历时四年，分布式文件系统 JuiceFS 正式开源

{"type":"doc","content":[{"type":"blockquote","content":[{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"JuiceFS 是为海量数据设计的分布式文件系统，使用对象存储来做数据持久化，避免重复造轮子，还能大大降低工程复杂度，让开发者专注解决元数据和访问协议部分的难题。 "}]}]},{"type":"heading","attrs":{"align":null,"level":2},"content":[{"type":"text","text":"云原生分布式文件系统 JuiceFS 正式开源"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":" "}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"1月11日， Juicedata果汁数据科技官方账号宣布JuiceFS "},{"type":"link","attrs":{"href":"https:\/\/mp.weixin.qq.com\/s\/asUFtorMB1pcOwhNvlDyKw","title":null,"type":null},"content":[{"type":"text","text":"正式开源"}]},{"type":"text","text":"，这是一个云原生分布式文件系统，经过了四年的持续迭代和累计几千万小时的线上考验。Davies（刘洪清，Juicedata 创始人、TGO 鲲鹏会会员）在文中表示："}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":" "}]},{"type":"blockquote","content":[{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"JuiceFS 的创新架构更符合云原生的发展趋势，我们一开始就以 SaaS 的形式将它提供给公有云的客户，让客户分钟级就可以获得 PB 级企业文件存储服务。同时，我们也和行业优秀的对象存储厂商一起服务私有云客户。"}]}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":" "}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"GitHub地址："},{"type":"link","attrs":{"href":"https:\/\/github.com\/juicedata\/juicefs","title":null,"type":null},"content":[{"type":"text","text":"https:\/\/github.com\/juicedata\/juicefs"}]}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":" "}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"2020年2月份，Davies曾在接受TGO采访时聊起了一路创办Juicedata的故事（可阅读："},{"type":"link","attrs":{"href":"https:\/\/mp.weixin.qq.com\/s?__biz=MzA4NTU2MTg3MQ==&mid=2655171742&idx=1&sn=f13ee09c27cfeaabb9ca2d05a48f6810&chksm=846013feb3179ae870a19a6db31c5be9b31fee94a333a248fa16d4add4f8e6ed5d764da927c5&scene=27#wechat_redirect","title":null,"type":null},"content":[{"type":"text","text":"《离开 Facebook 与百万美元后，他的技术故事才刚刚开始 | TGO 深焦》"}]},{"type":"text","text":"）。Davies在清华硕士毕业后即加入豆瓣成为早期员工，并研发了国内最早的开源 KV 存储 Beansdb 和 DPark （ Python clone of Spark ）；2013 年他加入 Facebook 总部负责 HDFS 方面的研发，2014 年加入 Databricks，帮助 Spark SQL 实现了上百倍的性能提升。2017 年，似乎是顺利成章，在大数据领域摸爬滚打了近十年的 Davies，开始了与几个伙伴的创业之旅。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":" "}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"“创业太艰难了，对一个团队的要求太全面了。你不能犯任何错误，要竭尽全力，然后要借助所有可能借助的力量，还要借助偶然的机遇才可能成功……（我）觉得这个事情是非常难的，还是不要创业。”"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":" "}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"话虽如此，但创业似乎对 Davies 有种奇妙的吸引力。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":" "}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"在他长达 13 年的职业生涯中，居然有近 12 年身在创业企业。2016 年，正是在硅谷创业企业 Databricks 任职期间，他萌生了创业的想法。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":" "}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"时值 Davies 负责为 Databricks 的存储层提速，虽然 AWS 已有相关的存储方案，但问题很多，且迟迟无法解决。于是，他提议，自研新的存储方案，系统性地解决问题。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"不过，在当时的 Databricks，从架构师到管理层，几乎全部认为风险太大，无人支持 Davies 的提议。Davies 说：“当时， CTO （注：Matei Zaharia，Apache Spark 作者）亲口对我说：‘存储这不是我们擅长的事情，能不碰尽量不要碰。’