Drizzle + Postgres in Production: Migrations, Pools, and Pitfalls

Why Drizzle on MVPs

Stay close to SQL. TypeScript inference is good enough. Migrations are reviewable. Teams can hire for Postgres, not ORM dialect.

Folder layout

db/
  schema/
    users.ts
    orders.ts
    index.ts      # barrel export
  migrations/
  client.ts       # single db instance pattern

// schema/users.ts
export const users = pgTable("users", {
  id: uuid("id").defaultRandom().primaryKey(),
  email: text("email").notNull().unique(),
  createdAt: timestamp("created_at", { withTimezone: true })
    .defaultNow()
    .notNull(),
});

Request path on Vercel

Migrations in CI

1. drizzle-kit generate in PR — human reads SQL
2. Apply to staging DB in CI
3. Prod apply in deploy step or manual gate
4. Never edit applied migration files

Serverless connection rules

Pitfalls we hit

1. N+1 in RSC — batch with inArray or joins
2. Missing FK indexes — filter columns need indexes
3. timestamp without TZ — use timestamptz
4. Transactions in Server Actions — know your isolation level

TL;DR

Drizzle rewards migration discipline early. Cheap at week 2, priceless at month 6.