db-rs

An ergonomic, embedded, single-threaded database for Rustaceans.

Strengths

• Define a schema in Rust.
• Use your types in the database as long as they implement Serialize and Deserialize. You don't have to fuss around with converting your data to database-specific types.
• All your database interactions are typesafe. When you type db., your tooling will suggest a list of your tables. When you select a table, you'll be greeted with that table-type's contract populated with your types. No need to wrap your db in a handwritten type safe contract.
• Supports a variety of simple data-structures, including LookupTables, Lists, and many more. Implementing your own table types is trivial.
• All table mutations are persisted to an append only log using the fast & compact bincode representation of your types.
• You can begin_transaction()s to express atomic updates to multiple tables.

Quickstart

Add the following to your Cargo.toml:

db-rs = "0.2.1"
db-rs-derive = "0.2.1"

Define your schema:

use db_rs_derive::Schema;
use db_rs::{Single, List, LookupTable};

#[derive(Schema)]
struct SchemaV1 {
    owner: Single<Username>,
    admins: List<Username>,
    users: LookupTable<Username, Account>,
}

Initialize your DB:

use db_rs::Db;
use db_rs::Config;

let mut db = SchemaV1::init(Config::in_folder("/tmp/test/"))?;
db.owner.insert("Parth".to_string())?;

println!("{}", db.owner.data().unwrap());

Table Types

Each table has an in-memory representation and a corresponding log entry format. For instance [List]'s in memory format is a [Vec], and you can look at it's corresponding [list::LogEntry] to see how writes will be written to disk.

Tables that start with Lookup have a HashMap as part of their in memory format. [LookupTable] is the most general form, while [LookupList] and [LookupSet] are specializations for people who want HashMap<K, Vec<V>> or HashMap<K, HashSet<V>>. Their reason for existence is better log performance in the case of small modifications to the Vec or HashSet in question (see [lookup_list::LogEntry] or [lookup_set::LogEntry]).

Log Compaction

At any point you can call [Db::compact_log] on your database. This will atomically write a compact representation of all your current tables. For example if there's a key in a LookupTable that was written to many times, the compact representation will only contain the last value. Each table type descibes it's own compact representation.

If your database is in an Arc<Mutex>> you can additionally use the [BackgroundCompacter] which will perform compactions periodically in a separate thread.

TXs and Batch Writing

You can [Db::begin_transaction] which will allow you to express batch operations that can be discarded as a set if your program is interrupted. Presently there is no way to abort a transaction. TXs are also a mechanism for batch writing, log entries are kept in memory until the transaction completes and written once to disk.

Active areas of thought and research

• Because the db implementation (like redis) is single threaded, it forces you to achieve application throughput via low latency rather than concurrency. Currently, this suits our needs. Simply being embedded gives us more than enough throughput compared to something like Postgres. For use in a server-style setting put the database in an Arc<Mutex<>>.
• The database offers no tools at the moment to define integrity constraints beyond what the Rust type system implicitly enforces (non-null for instance). At the moment for us, this is simply an application side concern.

Features

clone - derive clone on all table types. Consistency between cloned database is not provided. Useful in testing situations.

Used by

• Lockbook

License: BSD-3-Clause