Drizzle Orm Rules

Drizzle ORM Rules Skill

Schema Design

• Use integer('id').primaryKey().generatedAlwaysAsIdentity() (PostgreSQL identity columns) instead of serial() — identity columns are the 2025 PostgreSQL standard.
• Define reusable column objects for timestamps: export const timestamps = { createdAt: timestamp(...).defaultNow().notNull(), updatedAt: timestamp(...).$onUpdateFn(() => new Date()) }.
• Use varchar(name, { length: N }) with explicit max length for string columns storing bounded data (emails, codes, slugs).
• Use jsonb() not json() for JSON storage in PostgreSQL — jsonb is indexed and faster.
• Always call .notNull() on columns that must not be nullable.

Indexing

• Define indexes inside pgTable's second argument callback: (table) => [index('name').on(table.col)].
• Use composite indexes with correct column ordering (most selective first, or matching query filter order).
• Use uniqueIndex() for unique constraints on single or combined columns.
• For full-text search, use .withSearchIndex or a GIN index via raw SQL migration.

Queries

• Prefer db.query.<table>.findMany({ with: { relation: true } }) (relational API) for typed nested joins.
• Use db.select().from(table).where(eq(table.col, val)) for flat queries.
• Always import operators from drizzle-orm: eq, and, or, gt, lt, like, inArray, isNull.
• Use db.transaction(async (tx) => {...}) for multi-step writes that must be atomic.
• Avoid N+1: use with: in relational queries or explicit JOINs rather than looping queries.

Migrations

• Local development: drizzle-kit push (fast, no migration files) — never for production.
• Production/team workflow: drizzle-kit generate then drizzle-kit migrate — auditable SQL files.
• Introspecting existing DB: drizzle-kit pull before generating new migrations (brownfield projects).
• Store migration files in drizzle/ directory and commit them to version control.
• Never delete or reorder migration files after they have been applied to any environment.

Relations

• Define explicit relations() alongside table definitions in schema.ts.
• Use one() for many-to-one references and many() for one-to-many or many-to-many.
• Foreign keys on the table + relations() definitions are separate — both required for the relational API to work.

// Reusable timestamp columns export const timestamps = { createdAt: timestamp('created_at', { mode: 'date', withTimezone: true }).defaultNow().notNull(), updatedAt: timestamp('updated_at', { mode: 'date', withTimezone: true }) .defaultNow() .notNull() .$onUpdateFn(() => new Date()), };

export const users = pgTable('users', { id: integer('id').primaryKey().generatedAlwaysAsIdentity(), // NOT serial email: varchar('email', { length: 320 }).notNull().unique(), name: text('name').notNull(), meta: jsonb('meta'), // jsonb, not json ...timestamps, }, (table) => [ index('users_email_idx').on(table.email), ]);

export const posts = pgTable('posts', { id: integer('id').primaryKey().generatedAlwaysAsIdentity(), userId: integer('user_id').notNull().references(() => users.id), title: varchar('title', { length: 500 }).notNull(), ...timestamps, }, (table) => [ index('posts_user_id_idx').on(table.userId), ]);

// Relations (required for relational query API) export const usersRelations = relations(users, ({ many }) => ({ posts: many(posts), })); export const postsRelations = relations(posts, ({ one }) => ({ user: one(users, { fields: [posts.userId], references: [users.id] }), }));

// src/lib/db/queries.ts — typed relational query import { db } from './client'; import { eq } from 'drizzle-orm'; import { users } from './schema';

export async function getUserWithPosts(userId: number) { return db.query.users.findFirst({ where: eq(users.id, userId), with: { posts: true }, // nested join — no N+1 }); }

// Atomic transaction example export async function transferData(fromId: number, toId: number, amount: number) { return db.transaction(async (tx) => { await tx.update(accounts).set({ balance: sqlbalance - ${amount} }).where(eq(accounts.id, fromId)); await tx.update(accounts).set({ balance: sqlbalance + ${amount} }).where(eq(accounts.id, toId)); }); }

</examples>

## Iron Laws

1. **ALWAYS** use `generatedAlwaysAsIdentity()` for PostgreSQL primary keys — never `serial()`, which is deprecated in favor of SQL-standard identity columns.
2. **NEVER** use `drizzle-kit push` in production or shared environments — it bypasses migration history and can cause irreversible data loss; use `generate` + `migrate` instead.
3. **ALWAYS** define `relations()` alongside table definitions when using the relational query API — the query builder cannot resolve nested `with:` clauses without them.
4. **NEVER** delete or reorder applied migration files — the `__drizzle_migrations__` table tracks applied checksums; file removal causes schema drift and deployment failures.
5. **ALWAYS** import query operators (`eq`, `and`, `or`, `gt`, `inArray`, etc.) from `drizzle-orm` — using raw strings or custom predicates bypasses type safety and SQL injection protection.

## Anti-Patterns

| Anti-Pattern | Why It Fails | Correct Approach |
| --- | --- | --- |
| Using `serial()` for primary keys | `serial` is a PostgreSQL pseudo-type implemented via sequences; deprecated since PG 10 in favor of SQL-standard identity columns | Use `integer('id').primaryKey().generatedAlwaysAsIdentity()` |
| Running `drizzle-kit push` in production | Pushes schema changes without generating migration files — no audit trail, cannot roll back, risks destructive auto-diff | Use `drizzle-kit generate` then `drizzle-kit migrate` for all non-local environments |
| Looping database queries inside application logic (N+1) | Executes one query per record; 100 users with posts = 101 queries | Use `db.query.users.findMany({ with: { posts: true } })` to fetch nested data in a single optimized query |
| Omitting `relations()` but using relational query API | Drizzle throws runtime errors when `with:` keys are not mapped via `relations()` | Define `relations()` for every table that participates in relational queries |
| Using `json()` instead of `jsonb()` for JSON columns | `json` stores raw text, cannot be indexed; `jsonb` stores binary, supports GIN indexes and faster operations | Replace `json()` with `jsonb()` for all PostgreSQL JSON columns |

## Memory Protocol (MANDATORY)

**Before starting:**

```bash
cat .claude/context/memory/learnings.md

After completing: Record any new patterns or exceptions discovered.

ASSUME INTERRUPTION: Your context may reset. If it's not in memory, it didn't happen.