After diving into Alex Petrov’s *Database Internals*, the author spent 15 hours crafting a B‑tree from scratch in Go. The project yielded roughly 200 lines of code and 20+ unit tests, all achieving 100 % coverage. The exercise highlighted how TDD surfaces subtle bugs that would otherwise slip into production.

Unlike a binary tree, a B‑tree node can hold many children, a design that reduces costly disk reads. The author noted that root splits create a new root with a promoted key, making the tree grow taller rather than wider. Cascading splits require careful pointer rewiring, a detail that proved tricky to debug.

Building the structure revealed the practical gap between theory and implementation. The author compared B‑tree to B+tree, noting that the latter stores data only in leaves, improving range scans at the cost of early‑stop searches. Sequential inserts produced a 7‑level skewed tree, underscoring why bulk loading and REINDEX exist.

Next on the roadmap is implementing LSM‑tree logic, a structure favored by Cassandra for write‑heavy workloads. Understanding both B‑tree and LSM‑tree trade‑offs will explain why PostgreSQL opts for B+trees while Cassandra prefers append‑only logs. The author invites readers to explore the full code on GitLab.