Aug 19, 2019
The latest San Francisco Bay Area ClickHouse Meetup was in Silicon Valley on August 13th. We had between 25 and 30 attendees at H2O.ai, who kindly hosted the event at their offices in Mountain View. The crowd was enthusiastic, leading to a lot of back-and-forth questions during the presentations. We had a total of three talks.
- ML model generation using the H2O platform (Rafael Coss, H2O.ai) – Raphael gave a lightning talk showing how H2O generates models automatically from datasets. He also delved into the problem of integrating machine learning with databases. For example, it would be great to take rows describing customers, run a model to estimate customer lifetime value for each, and put the results back into the DBMS.
- How Clickhouse became the default analytics database for Mux (Adam Brown, Mux founder) Mux.com is an API-driven content delivery network for video. It uses ClickHouse to implement analytics for key aspects of video delivery. The web-based analytics have fast response and outstanding graphics. ClickHouse performance is sufficiently fast that Mux does not bother to aggregate or down-sample data, for example using materialized views. They just work directly off fact tables.
- Tricks every ClickHouse designer should know (Robert Hodges, Altinity) — My talk introduced a number of useful ClickHouse techniques ranging from encodings that can reduce data size dramatically to using replication to implement backups. The Q&A included a discussion of ClickHouse transactional semantics and how to recover from failures. That looks like a good topic for future blog articles.
The meetup touched on a number of areas that Altinity has worked on directly.
- Mux.com is running on Kubernetes and planning to adopt the ClickHouse Kubernetes operator, which Altinity developed. That was really good to hear. We want to build a big community around the operator. User feedback is gladly welcome.
- There was a question whether there’s a UI for ClickHouse. As of Wednesday, August 21st, the answer is “yes.” We’re doing a webinar to introduce it on that day.
- My talk covered ClickHouse encodings, which now cover a wide range of use cases to reduce data. Vasily Nemkov from Altinity implemented Gorilla and DoubleDelta among many other PRs he has submitted this year. Alexander Zaitsev described them and published encoding test results on this blog. Our work adds to existing encodings such as LowCardinality data type. ClickHouse is quite strong in this area.
We would like to thank H2O.ai for providing the venue, Raphael and Adam for presenting, and everyone else for attending and asking great questions. The next SF ClickHouse meetup is planned for early October in San Francisco. Please join us!