“"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":" "}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"在 Databricks 否决 Davies 的技术方案后，大概 Matei Zaharia 也没有想到，这个中国来的工程师颇有“美式英雄主义”精神。他不但没有放弃，反而用业余时间单枪匹马地写了个原型出来。之后，Davies 回忆道：“我找了一些朋友的公司去试用，发现效果也可以，所以我在想既然有这么不错的东西，就不能埋没它。”"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":" "}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"strong"}],"text":"2017 年，Davies 在美国远程敲定了国内的投资和早期客户，叫上当时也在创业的苏锐，共同创立了 Juicedata，并将产品命名为 JuiceFS。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":" "}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"strong"}],"text":"经过了四年的历练，整个团队最终决定将JuiceFS正式开源。"},{"type":"text","text":"Davies表示：在创业之初，我们认为 SaaS 可以为用户提供最佳的体验，同时让我们更快地迭代产品，决定优先把 SaaS 做好。经过 4 年的持续迭代和积累，JuiceFS 已经在几十家科技企业的大数据、AI、容器平台、归档、备份等场景中形成最佳实践， SaaS 使用量也持续快速增长，并且在过去的 2020 年首次实现了盈亏平衡。我们相信找到了可持续发展的模式，有信心保障 JuiceFS 的长期运营。"}]},{"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，Davies等人也发现闭源的基础软件会限制使用者对它的深度理解，不利于服务更多人，依靠 SaaS 产品的收入支撑和开源社区的力量或许可以让 JuiceFS 走得更远。"}]},{"type":"heading","attrs":{"align":null,"level":2},"content":[{"type":"text","text":"为什么我们需要新的文件系统？"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":" "}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"文件系统是计算机中一个非常重要的组件，为存储设备提供一致的访问和管理方式。在不同的操作系统中，文件系统会有一些差别，但也有一些共性几十年都没怎么变化："}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":" "}]},{"type":"bulletedlist","content":[{"type":"listitem","attrs":{"listStyle":null},"content":[{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"数据是以文件的形式存在，提供 Open、Read、Write、Seek、Close 等API 进行访问；"}]}]},{"type":"listitem","attrs":{"listStyle":null},"content":[{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"文件以树形目录进行组织，提供原子的重命名（Rename）操作改变文件或者目录的位置。"}]}]}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":" "}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"文件系统提供的访问和管理方法支撑了绝大部分的计算机应用，Unix 的“万物皆文件”的理念更是凸显了它的重要地位。文件系统的复杂性使得它的可扩展性未能跟得上互联网的高速发展，极大简化了的对象存储及时填补了空缺得以快速发展起来。因为对象存储缺乏树状结构也不支持原子重命名操作，跟文件系统有很大的差别，本文暂不讨论。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":" "}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"绝大多数文件系统都是单机的，在单机操作系统内为一个或者多个存储设备提供访问和管理。随着互联网的高速发展，单机文件系统面临很多的挑战："}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":" "}]},{"type":"bulletedlist","content":[{"type":"listitem","attrs":{"listStyle":null},"content":[{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"共享：无法同时为分布在多个机器中的应用提供访问，于是有了 NFS 协议，可以将单机文件系统通过网络的方式同时提供给多个机器访问。"}]}]},{"type":"listitem","attrs":{"listStyle":null},"content":[{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"容量：无法提供足够空间来存储数据，数据只好分散在多个隔离的单机文件系统里。"}]}]},{"type":"listitem","attrs":{"listStyle":null},"content":[{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"性能：无法满足某些应用需要非常高的读写性能要求，应用只好做逻辑拆分同时读写多个文件系统。"}]}]},{"type":"listitem","attrs":{"listStyle":null},"content":[{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"可靠性：受限於单个机器的可靠性，机器故障可能导致数据丢失。"}]}]},{"type":"listitem","attrs":{"listStyle":null},"content":[{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"可用性：受限於单个操作系统的可用性，故障或者重启等运维操作会导致不可用。"}]}]},{"type":"listitem","attrs":{"listStyle":null},"content":[{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"随着互联网的高速发展，这些问题变得日益突出，涌现出了一些分布式文件系统来应对这些挑战。"