All of these products are great, but in practice I used only about 40% of the features a database provides.

I found these databases too big and complex. In data warehousing, meanwhile, good solutions keep coming out one after another.

I doubt that Why should the primary key be mutable?

I doubt that rather than optimizing a query plan based cost every single query execution, how about optimizing data page layout?

In my experience, Read Committed and Repeatable Read were enough for most workloads.

In most modern applications and servers, developers no longer keep core business logic inside the database.

As micro services become popular, logic is separated across services and carried out as a chain of short transactions.

So I built KDBS, an OLTP-focused relational database (C++ implementation: CKDBS). Its core idea is a few self-imposed constraints.

No Procedure, No Triggers, only with necessary features like transactions, MVCC, Buffer Pool, WAL, Index, Backup, plus Assertions(SQL-92) which most databases never implemented,

and Transitions (define which value changes are allowed for a field, such as an order status moving only from “placed” to “paid” to “shipped”.. )

KDBS gives up generality in exchange for simplicity, predictability, extensibility, and performance.

The KDBS website provides a sandbox where you can try KDBS SQL with a single click or by typing your own queries.

https://kdbs.dev/sandbox

I would like to hear your opinions and advice. Do you think this project is worth building?

I believe it is, but I would greatly value your honest feedback.

All codes were written by using Claude Code CLI, mainly with Opus model.