Database engine
Strata
A relational database, from the page up.
A disk-backed B+Tree with a buffer pool, an LSM tree with compaction, a binary write-ahead log with redo/undo recovery, snapshot-isolation MVCC, a two-phase lock manager with deadlock detection, a hand-written SQL parser, a cost-based optimizer with join reordering, a vectorized executor, Raft consensus and a shard router.
nanosql is the single, explicit exception to the isolation rule: it imports all ten as real package dependencies and turns them into a transactional SQL engine with a REPL and a DB-API driver. No SQLite, DuckDB or Postgres anywhere in the stack — not even in the tests.
- Repositories
- 11
- Lines
- 27,698
- Commits
- 226
- Test files
- 86
- Built
- Jul 2026
- Mostly
- Python
L1 Distribution & HA
L2 Query engine
L3 Transactions
L4 Storage & persistence
↳ Flagship
↳ Flagship
Where it all comes together
L1
Distribution & HA
L2
Query engine
L3
Transactions
L4