[Bug]: Reusing page between tests when running serially adds extra teardown time per test

[david-olink]

Version

1.43.1

Steps to reproduce

Reusing page between tests when running serially, as described here https://playwright.dev/docs/test-retries#reuse-single-page-between-tests , adds extra time in in teardown for each test (even if the tests are empty).

Run this Test Suite:

import type { Page } from '@playwright/test';
import { test } from '@playwright/test';

// Based on the reuse-single-page-between-tests example:
// https://playwright.dev/docs/test-retries#reuse-single-page-between-tests

test.describe.configure({ mode: 'serial' });

let page: Page;

test.beforeAll(async ({ browser }) => {
  page = await browser.newPage();
});

test.afterAll(async () => {
  await page.close();
});

for (let i = 0; i < 5; i++) {
  test('Before: Empty test ' + i, async () => {});
}

test('First time we use the page object', async () => {
  await page.goto('/');
});

//
// Bug: These empty tests takes much more time than they should.
//

for (let i = 0; i < 5; i++) {
  test('After: Empty test (takes much more time) ' + i, async () => {});
}

Expected behavior

No test except the last should have a "After Hooks" that takes more than ~20ms instead of the ~700ms for the last tests as seen here:

Actual behavior

Running the above code adds extra "After Hooks" time as seen here:

Additional context

No response

Environment

System:
    OS: Windows 11 10.0.22631
    CPU: (12) x64 12th Gen Intel(R) Core(TM) i7-1255U
    Memory: 8.83 GB / 31.69 GB
  Binaries:
    Node: 20.11.1 - C:\Program Files\nodejs\node.EXE
    npm: 10.2.4 - C:\Program Files\nodejs\npm.CMD
  IDEs:
    VSCode: 1.88.1 - C:\Users\user\AppData\Local\Programs\Microsoft VS Code\bin\code.CMD
  Languages:
    Bash: 5.2.15 - C:\Users\user\AppData\Local\Programs\Git\usr\bin\bash.EXE

[dgozman]

@david-olink This is because you run tests in UI mode, which forces recording the trace, and tests after the first goto() have non-empty traces containing page resources to show you the snapshots. Why do you consider this to be a problem?

[david-olink]

Thanks for the clarification of what is happening. I think it would be clearer if the "After Hooks" sections show that it is the trace that takes that long time.

• Could it be possible to clarify that this is what the after hook does?
• Could it maybe be possible to speed up the performance of the trace after hook?

I want speed up our testing as much as possible for our developers when running in UI mode, we also need the tests to be run serially. So that is why I want to remove any unnecessary overhead if possible :)

[dgozman]

• Could it be possible to clarify that this is what the after hook does?

We can look into this.

• Could it maybe be possible to speed up the performance of the trace after hook?

We are always keen to improve Playwright's performance, but we do not have any specific ideas about tracing right now.

I want speed up our testing as much as possible for our developers when running in UI mode, we also need the tests to be run serially. So that is why I want to remove any unnecessary overhead if possible :)

Note that we strongly advise against the serial mode. Perhaps you just need to disable the parallelization with workers: 1? Making tests independent will improve everything, from reliability to performance.

[dgozman]

I'll tentatively close this given the improvements in #30946 that will be released in v1.45. If you encounter issues after upgrading to 1.45 once it's out, please file a new issue.