Database Workflow

Source of truth: db/schema.sql is the canonical schema.

Quick Commands

Workflow Decision Tree

Schema change needed?
├─ Early development / experimenting
│  └─ Edit db/schema.sql → just db-migrate → just db-schema
│
└─ Production / reviewable history needed
   └─ Edit db/schema.sql → just db-migration-new <name> → just db-migration-apply → just db-schema

Local Dev Setup

# 1. Start Postgres
docker compose up -d  # or: just setup

# 2. Apply schema
just db-migrate

# 3. Generate TypeScript types (after schema changes)
just db-schema

Schema Changes Checklist

1. Edit db/schema.sql
2. Apply locally: just db-migrate
3. Generate types: just db-schema
4. Update any affected services/routers
5. Run tests: pnpm test

Versioned Migrations (Atlas)

Even with versioned migrations, db/schema.sql stays the source of truth. Atlas uses it to infer changes.

# Create migration from schema diff
just db-migration-new add_user_roles

# Apply migrations
just db-migration-apply

# Check status
just db-migration-status

Conventions

• Use uuid primary keys
• Use created_at / updated_at timestamps
• Enforce tenancy at service layer (e.g., todos.user_id)
• Columns mapped to camelCase in TypeScript via Kysely plugin

Runtime DB Access

// packages/backend/core/src/db.ts
import { connectDB, getDB } from "@yourcompany/backend-core/db";

// At app startup (once)
await connectDB({ connectionString: config.DATABASE_URL });

// Anywhere after initialization
const db = getDB();
const users = await db.selectFrom("user").selectAll().execute();

Why Kysely (vs Drizzle/Prisma)

• Typed query builder without ORM "model" layer
• Works with SQL-first workflow (db/schema.sql) + codegen
• Easy to pass db dependency for testing

Gotchas

• Tests apply db/schema.sql into testcontainers Postgres
• After schema changes:
  • Update db/schema.sql
  • Re-apply locally (just db-migrate)
  • Re-generate TS types (just db-schema)
• Generated types: packages/backend/core/src/schema.ts