1432601 - Watchdog process

Status: NEW

(Core :: DOM: Content Processes, enhancement, P3)

Description

[Thomas Bertels]

An unknown percentage of hangs and crashes don't get reported by Breakpad.
A watchdog process would make sure these events are reported. By the way, Chrome already does that for unclean shutdowns, see https://bugs.chromium.org/p/chromium/issues/detail?id=412384.

I propose to use this process to also kill non responding processes after a set time.
It would prevent the browser from becoming unresponsive (hang). Those hangs could then be reported instead of staying under the radar.

This process could also be a start for bug 1374353, for testing maximum response delays when the goal is snappiness instead of just preventing browser hangs.

Comment 1

[Jim Mathies [:jimm]]

(In reply to Thomas Bertels from comment #0)
> An unknown percentage of hangs and crashes don't get reported by Breakpad.
> A watchdog process would make sure these events are reported. By the way,
> Chrome already does that for unclean shutdowns, see
> https://bugs.chromium.org/p/chromium/issues/detail?id=412384.

We catch and report unclean shutdowns internally through timing. I'm not sure another proceses running helps us catch additional data.

> 
> I propose to use this process to also kill non responding processes after a
> set time.
> It would prevent the browser from becoming unresponsive (hang). Those hangs
> could then be reported instead of staying under the radar.

We plan to do some work on this for Windows this year, separating the internal gecko thread from the native UI thread. This should help address hang issues we still see. I'd prefer to make those changes first.

Keeping this bug around for tracking the idea.