化·············265 6.7.2　BlockManager提供的方法·········266 6.8　BlockManagerMaster对Block-Manager的管理·················285 6.8.1　BlockManagerMaster的职责······285 6.8.2　BlockManagerMasterEndpoint详解·································286 6.8.3　BlockManagerSlaveEndpoint详解·····························289 6.9　Block传输服务····································290 6.9.1　初始化NettyBlockTransfer-Service···························291 6.9.2　NettyBlockRpcServer详解·········292 6.9.3　Shuf?e客户端·····························296 6.10　DiskBlockObjectWriter详解···········305 6.11　小结·······································308 第7章　调度系统·····································309 7.1　调度系统概述·······································310 7.2　RDD详解·····································312 7.2.1　为什么需要RDD························312 7.2.2　RDD实现的初次分析················313 7.2.3　RDD依赖····································316 7.2.4　分区计算器Partitioner················318 7.2.5　RDDInfo······································320 7.3　Stage详解········································321 7.3.1　ResultStage的实现·····················322 7.3.2　Shuf?eMapStage的实现·············323 7.3.3　StageInfo······································324 7.4　面向DAG的调度器DAGScheduler···326 7.4.1　JobListener与JobWaiter·············326 7.4.2　ActiveJob详解····························328 7.4.3　DAGSchedulerEventProcessLoop的简要介绍·······················328 7.4.4　DAGScheduler的组成················329 7.4.5　DAGScheduler提供的常用方法···330 7.4.6　DAGScheduler与Job的提交····334 7.4.7　构建Stage····································337 7.4.8　提交ResultStage························341 7.4.9　提交还未计算的Task·················343 7.4.10　DAGScheduler的调度流程······347 7.4.11　Task执行结果的处理··············348 7.5　调度池Pool······································351 7.5.1　调度算法·······························352 7.5.2　Pool的实现·································354 7.5.3　调度池构建器·····························357 7.6　任务集合管理器TaskSetManager···363 7.6.1　Task集合·····································363 7.6.2　TaskSetManager的成员属性······364 7.6.3　调度池与推断执行·····················366 7.6.4　Task本地性·································370 7.6.5　TaskSetManager的常用方法······373 7.7　运行器后端接口LauncherBackend···383 7.7.1　BackendConnection的实现········384 7.7.2　LauncherBackend的实现···········386 7.8　调度后端接口SchedulerBackend····389 7.8.1　SchedulerBackend的定义··········389 7.8.2　LocalSchedulerBackend的实现分析································390 7.9　任务结果获取器TaskResultGetter···394 7.9.1　处理成功的Task·························394 7.9.2　处理失败的Task·························396 7.10　任务调度器TaskScheduler··············397 7.10.1　TaskSchedulerImpl的属性·····397 7.10.2　TaskSchedulerImpl的初始化···399 7.10.3　TaskSchedulerImpl的启动·····399 7.10.4　TaskSchedulerImpl与Task的提交·······················400 7.10.5　TaskSchedulerImpl与资源分配···························402 7.10.6　TaskSchedulerImpl的调度流程······························405 7.10.7　TaskSchedulerImpl对执行结果的处理·····························406 7.10.8　TaskSchedulerImpl的常用方法···409 7.11　小结·······································412 第8章　计算引擎·····································413 8.1　计算引擎概述·······································414 8.2　内存管理器与执行内存·····················417 8.2.1　ExecutionMemoryPool详解·······417 8.2.2　MemoryManager模型与执行内存··························420 8.2.3　Uni?edMemoryManager与执行内存·······················421 8.3　内存管理器与Tungsten·····················423 8.3.1　MemoryBlock详解·····················423 8.3.2　MemoryManager模型与Tungsten···························425 8.3.3　Tungsten的内存分配器··············425 8.4　任务内存管理器··································431 8.4.1　TaskMemoryManager详解·········431 8.4.2　内存消费者·······················439 8.4.3　执行内存整体架构·····················441 8.5　Task详解······································443 8.5.1　任务上下文TaskContext············443 8.5.2　Task的定义·································446 8.5.3　Shuf?eMapTask的实现··············449 8.5.4　ResultTask的实现·······················450 8.6　IndexShuf?eBlockResolver详解······451 8.7　采样与估算···········································455 8.7.1　SizeTracker的实现分析·············455 8.7.2　SizeTracker的工作原理·············457 8.8　特质WritablePartitionedPair-Collection······················458 8.9　AppendOnlyMap的实现分析···········460 8.9.1　AppendOnlyMap的容量增长····461 8.9.2　AppendOnlyMap的数据更新····462 8.9.3　AppendOnlyMap的缓存聚合算法·····························464 8.9.4　AppendOnlyMap的内置排序····466 8.9.5　AppendOnlyMap的扩展············467 8.10　PartitionedPairBuffer的实现分析···469 8.10.1　PartitionedPairBuffer的容量增长······················469 8.10.2　PartitionedPairBuffer的插入···470 8.10.3　