ACID, isolation levels, indexing strategy, the N+1 trap, locking, and the day-to-day failure modes of relational databases — practical notes from real-world use.
7 articles · updated regularly
Storage engines must guarantee that a sudden power loss or database crash mid-transaction leaves the data file completely uncorrupted. We compare Shadow Paging (used in LMDB and SQLite journal mode) against Write-Ahead Logging (WAL, used in PostgreSQL, InnoDB, and RocksDB).
Two requests each removed a different admin from a project. Each one first checked that at least one other admin would remain, and each check passed, because at the moment it ran another admin really did still exist. Both committed. The project ended up with zero admins, and our Repeatable Read transactions never noticed a thing.
A simple ALTER TABLE that should have been instant. A user_id column. A live application. Nine hours of downtime. Here's the post-mortem and the playbook.
Two strategies for concurrent writes. Pick wrong and you either serialise all your traffic or lose data quietly.
Why a single missing index can turn your 10ms endpoint into a 10-second endpoint — and why adding the wrong one makes writes slower.
The single most common performance bug in ORM-backed apps. Here's how it sneaks in and the three reliable ways to stop it.
Atomicity, Consistency, Isolation, Durability — a plain-English tour plus the real-world price tag of each.