[Bug]: UI seems to lock on Windows making tests time out

[pbrzosko]

Version

1.42.1

Steps to reproduce

1. Clone my repo at https://github.com/pbrzosko/playwright-lock-issue
2. Run npm install
3. Run npm run pw on a Windows machine
4. Execute tests in the UI

Expected behavior

All tests should run similar amount of time and not time out. Here is a screenshot of them running on OSX:

Actual behavior

On Windows, it seems like tests are locking each other on some resource. Thus they start to time out and run very slowly:

Additional context

When run with one worker, it seems to work fine:

Environment

System:
    OS: Windows 10 10.0.19045
    CPU: (24) x64 12th Gen Intel(R) Core(TM) i7-12850HX
    Memory: 6.77 GB / 31.69 GB
  Binaries:
    Node: 18.12.0 - C:\Program Files\nodejs\node.EXE
    npm: 8.19.2 - C:\Program Files\nodejs\npm.CMD
  IDEs:
    VSCode: 1.87.2 - C:\Users\rgzc\AppData\Local\Programs\Microsoft VS Code\bin\code.CMD
  Languages:
    Bash: 5.0.17 - C:\Windows\System32\bash.EXE
  npmPackages:
    @playwright/experimental-ct-vue: ^1.42.1 => 1.42.1

[pbrzosko]

Seems like problem also exists when running from command line, but with smaller impact:

[dgozman]

@pbrzosko I have played with your repro a bit. Things are slowing down indeed, but some of that is expected due to increased parallelization between single worker and many workers. However, I did not reach 10 seconds.

UI mode is noticeably slower. I think this is due to trace being recorded by UI mode, and disk access slowing everything down. Perhaps the shared disk access may even slow down all workers in parallel, which looks like "locking effect".

[pbrzosko]

How about every worker using its own directory to gather trace and then everything is combined at the end of the run? Why it doesn't happen on OSX? What times do you get? It might be some of the company installed software that spoils things but we have no idea where to look?

[dgozman]

How about every worker using its own directory to gather trace and then everything is combined at the end of the run?

Each worker has its own directory already.

Why it doesn't happen on OSX?

I guess the filesystem is implemented differently? Note that it was just my first idea, and the problem might be entirely different.

What times do you get?

Everything is ~1.8x slower in UI mode. Going up to 8 workers slows down each test ~2x.

It might be some of the company installed software that spoils things but we have no idea where to look?

That could easily be the case with the antivirus. You can check that by running with --workers=1 and --trace=on/--trace=off without UI mode. This way you can isolate the impact of tracing, which writes to disk, on the test timing. Something like an antivirus could slow things down significantly.

[pbrzosko]

I will check the trace impact. But when we run with 1 worker in UI mode, times are as expected. So it seems like some locking issue which is only visible with many workers running.

[dgozman]

I'll close this issue, because there is no action item for Playwright right now. If you come up with some new findings, please file a new issue and link to this one, so we can continue the discussion.

[pbrzosko]

Yyy, say that again @dgozman? You have clear evodence that tests run over 10s instead of one and you say there is nothing to do from your side?