I wanted to understand how modern OLAP engines (like DuckDB or ClickHouse) work under the hood, so I built one from scratch. The goal wasn't just "make it work," but to use every systems programming trick available to maximize throughput on Linux.

Frigatebird is an OLAP engine built from first principles. It features a custom storage engine (Walrus) that uses io_uring for batched writes, a custom spin-lock allocator, and a push-based execution pipeline. I explicitly avoided async runtimes in favor of manual thread scheduling and atomic work-stealing to maximize cache locality. Code is structured to match the architecture diagrams exactly.

currently it only supports single table operations (no JOINS yet), would love to hear your thoughts on it.