prisma

Ensures Prisma code is type-safe, performant, and architecturally clean — covering the non-obvious patterns that trip up even experienced developers.

Core Principles

• One PrismaClient per process. Multiple instances create multiple connection pools and exhaust your database connection limit.
• Enums vs client imports. Import from the generated enums.ts in client/shared code; import from client.ts only in server-side code. Never import client.ts in client components.
• Repository pattern. Keep data access logic separate from business logic. Create repository modules for complex query sets.
• Transactions for multi-step writes. Any operation that modifies more than one record should use $transaction.
• Never modify existing migrations. Treat migrations as append-only history.

When to Use

• Writing or reviewing Prisma schema models
• Building or refactoring a database access layer
• Debugging N+1 queries or performance regressions
• Setting up Prisma in a new project (especially Next.js)

Implementation Guide

Client Setup (Singleton)

The hot-reload guard is required in Next.js — without it, dev mode creates a new PrismaClient on every file change and exhausts connections:

// lib/prisma.ts
import { PrismaClient } from "@prisma/client";

const globalForPrisma = globalThis as unknown as { prisma: PrismaClient };

export const prisma =
  globalForPrisma.prisma ?? new PrismaClient();

if (process.env.NODE_ENV !== "production")
  globalForPrisma.prisma = prisma;

Enums vs Client Imports

// ✅ Client component or shared validation code
import { Role, Status } from "@prisma/client/enums";

// ✅ Server-side only (API routes, server components, background jobs)
import { prisma } from "@/lib/prisma";
import type { User } from "@prisma/client";

// ❌ Never in client components
import { PrismaClient } from "@prisma/client";

Schema Design

• Use domain-driven model names. Keep schemas normalized and DRY.
• Declare all relations explicitly with @relation.
• Implement soft delete via deletedAt DateTime? — never hard-delete records that may be referenced.
• Use Prisma's native type decorators for database-level precision.

Queries & Performance

• N+1: Use nested include or select instead of looping with separate queries.
• Pagination: Use take and skip; add a cursor-based approach for large datasets.
• Field selection: Use select to fetch only needed fields; avoid over-fetching with blanket include.
• Join strategy: relationLoadStrategy ("join" or "query") is available but requires enabling the relationJoins preview feature flag. "join" uses a single LATERAL JOIN; "query" sends one query per table and joins at the application level.

Error Handling

Catch Prisma-specific errors at the repository boundary:

• PrismaClientKnownRequestError — structured DB errors. Check error.code:
  • P2002 — unique constraint violation
  • P2025 — record not found (replaces the removed NotFoundError from Prisma 5)
• PrismaClientUnknownRequestError — unstructured DB errors
• PrismaClientValidationError — invalid query shape (usually a type error)

Provide user-friendly messages upstream; log the full error with context for debugging.

Migrations

• Use descriptive names: prisma migrate dev --name add_user_email_index.
• Review generated SQL before applying to production.
• Never edit existing migration files — create a new migration instead.
• Keep migrations idempotent where possible.

Review Checklist

• Is there a single PrismaClient instance with the hot-reload guard in place?
• Are enums imported from enums.ts in any client or shared code?
• Do multi-record writes use $transaction?
• Are N+1 patterns avoided (no DB calls inside loops)?
• Are PrismaClientKnownRequestError codes handled at the repository boundary?
• Do migrations have descriptive names and remain unmodified after creation?