[Bug]: Spawning and unreferencing a child process prevents electron from exiting

[Josh-G]

Preflight Checklist

• I have read the Contributing Guidelines for this project.
• I agree to follow the Code of Conduct that this project adheres to.
• I have searched the issue tracker for a bug report that matches the one I want to file, without success.

Electron Version

19.0.7

What operating system are you using?

Other Linux

Operating System Version

Debian 10 (buster)

What arch are you using?

arm64 (including Apple Silicon)

Last Known Working Electron version

No response

Expected Behavior

Running the following:

spawn('sleep', ['10'], {
    detached: true,
    stdio: 'ignore',
}).unref();

should allow Electron to exit without waiting for the child process.

Actual Behavior

Electron emits the correct events, however some electron processes linger until the child process has completed.

Testcase Gist URL

https://gist.github.com/Josh-G/aadee165b50c7629d5f1c9340728a8ca

Additional Information

I've also created a repo with a workflow that demonstrates the issue; https://github.com/Josh-G/electron-spawn-detached

I believe this may be the same as, or a regression of, #2208 #13870.
I've reproduced this on Linux on arm64 + x64 with Debian and Ubuntu (thru GitHub Action).

[Prinzhorn]

I've noticed this in the past as well. Some processes don't exit:

  28712 ?        S      0:00 /Electron Fiddle/electron-bin/19.0.7/electron --type=zygote --no-zygote-sandbox --enable-crashpad --enable-crashpad
  28713 ?        S      0:00 /Electron Fiddle/electron-bin/19.0.7/electron --type=zygote --enable-crashpad --enable-crashpad
  28715 ?        S      0:00 /Electron Fiddle/electron-bin/19.0.7/electron --type=zygote --enable-crashpad --enable-crashpad

while these exit immediately (if you wrap process.exit in a setTimeout you will see them go away while the above stay until sleep is done):

  28971 ?        Sl     0:00 /Electron Fiddle/electron-bin/19.0.7/electron --type=gpu-process --enable-crashpad --enable-crash-reporter=a9f51b37-898e-4b52-a491-88193519d896,no_channel --user-data-dir=/illustrious-speaker-spend-o4irg --gpu-preferences=WAAAAAAAAAAgAAAIAAAAAAAAAAAAAAAAAABgAAAAAAA4AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAACAAAAAAAAAAIAAAAAAAAAABAAAAAAAAAAgAAAAAAAAACAAAAAAAAAAIAAAAAAAAAA== --shared-files --field-trial-handle=0,i,10080314104194432827,127912701232379743,131072 --disable-features=SpareRendererForSitePerProcess
  28986 ?        Sl     0:00 /Electron Fiddle/electron-bin/19.0.7/electron --type=utility --utility-sub-type=network.mojom.NetworkService --lang=en-US --service-sandbox-type=none --enable-crashpad --enable-crash-reporter=a9f51b37-898e-4b52-a491-88193519d896,no_channel --user-data-dir=/illustrious-speaker-spend-o4irg --shared-files=v8_context_snapshot_data:100 --field-trial-handle=0,i,10080314104194432827,127912701232379743,131072 --disable-features=SpareRendererForSitePerProcess --enable-crashpad

[Prinzhorn]

Well, I just had a deja vu and I commented the exact same thing a year ago 😄 #24520 (comment)

Electron processes are kept alive

For some reason this child process (that now belongs to systemd) keeps several Electron processes alive. To me this looks broken?

 $ ps ax | grep electron
  16478 ?        S      0:00 /home/alex/.config/Electron Fiddle/electron-bin/13.1.6/electron --type=zygote --no-zygote-sandbox
  16479 ?        S      0:00 /home/alex/.config/Electron Fiddle/electron-bin/13.1.6/electron --type=zygote
  16481 ?        S      0:00 /home/alex/.config/Electron Fiddle/electron-bin/13.1.6/electron --type=zygote

Back then what I meant with "normal unix behavior" is that the sleep process will be moved to the next parent (e.g. systemd) and does not exit with Electron. But it shouldn't keep Electron waiting for it.

[Josh-G]

When using electron-updater on Linux with the AppImage target, this can cause a significant amount of lingering processes on long-running hosts that apply multiple updates

[Prinzhorn]

Just started writing e2e tests and now this becomes a real problem. The tests will time out because Electron never exits.

Even this is not enough:

app.exit(0);
process.exit(0);

As long as the child process does not exit, playwrights await app.close() will never resolve. If I kill the child process manually, the test passes, because the lingering Electron processes can finally exit.