Performance regression when upgrading from Vitest v3 to v4 with single worker thread

[bpetetot]

Describe the bug

After upgrading from Vitest v3 to v4, I’m seeing a consistent ~2× slowdown in our test suite when running with a single worker thread.

Prior behavior (v3)
Using the following config:

export default defineConfig({
  test: {
    globals: true,
    environment: 'node',
    pool: 'threads',
    poolOptions: { threads: { singleThread: true } },
  },
})

Execution time:

Duration 846ms
(transform 137ms, setup 0ms, collect 153ms, tests 9ms, environment 0ms, prepare 193ms)

After upgrade (v4)
Updated config per migration guide, replacing deprecated poolOptions:

export default defineConfig({
  test: {
    globals: true,
    environment: 'node',
    pool: 'threads',
    maxWorkers: 1,
  },
})

Execution time:

Duration 1.79s
(transform 122ms, setup 0ms, collect 193ms, tests 15ms, environment 0ms, prepare 35ms)

Observations

• Test execution time is nearly doubled from 846ms to 1.79s. (results from the minimal reproductions on StackBlitz)
• On my main project, I have almost the same difference (from 54s to 138s)

Thanks for vitest, it is awesome 😊

Reproduction

• Reproduction Vitest v3 (before): https://stackblitz.com/edit/vitest-dev-vitest-dh3lzrnu?file=vite.config.ts

• Reproduction Vitest v4 (after): https://stackblitz.com/edit/vitest-dev-vitest-b2yoehcc?file=vite.config.ts

System Info

System:
    OS: macOS 15.6.1
    CPU: (12) arm64 Apple M3 Pro
    Memory: 405.28 MB / 18.00 GB
    Shell: 5.9 - /bin/zsh
  Binaries:
    Node: 22.18.0 - /Users/benjamin_petetot/.local/share/mise/installs/node/22.18.0/bin/node
    npm: 10.9.3 - /Users/benjamin_petetot/.local/share/mise/installs/node/22.18.0/bin/npm
    pnpm: 10.18.3 - /opt/homebrew/bin/pnpm
    Watchman: 2025.10.13.00 - /opt/homebrew/bin/watchman
  Browsers:
    Brave Browser: 141.1.83.118
    Chrome: 139.0.7258.155
    Firefox: 137.0
    Safari: 18.6
  npmPackages:
    vitest: ^4.0.3 => 4.0.3

Used Package Manager

npm

Validations

• Follow our Code of Conduct
• Read the Contributing Guidelines.
• Read the docs.
• Check that there isn't already an issue that reports the same bug to avoid creating a duplicate.
• Check that this is a concrete bug. For Q&A open a GitHub Discussion or join our Discord Chat Server.
• The provided reproduction is a minimal reproducible example of the bug.

[AriPerkkio]

When removing --singleThread/--singleForks in #8705, it was assumed it's equal to --maxThreads=1, --maxForks=1 and --maxWorkers=1. Looking at sources of v3, it also seems to disable isolation and clear modules between each test. You should achieve same with following config:

import { defineConfig } from "vitest/config";

export default defineConfig({
  test: {
    isolate: false,
    maxWorkers: 1,
    setupFiles: ["./setup.ts"],
  },
});

// setup.ts
import { beforeAll, vi } from "vitest";

beforeAll(() => {
  vi.resetModules();
  vi.restoreAllMocks();
});

[bpetetot]

Thanks @AriPerkkio!

I applied the isolate: false option along with the setup.ts reset, and performance is now back to what we had on v3.

But, after applying this configuration to a project with more test files (+100 test files), I see this warning:

(node:10105) MaxListenersExceededWarning: Possible EventEmitter memory leak detected. 11 error listeners added to [EventEmitter]. MaxListeners is 10. Use emitter.setMaxListeners() to increase limit
		at genericNodeError (node:internal/errors:983:15)
		at wrappedFn (node:internal/errors:537:14)
		at _addListener (node:events:592:17)
		at EventEmitter.addListener (node:events:610:10)
		at PoolRunner.on (file:///Users/ben/my-project/node_modules/vitest/dist/chunks/cli-api.D24RdiHp.js:6488:22)
		at Pool.schedule (file:///Users/ben/my-project/node_modules/vitest/dist/chunks/cli-api.D24RdiHp.js:6841:46)
		at Pool.schedule (file:///Users/ben/my-project/node_modules/vitest/dist/chunks/cli-api.D24RdiHp.js:6867:179)
		at processTicksAndRejections (node:internal/process/task_queues:105:5)

I'll try to create a minimal repro.

[AriPerkkio]

The stacktrace was enough for me to reproduce locally - fix coming up at #8820.

I applied the isolate: false option along with the setup.ts reset, and performance is now back to what we had on v3.

I would also recommend to try changing the maxWorkers in your project to see if performance improves even more.

[bpetetot]

I applied your fix locally to test it. And I can confirm that the previous “MaxListenersExceededWarning” warning has disappeared.

I didn't mention it in my previous message because I thought it was related to the previous error, but I also see this warning. (It is still present after the fix.)

(node:17664) MaxListenersExceededWarning: Possible EventTarget memory leak detected. 11 message listeners added to [MessagePort [EventTarget]]. MaxListeners is 10. Use events.setMaxListeners() to increase limit
    at MessagePort.[kNewListener] (node:internal/event_target:562:17)
    at MessagePort.<anonymous> (node:internal/worker/io:226:12)
    at MessagePort.addEventListener (node:internal/event_target:686:23)
    at MessagePort.on (node:internal/event_target:965:10)
    at on (file:///Users/ben/my-project/node_modules/vitest/dist/chunks/init-threads.BEmFEaEY.js:9:32)
    at createBirpc (file:///Users/ben/my-project/node_modules/vitest/dist/chunks/index.Bgo3tNWt.js:164:14)
    at createRuntimeRpc (file:///Users/ben/my-project/node_modules/vitest/dist/chunks/rpc.cD77ENhU.js:31:22)
    at execute (file:///Users/ben/my-project/node_modules/vitest/dist/chunks/init.DvvF4rpe.js:68:28)
    at run (file:///Users/ben/my-project/node_modules/vitest/dist/chunks/init.DvvF4rpe.js:105:9)
    at MessagePort.onMessage (file:///Users/ben/my-project/node_modules/vitest/dist/chunks/init.DvvF4rpe.js:163:19)

I would also recommend to try changing the maxWorkers in your project to see if performance improves even more.

I had changed the maxWorkers, and I confirm performance improves even more.

[AriPerkkio]

The second listener leak is addressed in #8821 🙃.

I'll next update the docs for singleThread migration so we can close this issue too.