Kylin 在 eBay 的成长历程与实践

{"type":"doc","content":[{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"在演讲之前，我想首先代表 eBay 祝贺 Apache Kylin 成立 5 周年。Apache Kylin 是出色的 OLAP 产品，希望 Apache Kylin 社区在未来能够带来更多了不起的功能，并对行业产生更大的影响。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"今天的议程包括："}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"bulletedlist","content":[{"type":"listitem","attrs":{"listStyle":null},"content":[{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"介绍 eBay 的开源文化"}]}]},{"type":"listitem","attrs":{"listStyle":null},"content":[{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"Apache Kylin 在 eBay 中的发展历程"}]}]},{"type":"listitem","attrs":{"listStyle":null},"content":[{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"Kylin 在 eBay 中的应用"}]}]},{"type":"listitem","attrs":{"listStyle":null},"content":[{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"eBay 未来的计划"}]}]}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"heading","attrs":{"align":null,"level":2},"content":[{"type":"text","text":"01 eBay 的开源文化"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"eBay 是一家拥有 30 年历史的电子商务公司，为全球的买卖双方提供线上市场。我们的业务发展速度极快，通过进行巨大的技术转型和发展，后端技术团队在促进 eBay 业务实现方面发挥着非常重要的作用。在过去的十年中，核心技术团队推动着开源文化成为了我们的 DNA，给我们带来了巨大的改变和丰硕的成果。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"image","attrs":{"src":"https:\/\/static001.infoq.cn\/resource\/image\/bd\/d6\/bd2617aeeca5206862c3e50de0fcb7d6.png","alt":null,"title":null,"style":[{"key":"width","value":"75%"},{"key":"bordertype","value":"none"}],"href":null,"fromPaste":true,"pastePass":true}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"eBay 鼓励开源内部项目。正如大家在 eBay 公共 GitHub 上看到的那样，eBay 有 100 多个开源项目，并且会越来越多。我们认为，他们值得与社区分享，使更多的用户从中受益。eBay 的一些重要产品，例如 Kylin，Eagle，Griffin等，有着更大的社区影响力，目前他们都已经贡献到了 Apache 软件基金会。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"除了鼓励内部项目开源之外，eBay 还参与并使用了大量外部开源产品来重建我们的基础架构和平台。我们利用 Kubernetes 构建了我们的第三代云平台。我们还使用了 Apache Hadoop 和 Spark 来支持大量数据处理和即席查询。我们也利用了 Kafka 和 Flink 来构建托管的实时数据平台。在使用这些产品的同时，我们也为它们做出了贡献。在 eBay 这样的大型环境中，我们检测出了许多问题，贡献了数千项对原有项目的改进，并以此回馈社区。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"开源文化就是我们的 DNA，我们也非常高兴我们的技术能与所有行业中领先的技术相结合。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"heading","attrs":{"align":null,"level":2},"content":[{"type":"text","text":"02 Kylin 在 eBay 的发展历程"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"接下来让我谈谈 Kylin 在 eBay 上的发展历程。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"image","attrs":{"src":"https:\/\/static001.infoq.cn\/resource\/image\/53\/75\/53458b0da49779cd49b30a367c928875.png","alt":null,"title":null,"style":[{"key":"width","value":"75%"},{"key":"bordertype","value":"none"}],"href":null,"fromPaste":true,"pastePass":true}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"eBay 和 Apache Kylin 有着悠久的历史与关系。2013 年，为了解决在 Hadoop 大型数据集上的 OLAP 查询性能需求， Kylin 项目在 eBay CCOE 中启动。它基于多维数据集技术构建，可以在亚秒级内响应查询。经过一年的发展，Kylin 成为 Apache 孵化器项目，并最终在 2015 年 11 月毕业成为 Apache 顶级项目。这是 eBay 团队的一项重大创新，并且在今天给更多的社区用户带来了益处。