RoughDB

Embedded key-value storage written in Rust — a port of LevelDB to Rust.

RoughDB is an LSM-tree key-value store with a LevelDB-compatible on-disk format. It supports persistent (WAL + MANIFEST + SSTables) and in-memory operation, multi-level compaction, and bidirectional iteration.

Getting started

cargo add roughdb

Opening a database

use roughdb::{Db, Options};

let mut opts = Options::default();
opts.create_if_missing = true;

let db = Db::open("/tmp/my_db", opts)?;

Use Db::default() for a lightweight in-memory database (no WAL, no flush):

let db = Db::default();

Basic reads and writes

use roughdb::{ReadOptions, WriteOptions};

// Write
db.put(b"hello", b"world")?;
db.delete(b"hello")?;

// Read
match db.get(b"hello") {
    Ok(value) => println!("got: {}", String::from_utf8_lossy(&value)),
    Err(e) if e.is_not_found() => println!("not found"),
    Err(e) => return Err(e),
}

Atomic batch writes

use roughdb::{WriteBatch, WriteOptions};

let mut batch = WriteBatch::new();
batch.put(b"key1", b"val1");
batch.put(b"key2", b"val2");
batch.delete(b"old_key");

db.write(&WriteOptions::default(), &batch)?;

Iterating over keys

Iterators start unpositioned — call seek_to_first, seek_to_last, or seek before reading.

use roughdb::ReadOptions;

let mut it = db.new_iterator(&ReadOptions::default())?;

// Forward
it.seek_to_first();
while it.valid() {
    println!("{:?} = {:?}", it.key(), it.value());
    it.next();
}

// Backward
it.seek_to_last();
while it.valid() {
    println!("{:?}", it.key());
    it.prev();
}

// Range scan from "b" onward
it.seek(b"b");
while it.valid() {
    println!("{:?}", it.key());
    it.next();
}

Snapshots

A snapshot pins the database to a specific point in time; reads through it see only writes that preceded the snapshot.

use roughdb::ReadOptions;

let snap = db.get_snapshot();

// Reads with this snapshot see the state at the moment it was taken.
let opts = ReadOptions { snapshot: Some(snap), ..ReadOptions::default() };
let value = db.get_with_options(&opts, b"key")?;
let mut it = db.new_iterator(&opts)?;

// Release when no longer needed (allows compaction to reclaim old data).
db.release_snapshot(snap);

Manual compaction

// Compact everything
db.compact_range(None, None)?;

// Compact a specific key range
db.compact_range(Some(b"aaa"), Some(b"zzz"))?;

Building and testing

cargo build
cargo test
cargo clippy
cargo bench      # Criterion benchmarks (benches/db.rs)

Status

RoughDB is in active development. The on-disk format is LevelDB-compatible. The following are not yet implemented:

• Compression (Snappy, Zstd) — all blocks written uncompressed
• Db::get_property / Db::get_approximate_sizes
• DestroyDB
• LOCK file (multi-process safety)
• Block cache, table cache, Bloom filters

License

Apache License 2.0 — see LICENSE.