It's a really interesting new (emerged in the last few years) piece of technology.
It's kind of a cross between SQLite and columnar analytical databases such as Snowflake or Cassandra.
You can run it happily on a laptop, but because it's column oriented it can calculate aggregates (group by/count queries for example) incredibly fast - so it's better for a lot of analytical workloads than regular row-oriented databases.
Cassandra isn't columnar, it's row-oriented. The original authors coined a new term "column family" which is just a nonsense (and confusing) term, meant to describe its original relatively unstructured API, which has unfortunately stuck in all descriptions of the database, and leads to confusion like this.
Cassandra is also not an analytics database, it is intended for "OLTP"-like workloads.
Thanks. I wanted another example other than Snowflake so I did a quick Google search and grabbed the first example from the top search summary box that I recognized.
I always considered Cassandra's 'column' designation a guarantee that OLTP transactions will apply atomically to column groups; or that you'll minimize IO writes for a row to the column group (as opposed to writing a new version of the entire row). But this is actually just a wild guess. Anyone want to chime in and confirm/deny?
DuckDB is a no-dependency SQL engine that's available from multiple programming languages. A bit like SQLite but optimised for analytics.
For instance, in Python, it can be installed using `pip install duckdb`. For things like unit tests, it can be very valuable because no additional services need to be running.
