My toy relational database, built from scratch in Zig — no dependencies, no prior database experience. I’m learning in public. The goal is a modular architecture where each layer is independently swappable. Early, unstable, and changing fast.
## 02 ─ internal/layers ─────────────────Each layer is independently swappable — plug in a different buffer pool, storage backend, or transaction strategy without touching the rest. This is also my live progress log: what’s built, what’s planned, and what I’m still working out.
Toy project, yes — but sloppy results aren't the goal. The aim is a well-tested implementation with growing coverage across SQL correctness, durability, and concurrency.
The SQLite project's standard test corpus — thousands of query-and-result assertions covering a wide surface of SQL. The clearest measure of how much standard SQL PagerDB actually gets right.
Repeatable crash-and-recovery scenarios to verify pager, WAL, and transaction behavior under controlled fault injection.
not published yet
Random SQL generation and state mutation to surface parser, planner, and execution engine edge cases.
not published yet
## 06 ─ frequently asked ─────────────────────────────────────────── Questions no one asked, but I'm answering anyway.
No. Not even close. It's not mature enough to be considered a toy database, let alone a production one. If you're asking this seriously, I'm concerned.
I just wanted to learn Zig. No grand vision, no benchmark slides. Just vibes.
I'm not accepting contributions right now — that would defeat the whole point of a learning project.
Yes, a lot — but I try not to let it just write everything for me. I slow down, read the output, understand it, and rewrite it myself when I can. Muscle memory in Zig won't build itself.