}]}]}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":" "}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"如今，我们耳熟能详的分布式文件系统有GlusterFS、CephFS、GFS以及HDFS等，每种方案各有优劣（详细对比可阅读"},{"type":"link","attrs":{"href":"https:\/\/mp.weixin.qq.com\/s\/ZwfRZW42e_2dh0u6vH0psg","title":"xxx","type":null},"content":[{"type":"text","text":"《分布式文件系统架构对比》"}]},{"type":"text","text":"一文），JuiceFS与这些相比的差异性在于更加适合云原生的环境。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":" "}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"具体而言，GFS、HDFS 和 MooseFS 都是针对自建机房这种软硬件环境设计的，将数据的可靠性和节点可用性合在一起用多机多副本的方式解决。但是在公有云或者私有云的虚拟机里，块设备已经是具有三副本可靠性设计的虚拟块设备，如果再通过多机多副本的方式来做，会导致数据的成本居高不下（实际上是 9 个拷贝）。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":" "}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"于是，整个团队针对公有云，改进 HDFS 和 MooseFS 的架构，设计了 JuiceFS，架构如下图所示："}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":" "}]},{"type":"image","attrs":{"src":"https:\/\/static001.geekbang.org\/infoq\/8a\/8ac68b6f9758f4ea4c363d2ecd832b73.jpeg","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":" "}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"JuiceFS 使用公有云中已有的对象存储来替换 DataNode 和 ChunkServer，实现一个完全弹性的 Serverless 的存储系统。公有云的对象存储已经很好地解决了大规模数据的安全高效存储，JuiceFS 只需要专注元数据的管理，也大大降低了元数据服务的复杂度（GFS 和 MooseFS 的 master 要同时解决元数据的存储和数据块的健康管理）。团队也对元数据部分做了很多改进，从一开始就实现了基于 Raft 的高可用。要真正提供一个高可用高性能的服务，元数据的管理和运维仍然是很有挑战的，元数据是以服务的形式提供给用户。因为 POSIX 文件系统 API 是应用最最广泛的 API，团队基于 FUSE 实现了高度 POSIX 兼容的客户端，用户可以通过一个命令行工具把 JuiceFS 挂载到 Linux 或者 macOS 中，像本地文件系统一样快速访问。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":" "}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"上图中右边虚线部分是负责数据存储和访问的部分，涉及用户的数据隐私，它们是完全在客户自己的账号和网络环境中，不会跟元数据服务接触。Juicedata没有任何方法接触到客户的内容（元数据除外，请不要把敏感内容放到文件名里）。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":" "}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"image","attrs":{"src":"https:\/\/static001.geekbang.org\/infoq\/0d\/0dee2f7d15b56a766a3722c450c2fa14.jpeg","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},"content":[{"type":"text","text":"上图中蓝色项目主要是给大数据场景使用的，实现的是 POSIX 的子集，而绿色的则是POSIX 兼容的文件系统。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":" "}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"他们中以 GFS 为代表的元数据和数据分离的系统设计能够有效平衡系统的复杂度，有效解决大规模数据的存储问题（通常也都是大文件），有更好的可扩展性。这个架构下支持元数据的分布式存储的 Colossus 和 WarmStorage 更是具有无限的可扩展能力。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":" "}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"JuiceFS 作为后来者，学习了 MooseFS 实现分布式 POSIX 文件系统的方式，也学习了 Facebook 的 WarmStorage 等元数据和数据彻底分开的思路，希望为公有云或者私有云等场景下提供最好的分布式存储体验。JuiceFS 通过将数据存储到对象存储的方式，有效避免了使用以上分布式文件系统时的双层冗余（块存储的冗余和分布式系统的多机冗余）导致的成本过高问题。JuiceFS 还支持所有的公有云，不用担心某个云服务锁定，还能平滑地在公有云或者区之间迁移数据。"}]},{"type":"heading","attrs":{"align":null,"level":2},"content":[{"type":"text","text":"为开源做好准备，架构再升级"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":" "}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"借助对象存储的帮助，JuiceFS 已经大大降低了分布式文件系统的复杂度，元数据管理是它最核心的问题。JuiceFS 的 SaaS 使用的元数据引擎，是专为文件系统打造的数据库，整个团队已经积累了丰富的运维经验，仍然如履薄冰。