In September 2014, Philip O’Toole published rqlite, combining SQLite with Raft consensus replication to ensure data survived node failures. According to O’Toole, it was the first open-source system to merge Raft and SQLite. “Nobody else was doing it, and for several years nobody else tried,” he writes. Today, the distributed SQLite category includes a dozen projects, several venture-funded and two acquired.
Early alternatives existed but remained niche. Expensify had replicated SQLite internally using Paxos since before 2014 and released Bedrock in 2016. ActorDB, an Erlang system pairing SQLite with Raft, emerged around the same time as rqlite but neither gained wide adoption outside their origins. rqlite’s contribution was making distributed SQLite a standalone, general-purpose, open-source offering.
Canonical built dqlite in 2017 to embed a replicated store in LXD and later MicroK8s without running etcd. O’Toole notes the key difference: dqlite is a library to embed; rqlite is a server to run. Both used consensus-based approaches prioritizing consistency over availability.
Ben Johnson’s Litestream, released in early 2021, took a different path. Rather than clustering, it watches SQLite’s write-ahead log and streams it to S3 for single-node disaster recovery at minimal cost. O’Toole credits Johnson’s “SQLite is enough” argument with influencing subsequent years of discussion. Fly.io hired Johnson in 2022 and shipped LiteFS, replicating SQLite at the filesystem level using Consul leases—the first attempt to make replicated SQLite a platform feature.
By 2022, the design space had three corners. Consensus-based systems like rqlite and dqlite prioritized strong consistency with a single leader. Streaming approaches like Litestream offered cheap recovery for single writers. Merge-based systems like cr-sqlite added CRDTs, enabling multiple writers with eventual consistency—suited for local-first applications.
ChiselStrike forked SQLite itself to make engine-level changes the SQLite project wouldn’t accept, eventually becoming Turso, which Supabase acquired in 2026. SQLite Cloud emerged offering hosted Raft-replicated SQLite. Cloudflare announced D1 in 2022, offering serverless SQLite in their edge network.
O’Toole identifies three observations about the twelve-year arc: the core approach—SQLite with external replication—has proven robust enough for at least eight independent teams to reinvent it; the design space correctly splits into three approaches for different use cases; and commercial success concentrated in platforms offering managed services rather than open-source tools. “The idea was open source for eight years before anyone sold it,” he notes. rqlite remains available as a free, self-hosted option.
Key facts
- rqlite, combining SQLite with Raft consensus, launched in September 2014 as the first open-source distributed SQLite system
- The distributed SQLite category now includes a dozen projects; several are venture-funded and two have been acquired
- Litestream (2021) introduced streaming-based replication to S3 for single-node recovery, differing from consensus approaches
- Turso (formerly ChiselStrike) forked SQLite to enable engine-level changes and was acquired by Supabase in 2026
- Cloudflare’s D1 and Turso’s service represent the shift from self-hosted tools to managed platform offerings
- Three design approaches emerged: consensus-based (rqlite, dqlite), streaming-based (Litestream), and merge-based (cr-sqlite)