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"heading","attrs":{"align":null,"level":3},"content":[{"type":"text","text":"Cube Planner"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"eBay 一直与 Apache Kylin 社区紧密合作，持续地贡献了有关功能改进以及错误修复的相关代码。在 Kylin 2.3.0 版本中的 Cube Planner 就是来自 eBay 贡献的一项主要功能。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"OLAP 解决方案权衡了在线查询速度和离线 Cube 构建成本，即构建 Cube 的计算资源，以及用于保存 Cube 数据的存储资源。资源效率是 OLAP 引擎最重要的能力。为了达到这个目的，选择性地只预构建那些最有价值的 Cube 是至关重要的。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"Kylin 拥有的 aggregation group 聚合组的功能可以减少 cuboid 的数量。借助 Cube Planner 能使 Apache Kylin 的资源利用率更高。它可以智能地构建 partial cube，以最大程度地降低构建 Cube 的成本，同时最大程度地为最终用户查询提供符合预期的服务，在运行时从查询中学习其模式和规律，并相应地动态地推荐 cuboids。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"image","attrs":{"src":"https:\/\/static001.infoq.cn\/resource\/image\/67\/80\/6732e8ed68744ee3c7d5yy55c2fa2c80.png","alt":null,"title":null,"style":[{"key":"width","value":"75%"},{"key":"bordertype","value":"none"}],"href":null,"fromPaste":true,"pastePass":true}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"在图中大家可以看到，在第 1 阶段中，Cube Planner 可以在构建 Cube 之前根据估计的 cuboid 大小来列出建议构建的 cuboid，在第 2 阶段中，可以根据查询统计信息为现有的 Cube 对 cuboid 列表作出建议。Cube 需要在线一段时间去获得足够的查询命中率，然后 Cube Planner 才能足够智能并且给出建议。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"此功能在 2.5.0 中默认启用，极多维度的 Cube 构建就可以从中受益。更多关于 Cuboid 剪枝优化和 Cube Planner 的介绍👉 "},{"type":"link","attrs":{"href":"http:\/\/mp.weixin.qq.com\/s?__biz=MzAwODE3ODU5MA==&mid=2653081972&idx=1&sn=8cfda39ab42322d53c7cb7259c1aeede&chksm=80a4af85b7d32693627fe71a11078e4f915ac7679a2710a48a5f484f65ae8a78028022c79294&scene=21#wechat_redirect","title":"","type":null},"content":[{"type":"text","text":"Kylin 新手必看：Cube 越用越好，存储越用越少"}]}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"heading","attrs":{"align":null,"level":3},"content":[{"type":"text","text":"实时 OLAP"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"Kylin 还在 Kylin 3.0 中提供了实时 OLAP 功能。有了这个功能，我们可以实时在生成数据后进行决策。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"image","attrs":{"src":"https:\/\/static001.infoq.cn\/resource\/image\/82\/a3\/829d0e9ff4bec37b804a100e17b5b0a3.png","alt":null,"title":null,"style":[{"key":"width","value":"75%"},{"key":"bordertype","value":"none"}],"href":null,"fromPaste":true,"pastePass":true}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"大家可以看到，在 Kylin 架构中，为了实现实时功能，我们添加了蓝色方框中的两个组件。一个是 Streaming coordinator，另一个是 Streaming receiver。Streaming coordinator 负责管理接收器的操作。Streaming receiver 负责从流式数据源获取数据并回答实时数据查询。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"实时段数据首先存储在内存中，然后以列式存储的片段文件形式刷新到接收器的本地磁盘，并且当磁盘上的片段过多时，它们将被自动合并，最后移交到完整的 Cube 构建阶段。无论 cube segment 的数据在哪里，查询都可以随时被回答。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"我们扩展了 Query Engine 查询引擎和 Build Engine 构建引擎，以支持实时数据查询和 Cube 构建。