PartitionedPairBuffer的迭代器···471 8.11　外部排序器·········································472 8.11.1　ExternalSorter详解·················473 8.11.2　Shuf?eExternalSorter详解······487 8.12　Shuf?e管理器····································490 8.12.1　Shuf?eWriter详解··················491 8.12.2　Shuf?eBlockFetcherIterator详解······························502 8.12.3　BlockStoreShuf?eReader详解···510 8.12.4　SortShuf?eManager详解········513 8.13　map端与reduce端的Shuf?e组合······························516 8.14　小结·········································519 第9章　部署模式········································520 9.1　心跳接收器HeartbeatReceiver·········521 9.2　Executor的实现分析··························527 9.2.1　Executor的心跳报告··················528 9.2.2　运行Task·····································530 9.3　local部署模式······································535 9.4　持久化引擎PersistenceEngine··········537 9.4.1　基于文件系统的持久化引擎·····539 9.4.2　基于ZooKeeper的持久化引擎···541 9.5　领导选举代理·······································542 9.6　Master详解···········································546 9.6.1　启动Master·································549 9.6.2　检查Worker超时························553 9.6.3　被选举为领导时的处理·············554 9.6.4　一级资源调度·····························558 9.6.5　注册Worker·································568 9.6.6　更新Worker的最新状态············570 9.6.7　处理Worker的心跳····················570 9.6.8　注册Application··························571 9.6.9　处理Executor的申请·················573 9.6.10　处理Executor的状态变化·······573 9.6.11　Master的常用方法···················574 9.7　Worker详解································578 9.7.1　启动Worker·································581 9.7.2　向Master注册Worker···············584 9.7.3　向Master发送心跳····················589 9.7.4　Worker与领导选举·····················591 9.7.5　运行Driver··································593 9.7.6　运行Executor······························594 9.7.7　处理Executor的状态变化·········599 9.8　StandaloneAppClient实现·················600 9.8.1　ClientEndpoint的实现分析········601 9.8.2　StandaloneAppClient的实现分析······························606 9.9　StandaloneSchedulerBackend的实现分析························607 9.9.1　StandaloneSchedulerBackend的属性····························607 9.9.2　DriverEndpoint的实现分析·······609 9.9.3　StandaloneSchedulerBackend的启动··························614 9.9.4　StandaloneSchedulerBackend的停止·························617 9.9.5　StandaloneSchedulerBackend与资源分配················618 9.10　CoarseGrainedExecutorBackend详解····························619 9.10.1　CoarseGrainedExecutorBackend进程··························620 9.10.2　CoarseGrainedExecutorBackend的功能分析·························622 9.11　local-cluster部署模式·······················625 9.11.1　启动本地集群····························625 9.11.2　local-cluster部署模式的启动过程·································627 9.11.3　local-cluster部署模式下Executor的分配过程·················628 9.11.4　local-cluster部署模式下的任务提交执行过程····························629 9.12　Standalone部署模式·························631 9.12.1　Standalone部署模式的启动过程························632 9.12.2　Standalone部署模式下Executor的分配过程················634 9.12.3　Standalone部署模式的资源回收·····························635 9.12.4　Standalone部署模式的容错机制······························636 9.13　其他部署方案·····································639 9.13.1　YARN·········································639 9.13.2　Mesos·········································644 9.14　小结·······································646 第10章　SparkAPI································647 10.1　基本概念·····································648 10.2　数据源DataSource····························650 10.2.1　DataSourceRegister详解··········650 10.2.2　DataSource详解························651 10.3　检查点的实现···································655 10.3.1　CheckpointRDD的实现············655 10.3.2　RDDCheckpointData的实现····660 10.3.3　ReliableRDDCheckpointData的实现························662 10.4　RDD的再次分析·······························663 10.4.1　转换API····································663 10.4.2　动作API····································665 10.4.3　检查点API的实现分析···········667 10.4.4　迭代计算···································669 10.5　数据集合Dataset·······························671 10.6　DataFrameReader详解·····················673 10.7　SparkSession详解·····························676 10.7.1　SparkSession的构建器Builder···676 10.7.2　SparkSession的API·················679 10.8　wordcount例子·································679 10.8.1　Job准备阶段·····························680 10.8.2　Job的提交与调度·····················685 10.9　小结········································689 附录···········································690