Prisma 7 dropped the Rust engine - is it finally fast enough or are you still switching to Drizzle?

Backend
, ,
1

So Prisma 7 shipped with the big architectural change everyone was waiting for, the Rust query engine is gone and the whole client is pure TypeScript now. They’re claiming 3x faster queries from eliminating the serialization layer.

I’ve been running Prisma 5 in a mid-size Node.js API (about 40 models, PostgreSQL) and the cold start overhead on Lambda has been a constant pain point. The Rust binary added ~8MB to the bundle and cold starts were consistently 1.5-2s slower than what my team was seeing with lighter ORMs.

I did some quick benchmarks comparing Prisma 7 against Drizzle on the same schema:

// Simple select benchmark - 1000 iterations
// Prisma 7
const users = await prisma.user.findMany({
  where: { status: 'active' },
  include: { orders: { take: 5 } }
});
// Avg: 2.1ms per query

// Drizzle equivalent
const users = await db.query.users.findMany({
  where: eq(users.status, 'active'),
  with: { orders: { limit: 5 } }
});
// Avg: 1.4ms per query

Drizzle is still faster in raw overhead, but the gap is way smaller than it used to be. The bundle size difference is still significant though. Drizzle’s output is about 50KB vs Prisma 7 at roughly 2MB (better than the old 8MB, but still).

What I’m trying to figure out is whether it’s worth migrating a production codebase. The schema migration part seems straightforward since the underlying tables don’t change, but rewriting every query in a 40-model API is probably a week of work plus testing.

Has anyone actually done this migration on a non-trivial codebase? Specifically curious about:

  1. How painful was rewriting relational queries? Prisma’s include maps pretty cleanly to Drizzle’s with, but what about the edge cases?
  2. Did you see meaningful cold start improvements on serverless?
  3. Any gotchas with Drizzle’s migration tooling compared to prisma migrate?

Part of me thinks Prisma 7 is “good enough” now, but the other part knows that bundle size still matters a lot in our Lambda-heavy architecture.

Seed content posted by the DevForums team to help get our community started. Have a better answer? Jump in!