SecantusDB Rust DB
A whole second server, written in Rust from the wire up — dispatch, cursors, change streams, and WiredTiger storage — with no Python in the request path. Same wire protocol, same conformance gauges, a single standalone binary.
Why a second engine in Rust
The pure-Python server is the reference — the broadest feature surface and the easiest to read and extend. The Rust server is the same database with a different runtime underneath, built for speed and standalone deployment.
One standalone binary
The Rust server ships as a single executable with WiredTiger statically linked — secantusd-rs. No interpreter, no virtualenv, no system Mongo. Drop it in a container or a CI image and run.
No Python in the request path
Wire parsing, command dispatch, cursors, the operator engines, and storage are all pure Rust. The GIL never touches a query, and the accept loop runs on a native thread.
Same wire protocol
It answers the same OP_MSG handshake, CRUD, aggregation, and change-stream commands. Every driver that talks to the Python server talks to the Rust server unchanged.
Same conformance bar
The unmodified driver-conformance gauges run against the Rust server too — the same upstream suites the driver maintainers run against a real mongod. It passes 99.5% of pymongo's own test suite, level with the Python server; the feature comparison maps both servers against real MongoDB, feature by feature.
Real WiredTiger storage
The same C storage engine MongoDB ships with, linked into the binary. B-trees, durability, write-ahead logging — identical on-disk semantics to the Python server.
Embeddable from Python too
A thin lifecycle handle (start / stop / address) runs the Rust accept loop in-process on a GIL-released thread; pymongo connects over real TCP. Launch it from a test, get a Rust server.
Measured, not promised
Six workloads, median of five runs, end-to-end through pymongo on on-disk WiredTiger — real mongod as the reference. The Rust server runs at 2.1×–4.5× of mongod per operation, and ~2.7×–5.2× faster than the Python server workload-for-workload.
| Workload | mongod | Rust server | ×mongod |
|---|---|---|---|
| insert (10k docs) | 55.6 ms | 118.8 ms | 2.1× |
| find indexed range | 4.3 ms | 10.1 ms | 2.4× |
| find full scan | 7.7 ms | 34.8 ms | 4.5× |
| update_many (half) | 35.2 ms | 93.0 ms | 2.6× |
aggregate $group | 5.8 ms | 24.6 ms | 4.3× |
| delete_many (half) | 21.6 ms | 80.5 ms | 3.7× |
Apple Silicon, dataset of 10,000 small docs, every number includes the wire + driver overhead a real client pays. Full benchmark & methodology →
Get it
Two ways to run the Rust server — bundled in the Python package, or as a prebuilt standalone binary.
Bundled in the wheel
The secantus wheel ships the Rust server as the secantusd-rs command, alongside the pure-Python secantusdb.
pip install SecantusDB
secantusd-rs --port 27017 --storage-path ./dataStandalone binary
Prebuilt static-WiredTiger archives are published on GitHub Releases (Linux x86_64 and macOS arm64) — no Python at all. Look for the secantusdb-v<version> tags; each archive ships with a .sha256 checksum.
# download + extract from the releases page,
# then run the secantusd-rs binary:
./secantusd-rs --port 27017 --storage-path ./dataPython or Rust?
Same database, same wire protocol, same storage. The runtime is the choice.
| Python server | Rust server | |
|---|---|---|
| Request path | Pure Python | Pure Rust |
| Best for | Hackability, the widest feature surface, the reference | Speed and a standalone, dependency-free binary |
| How you run it | pip install SecantusDB → SecantusDBServer(...) | secantusd-rs binary (bundled in the wheel or from Releases) |
| Python required at runtime | Yes | No |
| Storage engine | WiredTiger | WiredTiger |
| Wire protocol & drivers | MongoDB OP_MSG — all drivers | MongoDB OP_MSG — all drivers |
| Embeddable in Python | Native (in-process) | Yes — in-process on a GIL-released thread |
Rust-server documentation: Installation · Running the daemon · Embedded in Python · Backup & PITR · Conformance · Feature comparison