Learning a few things about running SQLite

Julia Evans reflects on running a Django site backed by SQLite — her fourth SQLite-backed web project — and what she's learned from the rough edges encountered in practice.

A slow FTS5 full-text search query (5 seconds on a 4,000-row table) was fixed by running ANALYZE, which dropped it to ~0.05 seconds. She suspects a query-planning issue and muses about eventually learning to read query plans.

DELETE-heavy cleanup jobs would lock the database for over 5 seconds, causing other workers to hit timeouts and crash the VM. Her workaround: batch deletes into smaller chunks. This gave her more appreciation for why someone might reach for Postgres when concurrent writes matter.

She found ORM performance from Django's ORM mostly fine on a small dataset (~10,000 rows). Backup is handled two ways: restic with VACUUM INTO (plagued by OOM kills), and Litestream (newer, still evaluating). She backs up to AWS S3 but finds credential generation annoying.

Her Mess with DNS project has run on SQLite since 2022 after migrating from Postgres — a move she calls a great choice. She closes by noting how long it takes to learn basic things about the tools you use every day: she'd been using SQLite for web projects for four years before learning about ANALYZE, and expects to discover another basic feature in another year or two.

Sources: raw/learning-about-running-sqlite Last updated: 2026-08-01