[urthor]

There's reasons for one wide table.

The traditional SQL database uses row based data structures.

When you shift to columnar data formats for the table, parquet or any columnar DB, the normalization rules which were developed for row based become extremely different.

The brave new world now is in-memory DBs.

All those 1970s rules, which make sense for row based tables stored on disk, don't apply to data in RAM.

At all.

So it's going to get very interesting.

[goto11]

Shouldn't the normalization rules be exactly the same whether the data is stored row-oriented or column-oriented and on disk or memory?

[mulmen]

This is a common misconception.

Columnar stores do indeed enable wide tables but they don’t replace the need for normalization or the benefits. Wide tables and normalization (or denormalization) solve different problems.

From the perspective of a database the difference between RAM and disk is latency. Normalization is still a factor in query performance.

[Akronymus]

> There's reasons for one wide table.

There are indeed. It should still be a conscious decision. For the DB I work with, it wasn't.