ClickHouse 在这里
Video analytics is used by millions of Vimeo users, ranging from individual content creators to large enterprises, facilitating in-depth insights into viewing patterns for hundreds of millions of videos. With real-time data spanning any desired time frame, our analytical platform beautifully handles over a billion viewing log events every day.
视频分析被数百万Vimeo用户使用，从个人内容创作者到大型企业，为数亿个视频的观看模式提供深入洞察。我们的分析平台每天处理超过10亿个实时数据，时间跨度可根据需要设定。
In this post, I’ll outline our journey from a traditional architecture anchored in Apache Phoenix on HBase to our embrace of ClickHouse just eighteen months ago. Along the way, I’ll pepper in key learnings and tips, making this a read you won’t want to skip.
在本文中，我将概述我们从传统架构（基于Apache Phoenix和HBase）到我们在仅仅18个月前开始使用ClickHouse的过程。在此过程中，我将穿插关键的经验和技巧，使本文成为您不容错过的阅读。
From Copenhagen Airport in Kastrup, Denmark. Photo by Nazrin Babashova on Unsplash.
来自丹麦Kastrup的哥本哈根机场。照片由Nazrin Babashova拍摄，来源于Unsplash。
What is video analytics data?
什么是视频分析数据？
Video analytics are broken down into user sessions. Think of these as individual viewing experiences, characterized by dimensions like device, location, and source to derive viewing metrics such as views, view time, and much more. Keep in mind, a single user can have multiple sessions for a video — they might watch a portion, pause, then resume later or even revisit days later.
视频分析被分解为用户会话。可以将其视为个体的观看体验，其特征包括设备、位置和来源，以推导出观看指标，如观看次数、观看时间等。请记住，单个用户可以对一个视频有多个会话 - 他们可以观看一部分，暂停，然后稍后继续观看，甚至几天后再次观看。
On the back end, we capture the micro events that shape each session in real time: plays, loads, regular heartbeats, and user actions. This granular data enables robust video metric construction, enabling aggregation over any time frame and refinement using the myriad of dimensions we collect.
在后端，我们实时捕获塑造每个会话的微事件：播放、加载、定期心跳和用户操作。这种细粒度的数据能够实现强大的视频指标构建，可以在任何时间范围内进行聚合，并使用我们收集的各种维度进行细化。
What problem are we trying to solve
我们试图解决什么问题
For years, our analytics platform rested on a widespread architectural approach: an ACID key-value storage leveraging HBase, which is a Hadoop JVM...