Philip O'Toole published rqlite in September 2014 — a SQLite database with a Raft consensus log in front of it, replicated across a cluster so that losing a node did not mean losing data. As far as anyone can document, it was the first open-source system to combine Raft and SQLite. For several years nobody else tried. The design space filled in three phases. Consensus replication came first: rqlite (2014), Canonical's dqlite (2017, built as an embedded library for LXD and MicroK8s). Streaming came second: Ben Johnson's Litestream (2021) watched the SQLite WAL and shipped it to S3, giving disaster recovery without a cluster. Merge came third: cr-sqlite (2022) added CRDTs so multiple writers could diverge and converge without a leader. Each approach is correct for a different failure mode, and most of the noise in the category is people arguing about which failure mode matters. The money, predictably, went to the platform layer. Turso forked SQLite itself, raised venture capital, and was acquired by Supabase in 2026. Cloudflare shipped D1 as serverless SQLite at the edge. SQLite Cloud wrapped Raft-replicated SQLite in a hosted service. The category went from "a thing you run" to "a thing you buy" in roughly two years. rqlite, dqlite, and Litestream remain open source and free. O'Toole's account is notable for what it does not claim. He names Expensify's Bedrock (Paxos-based, internal before 2016) and ActorDB (Erlang, per-actor replication) as contemporaries, conceding the idea was in the air. He credits Ben Johnson's Litestream argument — that most apps need cheap recovery, not a cluster — as correct for most applications. He records that Mark Shuttleworth called him directly when Canonical was exploring options, but the fit was wrong: dqlite needed a library, rqlite was a server. The structural pattern here is familiar from every infrastructure layer: the open-source proof-of-concept establishes that a thing works, a half-dozen teams re-derive the idea with different trade-offs, and the venture-funded entrants capture the revenue by wrapping the proven architecture in a managed service. The pioneers keep the users who want to understand and control their own stack. The platforms get everyone else. What distinguishes rqlite twelve years in is the clarity of its CP commitment: writes go through the leader, a majority must acknowledge, a partitioned minority refuses writes. That decision, made in 2014, was never revisited. It is the design choice that makes rqlite the thing you reach for when you need a fault-tolerant relational store you can run yourself, understand fully, and owe nobody for. O'Toole says he is content with that position. The record suggests he should be — he wrote a category into existence and it is still his name at the origin of the timeline, even if the timeline most people encounter starts in 2021.