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"在这里概述几个要点："}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"bulletedlist","content":[{"type":"listitem","attrs":{"listStyle":null},"content":[{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"strong"}],"text":"数据准备只是毫秒级延迟"},{"type":"text","text":" ：一旦 Kylin Streaming Receiver 接收了流数据，Cube 的预计算就会在内存中进行，那时实时 Cube 已经能够回答查询请求；"}]}]},{"type":"listitem","attrs":{"listStyle":null},"content":[{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"strong"}],"text":"支持 Lambda 架构"},{"type":"text","text":" ：实时数据通常不可靠，这可能是由多种原因引起的，例如，上游处理系统存在错误，或者数据一段时间后需要更改等。我们支持 Lambda 架构，从批处理源（例如 Hive）刷新 Cube 历史记录；"}]}]},{"type":"listitem","attrs":{"listStyle":null},"content":[{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"strong"}],"text":"更少的 MR 作业和 HBase 表"},{"type":"text","text":" ：在 Kylin 1.6 中，社区提供了一种流解决方案，该解决方案允许用户使用 MR 作业（如几分钟内的微批次）来摄入 Kafka 数据。它将导致系统中存在大量的 Hadoop 作业和小的 HBase 表，增加 Hadoop 系统的负载和发生潜在问题的可能性。"}]}]}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"此功能已在 eBay 中使用，我认为它对于实时场景应该非常有吸引力。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"heading","attrs":{"align":null,"level":2},"content":[{"type":"text","text":"03 Kylin 在 eBay 中的应用"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"strong"}],"text":"目前在 eBay 中，Kylin 平台是为决策分析用户案例提供支持的主要服务。"},{"type":"text","text":" 我们在多个数据中心设置 Kylin 服务，以与 Hadoop 基础架构和业务数据分布保持一致。数据能够以精心设计的安全方式从 Hadoop 和 Rheos 中提取。我们拥有专用的 HBase 集群来稳定服务器的查询延迟。此外，作为平台团队，我们提供 24 * 7 紧急事件响应和技术支持。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"image","attrs":{"src":"https:\/\/static001.infoq.cn\/resource\/image\/54\/66\/54c08c15c2e943bd56101ea76a796d66.png","alt":null,"title":null,"style":[{"key":"width","value":"75%"},{"key":"bordertype","value":"none"}],"href":null,"fromPaste":true,"pastePass":true}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"分享一些重点如下："}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"bulletedlist","content":[{"type":"listitem","attrs":{"listStyle":null},"content":[{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"我们在 40 个项目中分配了 300 多个 Cube。"}]}]},{"type":"listitem","attrs":{"listStyle":null},"content":[{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"最大的 Cube 在源表中具有 1.18 万亿个原始数据，最终的 Cube 大小为 96.2 TB。"}]}]},{"type":"listitem","attrs":{"listStyle":null},"content":[{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"平均构建时间为 16 s\/MB 源数据，平均 Cube 膨胀率为 0.46。"}]}]},{"type":"listitem","attrs":{"listStyle":null},"content":[{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"我们每天处理大约 80 万条查询，平均延迟时间约为 0.68 秒。"}]}]}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"image","attrs":{"src":"https:\/\/static001.infoq.cn\/resource\/image\/7f\/c3\/7fbb3ea93fce224282c31118f5a405c3.png","alt":null,"title":null,"style":[{"key":"width","value":"75%"},{"key":"bordertype","value":"none"}],"href":null,"fromPaste":true,"pastePass":true}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"eBay 中有一些典型的使用模式。 "},{"type":"text","marks":[{"type":"strong"}],"text":"第一种模式即最主要的模式，是为决策提供支撑"},{"type":"text","text":" 。通常，这些用例具有精心设计的业务模型，例如星型模式或雪花模式，它具有清晰的过滤维度和按维度分组以及目标度量。