MastiDB

Why I built it

I wanted to understand databases from the inside — not as a user of Postgres or Druid, but as someone who has to make the storage, planning, and execution choices himself. MastiDB is that learning engine: an on-disk OLAP system built from scratch so the abstractions stop being magical.

How it works

The path of a query is deliberately simple and readable:

1. Storage — data lands in columnar, on-disk layouts chosen for analytical scans rather than row-oriented lookups.
2. Planning — a query planner turns a request into an executable plan, choosing how to read and combine data.
3. Execution — an execution engine walks that plan, pulling batches through operators.
4. Scans — vectorised scans move through columns in chunks, keeping the hot path tight and inspectable.

Each stage is small enough to read, change, and reason about — which is the point of building it this way.