[BUG] regression in killing web server process in 1.28.0

Context:

• Playwright Version: 1.28.0
• Operating System: Linux and Mac
• Node.js version: 16 and 18
• Browser: All
• Extra: Windows is okay

Code Snippet

You can run locally with:

npm create svelte@latest myapp (choose all defaults, but add playwright)
cd myapp
edit package.json to set playwright version to 1.28.0
npm install
npm test (run it twice)

Describe the bug

Server process is no longer being killed on Mac and Linux. It’s possible we have some issue on our side, but it’s not clear to me what that would be or how to debug it. I do experience this locally, but can kill the process with kill after Playwright fails to do so.

I see @aslushnikov sent a PR (https://github.com/microsoft/playwright/pull/18220) affecting this section of code to address @rmunn’s issue (https://github.com/microsoft/playwright/issues/18209), which was later reverted (https://github.com/microsoft/playwright/pull/18220). The issue was not reopened - should it be? Since that change was reverted, I’m not quite sure where else this might be coming from.

[benmccann]

+1 for treeKill by default. Kill it all with fire

[dominikg]

changing from SIGKILL to SIGTERM is a breaking change in my opinion. This is no longer ending child processes by default and relies on the started process to handle SIGTERM to end children as well.

A backward-compatible way to introduce this would be to have the signal as an option, default to SIGKILL until the next major and flip it then, allowing users to still use SIGKILL if their server process isn’t handling SIGTERM.

sidenote: even SIGKILL can be problematic still, i have run into situations where that wasn’t enough and resorted to use tree-kill on the process pid to ensure everything is gone. https://github.com/sveltejs/vite-plugin-svelte/blob/90ff974fb81c7fff05f5e6e4ae1497a24d3078c4/packages/e2e-tests/e2e-server.js#L118