如果开源的话，让社区用户自己运维仍然会是一个大的挑战和负担，一旦运维失误导致数据丢失，后果非常严重。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":" "}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"带着这个问题，团队将元数据服务改造为支持多引擎的插件式架构，可以利用已有的开源数据库实现元数据存储。这样可以更灵活地适应不同场景，根据场景的规模、性能和成本需求，选用不同的元数据实现。这是 JuiceFS 的架构再升级，为未来的发展翻开新的篇章。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":" "}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"至于为什么选用 Redis 作为第一个开源存储引擎，是因为该引擎："}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":" "}]},{"type":"bulletedlist","content":[{"type":"listitem","attrs":{"listStyle":null},"content":[{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"是全内存的，可以满足元数据的低延时和高 IOPS 要求；"}]}]},{"type":"listitem","attrs":{"listStyle":null},"content":[{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"支持乐观事务，能够满足文件系统元数据操作的原子性要求；"}]}]},{"type":"listitem","attrs":{"listStyle":null},"content":[{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"有丰富的数据结构，易于实现文件系统的诸多 API；"}]}]},{"type":"listitem","attrs":{"listStyle":null},"content":[{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"有着非常广泛的社区和成熟的生态，运维 Redis 不会是一个问题；"}]}]},{"type":"listitem","attrs":{"listStyle":null},"content":[{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"在各个云上都有托管的服务，在云上使用会更简单；"}]}]}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":" "}]},{"type":"image","attrs":{"src":"https:\/\/static001.geekbang.org\/infoq\/ff\/ffe83b481c2edd0c3d9513458ab23349.jpeg","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":"未来，团队还会增加 SQL 数据库、TiKV 等支持事务的 KV 数据库支持。"}]},{"type":"heading","attrs":{"align":null,"level":2},"content":[{"type":"text","text":"未来规划"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":" "}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"最近几年，数据库领域发生了一件有趣的事情：当 NoSQL 数据库在满足了数据的快速增长后，它在一致性、访问便捷性和管理能力方面的不足逐渐显露，把这些复杂性转嫁到了业务系统和运维上，开始被人诟病。同时， SQL 数据库也有了长足的进展，已经能够满足现在的数据规模需求，经过全面的对比分析后，大家又在回归 SQL 数据库，曾经的 NoSQL 运动也逐渐显出颓势。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":" "}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"估计类似的事情也会发生在非结构数据领域。对象存储在媒体文件等场景取得了巨大的成功，但当人们以为它就是未来的存储形态，开始推广到更大范围时，它牺牲掉的树形目录结构、可修改性、元数据性能、一致性等等，变成了一只只拦路虎，影响它在其他场景的使用效果。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":" "}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"团队坚信文件系统是最好的管理非结构化数据的方式，对象存储只适用于某些简单场景。分布式文件系统一直是基础软件中难啃的骨头，JuiceFS 通过对文件系统中元数据和数据的独立抽象，大大减低了系统复杂度，使得文件系统能够借助这些年来对象存储和分布式数据库的进展，管理超大规模的数据。同时，复杂度的降低可以让更多的开发者参与进来，未来更多的应用也会建立在文件系统接口之上。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":" "}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"团队将通过开源社区的相互协作，一方面为各个应用提供更好的存储支持，也会在底层存储引擎和对象存储上加深协作，一起推动文件存储的快速发展，打造未来数据生态的坚实底座。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":" "}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"消息来源："}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":" "}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"link","attrs":{"href":"https:\/\/mp.weixin.qq.com\/s\/asUFtorMB1pcOwhNvlDyKw","title":null,"type":null},"content":[{"type":"text","text":"https:\/\/mp.weixin.qq.com\/s\/asUFtorMB1pcOwhNvlDyKw"}]}]}]}