rwsdk Database DO

Overview

rwsdk/db provides SQLite databases using Durable Objects with type-safe queries via Kysely. Each database instance runs in isolated Durable Objects with schema types inferred directly from migrations. No code generation needed - TypeScript understands your schema automatically.

When to Use

Use when:

• Building Cloudflare Workers apps needing SQL databases
• Want runtime database creation with minimal setup
• Need isolated database instances (multi-tenant, modular components)
• Prefer SQL query builder over ORMs
• Working with rwsdk framework

Don't use when:

• Need globally distributed low-latency reads (single location limitation)
• Require production-grade backup features (preview feature)
• Complex ORM requirements outweigh simplicity benefits

Quick Setup (5 Files)

1. Define Migrations (db/migrations.ts)

import { type Migrations } from 'rwsdk/db';

export const migrations = {
  '001_initial_schema': {
    async up(db) {
      await db.schema
        .createTable('todos')
        .addColumn('id', 'text', (col) => col.primaryKey())
        .addColumn('text', 'text', (col) => col.notNull())
        .addColumn('completed', 'integer', (col) => col.notNull().defaultTo(0))
        .addColumn('createdAt', 'text', (col) => col.notNull())
        .execute();
    },
    async down(db) {
      // CRITICAL: Use .ifExists() for SQLite's non-transactional DDL
      await db.schema.dropTable('todos').ifExists().execute();
    },
  },
} satisfies Migrations;

2. Create Database Instance (db/db.ts)

import { env } from 'cloudflare:workers';
import { type Database, createDb } from 'rwsdk/db';
import { type migrations } from '@/db/migrations';

// Export inferred types
export type AppDatabase = Database<typeof migrations>;
export type Todo = AppDatabase['todos'];

// Create database instance
export const db = createDb<AppDatabase>(
  env.APP_DURABLE_OBJECT,
  'todo-database' // unique key for this instance
);

3. Create Durable Object Class (db/durableObject.ts)

import { SqliteDurableObject } from 'rwsdk/db';
import { migrations } from '@/db/migrations';

export class AppDurableObject extends SqliteDurableObject {
  migrations = migrations;
}

4. Export from Worker (worker.ts)

export { AppDurableObject } from '@/db/durableObject';
// ... rest of worker code

5. Configure Wrangler (wrangler.toml)

[durable_objects]
bindings = [
  { name = "APP_DURABLE_OBJECT", class_name = "AppDurableObject" }
]

[[migrations]]
tag = "v1"
new_sqlite_classes = ["AppDurableObject"]

Type Inference

Types are automatically inferred from migrations - no code generation:

// TypeScript knows all columns and types
const user = await db.selectFrom('users').selectAll().executeTakeFirst();
// Type: { id: string, username: string } | undefined

CRUD Operations Reference

Joins and Nested Data

Use Kysely helpers for ORM-like nested results:

import { jsonObjectFrom } from 'kysely/helpers/sqlite';

const posts = await db
  .selectFrom('posts')
  .selectAll('posts')
  .select((eb) => [
    jsonObjectFrom(
      eb
        .selectFrom('users')
        .select(['id', 'username'])
        .whereRef('users.id', '=', 'posts.userId')
    ).as('author'),
  ])
  .execute();

// Result: [{ id: "post-123", title: "...", author: { id: "...", username: "..." } }]

Database Seeding

Create seed script with default export:

// scripts/seed.ts
import { db } from '@/db/db';

export default async () => {
  console.log('🌱 Seeding...');
  await db.deleteFrom('todos').execute();

  await db
    .insertInto('todos')
    .values([
      {
        id: crypto.randomUUID(),
        text: 'First todo',
        completed: 0,
        createdAt: new Date().toISOString(),
      },
    ])
    .execute();

  console.log('✓ Seeded');
};

Add to package.json:

{
  "scripts": {
    "seed": "rwsdk worker-run ./src/scripts/seed.ts"
  }
}

Run: npm run seed

Migration Timing

Migrations run when createDb() is called:

• Development: On dev server start
• Production: During deployment (initial request triggers migration)

Migration Failures and Rollback

CRITICAL: SQLite doesn't support transactional DDL. Failed migrations can leave partial state.

Automatic behavior: Failed up() triggers down() automatically

Write defensive down() functions:

async down(db) {
  // Always use .ifExists() in case up() partially failed
  await db.schema.dropTable("posts").ifExists().execute();
  await db.schema.dropTable("users").ifExists().execute();
}

Why auto-rollback: Returns database to known-good state since manual intervention isn't feasible in isolated runtime environments.

Common Mistakes

API Quick Reference

createDb()

createDb<T>(
  durableObjectNamespace: DurableObjectNamespace,
  key: string
): Database<T>

Migration structure

{
  "001_migration_name": {
    async up(db) { /* create schema */ },
    async down(db) { /* undo with .ifExists() */ }
  }
} satisfies Migrations

Type exports

export type AppDatabase = Database<typeof migrations>;
export type TableName = AppDatabase['tableName'];

Performance Notes

• Latency: Single-location execution (not globally distributed)
• Isolation: Perfect for modular components, multi-tenant apps
• Scale: Can create multiple instances with different keys for geographic distribution
• Backups: No built-in backup features (implement external replication for critical data)

Complete Kysely Documentation

For full query builder capabilities, see Kysely docs. Everything in Kysely works with rwsdk/db.

Real-World Patterns

Multiple database instances:

const userDb = createDb(env.DO, 'users');
const analyticsDb = createDb(env.DO, 'analytics'); // isolated

Type-safe table access:

type User = AppDatabase['users']; // Inferred from migrations
const users: User[] = await db.selectFrom('users').selectAll().execute();

Conditional queries:

let query = db.selectFrom('posts').selectAll();
if (userId) {
  query = query.where('userId', '=', userId);
}
const posts = await query.execute();