Memory leak when instantiating arktype validators

[VanTanev]

Report a bug

Memory leak when an arktype validator is instantiated, even with something as simple as type("string"), and anything else that instantiates a type, like type.pick() or type.and(other).

Memory does not leak on type.to(other), or type.keyof()

🔎 Search Terms

OOM, memory leak

🧩 Context

• ArkType version: 2.1.29
• TypeScript version (5.1+): 5.9.3
• Other context you think may be relevant (JS flavor, OS, etc.):
  Node 24.13.0

🧑‍💻 Repro

// test-arktype-memory.ts
import { type } from "arktype";
// import { z } from "zod";

const BATCH = 1_000;
let i = 0;

while (true) {
  { 
    testArkType();
  }
  if (++i % BATCH === 0) {
    i = 0
    global.gc?.();
    console.log(
      `${(process.memoryUsage().heapUsed / 1024 / 1024).toFixed(2)} MB`,
    );
  }
}

// function testZod() {
//   z.string().transform((x) => parseInt(x)).parse('1');
// }

function testArkType() {
  type("string")('1');
}

Run with: node --expose-gc --no-warnings test-arktype-memory.ts

[pullfrog]

  ^{｜ Build this ➔ ｜ Make a plan ➔ ｜ pullfrog.com}

[chris-kreidl]

I'm not a maintainer here, just a fan of the project, but I think this is expected behavior. Arktype registers every instantiated schema in a global registry, so if you create unbounded schemas that registry will grow unbounded as well.

I don't think this is how the library was meant to be used. If you were to cache that schema outside of the loop you'd observe that no additional memory is consumed.

For example, building on your demo:

// test-arktype-memory.ts
import { nodesByRegisteredId } from "@ark/schema";
import { type } from "arktype";
// import { z } from "zod";

const BATCH = 1_000;
const REUSE_SCHEMA = false;
let i = 0;

const cached = REUSE_SCHEMA ? type("string") : null;

while (true) {
  { 
    testArkType();
    // testZod();
  }
  if (++i % BATCH === 0) {
    i = 0
    global.gc?.();
    console.log(
      `${Object.keys(nodesByRegisteredId).length} nodes`,
      `${(process.memoryUsage().heapUsed / 1024 / 1024).toFixed(2)} MB`,
    );
  }
}

// function testZod() {
//   z.string().transform((x) => parseInt(x)).parse('1');
// }

function testArkType() {
  if (cached) return cached("1");
  type("string")("1");
}

Run once with REUSE_SCHEMA = false and again with REUSE_SCHEMA = true and observe the difference.

[eralmansouri]

How would it be expected behavior to run out of memory?

[chris-kreidl]

How would it be expected behavior to run out of memory?

I think it's unexpected behavior to instantiate millions of schemas.

This is akin to putting a brick on your car's gas pedal, walking away, and asking why your engine blew up.

If you use the library as intended then this is a non-issue.

[VanTanev]

I have confirmed with @ssalbdivad that this is indeed a bug and not intended behavior. As far as I'm aware, none of the other validation libraries suffer from a similar memory leak on schema instantiation.

[eralmansouri]

How would it be expected behavior to run out of memory?

I think it's unexpected behavior to instantiate millions of schemas.

This is akin to putting a brick on your car's gas pedal, walking away, and asking why your engine blew up.

If you use the library as intended then this is a non-issue.

"You're holding it wrong."

Does it really make sense to you that had to implement "caching" to avoid the library caching? What did the library cache, then, if each instantiation consumes more memory?

I have confirmed with @ssalbdivad that this is indeed a bug and not intended behavior. As far as I'm aware, none of the other validation libraries suffer from a similar memory leak on schema instantiation.

Thank you.