Hacker News Reader

> interest in NoSQL-style computation seemed to be on the wane

Well that's just ridiculous. With machine/deep learning and big data analytics unquestionably on the rise there is actually more demand for a heterogenous architecture. And since in these fields you are often doing lazy-schema application having tools that support this paradigm are important.

In fact what does seem to be on the rise is having SQL on top of NoSQL stores e.g. Phoenix/HBase, Spark SQL. And they support all of SQL 2003 which by the looks of it this doesn't.

Is this like a data table like in the pandas package for Python, or in R?

I don't think those data tables are schemaless. It seems like the value prop here is that it efficiently handles heterogenous data shapes.

I would love to see the implementation of this. In the article doesn't say anything about the implementation details or the space tradeoffs.

https://github.com/kevinlawler/kerf

It has no benchmarks so that's not a good sign. But there is another time series database that also allows you to use basic SQL: InfluxDB.

I know about InfluxDB but I was interested in the implementation of the data structure. I'll check the git, thanks

-- edit --

Oh, that's shitty! The git is not the code of Kerf, is some documentation and examples :(

Atlas Data Structure