Distributed SQLite designs answer different failure questions

Rqlite creator Philip O’Toole distinguishes three ways to distribute SQLite: consensus replication for continued service, log streaming for recovery after losing one machine, and mergeable replicas for independently changing copies. Rqlite sends writes through a leader and requires a majority, so a disconnected minority refuses writes. Those guarantees solve a different problem from a single-node backup stream or a local-first synchronization system.

Philip O’Toole · Oct 23, 2026

Why it matters

Choosing replication starts with acceptable downtime, write availability and conflict behavior rather than a shared SQLite label.

More context

The October 23 retrospective compares rqlite, embedded dqlite, Litestream and local-first extensions. This card focuses on their architectural tradeoffs, not the author’s priority claims or acquisition history. Write down the failure scenario and required recovery time before choosing a design.

Keep in mind

This is a maintainer’s historical perspective, not a comparative benchmark. Products and read-consistency modes differ; the broad categories do not replace checking current guarantees and operating requirements.

Original read