[Bug]: crashReporter.start blocks main process takes more than 600ms on macOS

[cupools]

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

18.2.4

What operating system are you using?

macOS

Operating System Version

macOS Monterey 12.3.1

What arch are you using?

x64

Last Known Working Electron version

18.2.4

Expected Behavior

Expect crashReporter.start to return quickly. It takes only 33ms in Electron@11 with the same code on the same machine.

Actual Behavior

It seems that #33072 does not the same question. crashReporter.start takes a long time while spawn returns quickly, and seriously delay the app startup.

/** Electron@18.2.4, node@v16.13.0 */

console.time('crashReporter')
crashReporter.start({ submitURL: 'https://your-domain.com/url-to-submit' })
console.timeEnd('crashReporter')
// => crashReporter: 762.32ms

console.time('spawn')
child_process.spawn('ls')
console.timeEnd('spawn')
// => spawn: 3.836ms

/** Electron@11.5.0, node@v16.13.0 */

console.time('crashReporter')
crashReporter.start({ submitURL: 'https://your-domain.com/url-to-submit' })
console.timeEnd('crashReporter')
// => crashReporter: 32.330ms

console.time('spawn')
child_process.spawn('ls')
console.timeEnd('spawn')
// => spawn: 6.081ms

Testcase Gist URL

https://gist.github.com/cupools/9336a79b76ef5da2b61105a63c796824

Additional Information

No response

[cupools]

Here's the performance profile.

[RaisinTen]

I've sent https://chromium-review.googlesource.com/c/crashpad/crashpad/+/3641386 to fix the performance problem. Waiting for approvals.

[RaisinTen]

My Crashpad CL landed yesterday, so I submitted a backport PR - #34609. With that change, initializing the crashReporter should just take 257 milliseconds on Intel Macs.

Will try to submit a follow up CL to reduce that even further if possible.

EDIT:

Sent another CL to crashpad which reduces the 257 milliseconds duration to 1 millisecond - https://chromium-review.googlesource.com/c/crashpad/crashpad/+/3711503.

However, initializing the crashpad would take around 110 milliseconds the first time you run Electron and then that reduces to 10 milliseconds in subsequent runs and all of that time is consumed here - https://source.chromium.org/chromium/chromium/src/+/main:third_party/crashpad/crashpad/client/crashpad_client_mac.cc;l=360-361;drc=300938d278c0fe0e1485a5f1731d242eb47ccc27;bpv=0;bpt=0. Will see if I can find the reason behind this.

[bpasero]

@RaisinTen thanks a ton for this! you said:

Sent another CL to crashpad which reduces the 257 milliseconds duration to 1 millisecond - https://chromium-review.googlesource.com/c/crashpad/crashpad/+/3711503.

Curious if that is something we could get too or whether that change is breaking?

[RaisinTen]

@bpasero

thanks a ton for this!

🙌

Curious if that is something we could get too or whether that change is breaking?

Not in the current state of the CL. I do plan to resume my work on it again when I get some time. My next approach for replacing the final fork() call would be to build a tiny standalone executable and spawn that with posix_spawn() and that executable would in turn call setsid() and spawn the crashpad handler executable with fork() + exec(). Note that the fork() call in this case wouldn't be problematic because it's getting spawned in the standalone executable which wouldn't have any mmaped pages.