Drizzle ORM

Gotchas

• drizzle(client, { schema }) — omit { schema } and the db.query.* relational API silently does not exist.
• relations() is ORM-level metadata only and creates no constraint. Define references(() => users.id, { onDelete: 'cascade' }) as well; you need both.
• Types come from typeof table.$inferSelect / $inferInsert — never hand-write them.
• SQL-like API (db.select().from()) for joins and aggregates; relational API (db.query.x.findMany({ with: {...} })) for nested reads. insert/update/delete give back nothing usable without .returning().
• Inside db.transaction(async (tx) => ...) every call must use tx; a stray db call runs outside the transaction.
• Prepared statements need sql.placeholder('id') + .prepare('name'), then .execute({ id }).
• Partial selects (db.select({ id: users.id })) avoid loading unused columns; declare indexes in the table definition.

drizzle-kit

• generate → review the SQL in the out dir → migrate. push is dev-only. Never edit generated SQL; fix the schema and regenerate. check validates migration history consistency.
• A column rename generates DROP + ADD, i.e. data loss. Hand-write the migration file instead — ALTER TABLE users RENAME COLUMN old_name TO new_name; — then migrate applies it.
• A new non-nullable column takes two migrations: add nullable → backfill → add .notNull() → migrate again.
• The programmatic migrator needs a single connection: postgres(url, { max: 1 }), then migrate(drizzle(client), { migrationsFolder: './drizzle' }), then close it.
• drizzle.config.ts must supply dialect, schema, out, and dbCredentials.url before any command runs.

Docs: https://orm.drizzle.team/