Data access layer

Data Access Layer

Overview

Don't scatter ORM calls across components and route handlers. Put every query behind a server-only data-access layer (DAL): small typed functions that take an explicit orgId and never return another tenant's rows. The rest of the app calls getProjects(orgId), not db.select(...). This centralizes tenant scoping, caching, and the place you write tests.

When to use

• Any read or write to the database.
• You catch yourself importing db into a React component.
• You want one place to enforce "every query is scoped to the current org".

The pattern

// src/server/services/projects.ts
import "server-only";
import { and, eq, desc } from "drizzle-orm";
import { db } from "@/server/db/client";
import { projects } from "@/server/db/schema";

export async function listProjects(orgId: string) {
  return db.select().from(projects)
    .where(eq(projects.orgId, orgId))
    .orderBy(desc(projects.createdAt));
}

export async function getProject(orgId: string, id: string) {
  const [row] = await db.select().from(projects)
    .where(and(eq(projects.id, id), eq(projects.orgId, orgId)))  // scope by BOTH
    .limit(1);
  return row ?? null;
}

export async function createProject(orgId: string, input: { name: string }) {
  const [row] = await db.insert(projects).values({ orgId, ...input }).returning();
  return row;
}

Callers (server components, actions, route handlers) pass the orgId resolved from the session — they never invent it.

Why a DAL

• Isolation by construction. Lookups filter on id AND org_id, so a guessed UUID from another tenant returns null, not data.
• Typed end-to-end. returning() gives you the row type for free.
• Testable. Mock or seed the DB and test functions directly, no HTTP needed.
• Cacheable. Wrap reads in cache() / unstable_cache in one spot.

Pitfalls

• Calling db from components — couples UI to the schema and risks shipping queries to the client. Always go through the DAL.
• Lookups by id only — where(eq(projects.id, id)) without orgId is a cross-tenant read. Scope by both.
• Returning raw rows with secrets — strip sensitive columns (tokens, hashes) in the DAL.
• N+1 in loops — batch with inArray or a join instead of awaiting per item.
• Forgetting server-only — without it a refactor can pull DB code into the client bundle.

Hand-off

A typed, tenant-safe query layer. api-routes-and-validation and Server Actions call it after validating input; authorization-rbac gates which functions a role may invoke.