btrdb - B-tree DataBase

btrdb is a NoSQL database engine with B-tree Copy-on-Write mechanism inspired by btrfs.

• Deno runtime
• Node.js runtime
• Single file
• B-tree Copy-on-Write (reference paper, slides)
• Good performance even written in pure TypeScript
  • Set 100k key-value pairs in 1.2s
  • Insert 100k documents in 2.3s
• Snapshots
  • Named snapshots
• Key-Value sets
• Document sets
  • Indexes
  • Query functions
  • Query tagged template parser
  • Serialize to "binval" format on disk
  • Binary data value support
• ACID
  • Readers/writer lock
  • Isolation with concurrent reader on snapshots
• Auto-commit
• Space reclamation with refcount tree
• Client / Server (?)
• Replication (?)

⚠️ Warning ⚠️

This project is just started. It's under heavy development!

The on-disk format structure and the API are NOT stable yet.

Please do NOT use it in any serious production.

btrdbfs

btrdbfs is a project to run filesystem on btrdb using FUSE.

Getting Started

Import the module

Deno:

import { Database } from "https://deno.land/x/btrdb/mod.ts";

Node.js:

Install from NPM registry:

npm i @yuuza/btrdb

Import from ES module:

import { Database } from "@yuuza/btrdb";

Require from CommonJS module:

const { Database } = require("@yuuza/btrdb");

Create/open database file

const db = new Database();
await db.openFile("data.db");
// Will create new database if the file doesn't exist.

Key-value set

const configSet = await db.createSet("config");
// Get the set or create if not exist.

await configSet.set("username", "yuuza");
console.info(await configSet.get("username")); // "yuuza"

await db.commit();
// Commit to persist the changes.

Use document set

Create set

interface User {
  id: number; // A property named "id" is required.
  username: string;
  status: "online" | "offline";
}

const userSet = await db.createSet<User>("users", "doc");
// Get the set or create if not exist.

Insert

await userSet.insert({ username: "yuuza", status: "offline" });
// Insert a new document, auto id when it's not specified.

console.info(await userSet.get(1));
// { id: 1, username: "yuuza", status: "offline" }

await db.commit();
// Commit to persist the changes.

Upsert

upsert will update the document with the same id, or insert a new document if the id does not exist.

const user = await userSet.get(1);
user.status = "online";
// Get user and set its status

await userSet.upsert(user);
// Use upsert to apply the change.

console.info(await userSet.get(1));
// { id: 1, username: "yuuza", status: "online" }

await db.commit();
// Commit to persist the changes.

Indexes

interface User {
  id: number;
  username: string;
  status: "online" | "offline";
  role: "admin" | "user";
}

const userSet = await db.createSet<User>("users", "doc");

// Define indexes on the set and update indexes if needed.
await userSet.useIndexes({
  status: (u) => u.status,
  // define "status" index, which indexing the value of user.status for each user in the set

  role: (user) => user.role,

  username: { unique: true, key: (u) => u.username },
  // define "username" unique index, which does not allow duplicated username.

  onlineAdmin: (u) => u.status == "online" && u.role == "admin",
  // define "onlineAdmin" index, the value is a computed boolean.
});

await userSet.insert({ username: "yuuza", status: "online", role: "user" });
await userSet.insert({ username: "foo", status: "offline", role: "admin" });
await userSet.insert({ username: "bar", status: "online", role: "admin" });
await db.commit();

// Get all online users
console.info(await userSet.findIndex("status", "online"));
// [
//   { username: "yuuza", status: "online", role: "user", id: 1 },
//   { username: "bar", status: "online", role: "admin", id: 3 }
// ]

// Get all users named 'yuuza'
console.info(await userSet.findIndex("username", "yuuza"));
// [ { username: "yuuza", status: "online", role: "user", id: 1 } ]

// Get all online admins
console.info(await userSet.findIndex("onlineAdmin", true));
// [ { username: "bar", status: "online", role: "admin", id: 3 } ]

Query (tagged template)

Querying on indexes is supported.

Queries can be created from the query tagged template parser for better readability.

Operators: ==, !=, >, <, <=, >=, AND, OR, NOT, SKIP, LIMIT, (, )

Always use ${} to pass values.

// Get all offline admins
console.info(
  await userSet.query(query`
    status == ${"offline"}
    AND role == ${"admin"}
  `),
);
// [ { username: "foo", status: "offline", role: "admin", id: 2 } ]

// Get all online users, but exclude id 1.
console.info(
  await userSet.query(query`
    status == ${"online"}
    AND NOT id == ${1}
  `),
);
// [ { username: "bar", status: "online", role: "admin", id: 3 } ]

Query (functions)

Query functions: EQ (==), NE (!=), LT (<), GT (>), LE (<=), GE (>=), AND, OR, NOT, SLICE.

// Get all offline admins
console.info(
  await userSet.query(AND(EQ("status", "offline"), EQ("role", "admin"))),
);
// [ { username: "foo", status: "offline", role: "admin", id: 2 } ]

// Get all online users, but exclude id 1.
console.info(
  await userSet.query(
    AND(
      EQ("status", "online"),
      NOT(EQ("id", 1)), // "id" is a special "index" name
    ),
  ),
);
// [ { username: "bar", status: "online", role: "admin", id: 3 } ]

Transactions

Database.runTransaction(async () => { ... }) could be used for auto commiting and rolling back.

It guarantees:

• The promise is resolved when it committed.
• Other transactions could be concurrently executed.
• Only commits when all transactions are completed.
• Rollback when any transaction is failed, and rerun other successful concurrent transactions.

The transaction function might be re-run in case of replaying.

Snapshots

Since btrdb uses CoW mechanism and never overwrites data on-disk, creating "snapshot" have almost no cost.

const dataSet = await db.createSet("data");
await dataSet.set("foo", "bar");

// Commit then create a "named snapshot"
await db.createSnapshot("backup");

await dataSet.set("someone", "messed up your data!");
await dataSet.set("foo", "no bar!");
await db.commit();

// Get a "named snapshot".
const snap = await db.getSnapshot("backup");

// Read data from the snapshot
console.info(await snap.getSet("data").get("foo"));

More example in the test code

See test.ts.

Design

(Outdated. To be added: documents tree, indexes tree, data pages, named snapshots)

License

MIT License