这种模型需要庞大的数据集作为分析基础，以便对每周，每月，甚至每年的趋势进行比较。同时，企业所有者还需要深入分析和汇总分析的功能，并期望他们的决策案例能得到快速响应。这里是一个用例。此案例为库存计划员和采购员提供了类别表现的概述。他们可以进行站点过滤和类别深入分析以及汇总分析，以获取深刻的见解。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"image","attrs":{"src":"https:\/\/static001.infoq.cn\/resource\/image\/13\/fd\/13683efc1c9782ae73f29c9abb6635fd.png","alt":null,"title":null,"style":[{"key":"width","value":"75%"},{"key":"bordertype","value":"none"}],"href":null,"fromPaste":true,"pastePass":true}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"strong"}],"text":"第二种模式是为分析师提供 eBay 自助服务分析指导，以探索更多的商机"},{"type":"text","text":" 。用户案例所有者（如 eBay 数据服务和解决方案团队）将在 IDL 数据集上构建 Cube。IDL 是集成数据层，它是一组用通用业务逻辑进行了转换的表，可以为一个域或跨域内的一个或多个业务流程提供服务。基于这些数据的 Cube 具有许多相关的度量和维度。统一的UI解决方案为最终用户提供了灵活性，可以轻松地拖放不同的维度和度量组合以进行特定分析。这是一个用例，最终用户可以从 30 个维度和 200 个度量中选择一种组合来分析 B2C 业务绩效。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"image","attrs":{"src":"https:\/\/static001.infoq.cn\/resource\/image\/69\/cf\/699e320c1247134237b54d28a51f2fcf.png","alt":null,"title":null,"style":[{"key":"width","value":"75%"},{"key":"bordertype","value":"none"}],"href":null,"fromPaste":true,"pastePass":true}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"strong"}],"text":"第三种模式是实时 OLAP"},{"type":"text","text":" 。最重要的需求是数据处理后的度量的实时反映。而且，由于实时事件管道中存在一些不确定性，用户要求 Lambda 架构从 ETL 流中调整度量精度。这是 BU 需要实时监视站点电子邮件号码以了解其性能的一个示例。同时他们还需要回顾庞大数据集中的逐年指标。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"heading","attrs":{"align":null,"level":2},"content":[{"type":"text","text":"04 eBay 未来计划"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"由于 eBay 朝着向目标用户（如分析师和业务部门）提供更多自助服务和智能数据分析功能的方向发展，我们的数据服务和解决方案团队正在构建增强版分析解决方案，该解决方案需要更多来自 Kylin 平台的灵活性和弹性支持，例如从多个 Hadoop 群集中提取数据，对庞大查询量的系统弹性，以及构建高维 Cube 的灵活性和性能。因此，建立更高的灵活性和弹性系统将是我们明年的首要任务。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"此外，我们还在考虑替代 HBase 的存储解决方案。Kylin 社区已经开始了一项重大计划，也就是使用 Parquet 提供轻量，列式存储，云友好的解决方案。我们明年也计划对存储解决方案部分进行一些调研。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"在过去的两年中，我们的团队致力于一项重大计划，该计划通过在 SparkSQL 上进行大量软件优化来提供对 Hadoop 数据的快速特定查询体验。我们也会考虑将 Kylin 用作整个蓝图中的加速层之一。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"strong"}],"text":"作者介绍"},{"type":"text","text":"："}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"Lisa Li，在 eBay 中国研发中心大数据平台部门担任研发主管。带领的团队主要负责 SQL on Hadoop 的方案，给使用 SQL 查询语言的数据分析师在开源的 Hadoop 数据平台上提供更加快速稳定的数据处理和查询体验。主要包括 Kylin 和 Spark 两个软件产品的深度开发和服务运维工作。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"strong"}],"text":"本文转载自公众号apachekylin（ID：ApacheKylin）。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"strong"}],"text":"原文链接"},{"type":"text","text":"："}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"link","attrs":{"href":"https:\/\/mp.weixin.qq.com\/s?__biz=MzAwODE3ODU5MA==&mid=2653082008&idx=1&sn=1df8f2501cf72d43e5f25f9f6f5b2943&chksm=80a4af69b7d3267f81580563ef38e4cbe0694a938ae9839d3704c5f82c3e84e98569b98ebca1&token=1340822333&lang=zh_CN#rd","title":"","type":null},"content":[{"type":"text","text":"Kylin 在 eBay 的成长历程与实践"}]}]}]}