SQLite replication moves from open-source projects to paid services
Rqlite creator Philip O’Toole traces a dozen years of efforts to add replication to SQLite, from consensus clusters to recovery tools and hosted platforms.
In a retrospective published by rqlite creator Philip O’Toole, the project’s history begins with a practical problem: how to keep a small relational database available when a server fails. Launched in September 2014, rqlite placed a Raft log in front of SQLite and replicated writes across a cluster.
O’Toole describes rqlite as a consensus-based system: writes pass through a leader and require acknowledgement from a majority. If a network partition leaves a minority of nodes, they refuse writes. He says the design aims to keep service available after a node failure, without waiting to restore a single-node database.
The retrospective groups distributed SQLite approaches into three broad models. Rqlite and Canonical’s dqlite use consensus clustering; Litestream streams SQLite’s write-ahead log for recovery; and cr-sqlite uses conflict-free data types to let multiple writers merge changes. O’Toole says each model addresses a different need, from strong consistency to inexpensive recovery or local-first applications.
Other projects have extended the model into products. The article cites LiteFS, libSQL/Turso, SQLite Cloud and Cloudflare D1, among others. O’Toole’s account traces a shift from open-source tools people run themselves towards hosted services and managed products.
O’Toole argues that open-source efforts preceded the category’s commercial funding and acquisitions by years. He says rqlite, dqlite and Litestream remain free open-source projects, while commercial attention has focused on services built around replicated SQLite.
