redb

A simple, portable, high-performance, ACID, embedded key-value store.

redb is written in pure Rust and is loosely inspired by lmdb. Data is stored in a collection of copy-on-write B+trees. For more details, see the design doc

use redb::{Database, Error, ReadableDatabase, TableDefinition};

const TABLE: TableDefinition<&str, u64> = TableDefinition::new("my_data");

fn main() -> Result<(), Error> {
    let db = Database::create("my_db.redb")?;
    let write_txn = db.begin_write()?;
    {
        let mut table = write_txn.open_table(TABLE)?;
        table.insert("my_key", &123)?;
    }
    write_txn.commit()?;

    let read_txn = db.begin_read()?;
    let table = read_txn.open_table(TABLE)?;
    assert_eq!(table.get("my_key")?.unwrap().value(), 123);

    Ok(())
}

Status

Stable and maintained.

The file format is stable, and a reasonable effort will be made to provide an upgrade path if there are any future changes to it.

Features

• Zero-copy, thread-safe, BTreeMap based API
• Fully ACID-compliant transactions
• MVCC support for concurrent readers & writer, without blocking
• Crash-safe by default
• Savepoints and rollbacks
• Multi-process access to one database, with a single or several writer processes

no_std

Turning off the std feature, with experimental-api-5 on, builds redb without the standard library. It needs alloc, panic = "abort", and a target with atomic compare-and-swap -- Cortex-M3 and above, but not Cortex-M0. There is no filesystem, so storage is supplied by implementing StorageBackend and passing it to Builder::create_with_backend(). The cache_metrics, chrono_v0_4 and uuid features are unavailable.

Development

just and rootless podman are required. Toolchain setup, dependency vendoring, and cargo-deny checks may use network access without compiling project dependencies. Other development recipes continue to run on the host. Cargo build artifacts are cached in the redb-sandbox-target Podman volume; remove that volume to force a clean sandbox build.

Benchmarks

redb has similar performance to other top embedded key-value stores such as lmdb and rocksdb

	redb	lmdb	rocksdb	fjall	sqlite
bulk load	17063ms	9232ms	13969ms	18619ms	15341ms
individual writes	920ms	1598ms	2432ms	3488ms	7040ms
batch writes	1595ms	942ms	451ms	353ms	2625ms
len()	0ms	0ms	749ms	1181ms	30ms
random reads	1138ms	637ms	2911ms	2177ms	4283ms
random reads	934ms	631ms	2884ms	2357ms	4281ms
random range reads	1174ms	565ms	2734ms	2564ms	8431ms
random range reads	1173ms	565ms	2742ms	2690ms	8449ms
random reads (4 threads)	1390ms	840ms	3995ms	2606ms	7000ms
random reads (8 threads)	757ms	427ms	2147ms	1352ms	8123ms
random reads (16 threads)	652ms	216ms	1478ms	963ms	23022ms
random reads (32 threads)	410ms	125ms	1100ms	576ms	26536ms
removals	23297ms	10435ms	6900ms	6004ms	10323ms
uncompacted size	4.00 GiB	2.61 GiB	893.18 MiB	1000.95 MiB	1.09 GiB
compacted size	1.69 GiB	1.26 GiB	454.71 MiB	1000.95 MiB	556.85 MiB

Source code for benchmark here. Results collected on a Ryzen 9950X3D with Samsung 9100 PRO NVMe.

License

Licensed under either of

• Apache License, Version 2.0
• MIT License

at your option.

Contribution

Unless you explicitly state otherwise, any contribution intentionally submitted for inclusion in the work by you, as defined in the Apache-2.0 license, shall be dual licensed as above, without any additional terms or conditions.