On Posix, make all child processes quit when the browser dies, not just the renderers.

On Posix, make all child processes quit when the browser dies, not just the renderers.

On bots, seeing that sometimes child processes are hanging around after the browser process is gone. This confuses the sharding scripts.

BUG=90448
Committed: https://src.chromium.org/viewvc/chrome?view=rev&revision=149023

[jam, 2012-07-30 20:05:17]

switching reviewers to Julien, since I've talked to him about this issue

[Nirnimesh, 2012-07-30 20:13:24]

+asharif, as FYI

[asharif1, 2012-07-30 20:37:06]

On 2012/07/30 20:13:24, Nirnimesh wrote:
> +asharif, as FYI

jam, could you change chrome/test/functional/perf.py to use this flag instead of
--renderer-clean-exit?

Thanks,

[jam, 2012-07-30 20:40:56]

On 2012/07/30 20:37:06, asharif1 wrote:
> On 2012/07/30 20:13:24, Nirnimesh wrote:
> > +asharif, as FYI
> 
> jam, could you change chrome/test/functional/perf.py to use this flag instead
of
> --renderer-clean-exit?

Done

[jln (very slow on Chromium), 2012-07-30 20:51:19]

We need to allow the alarm() system call in the sandbox as well. This is in
content/common/sandbox_init_linux.cc. Since this would be an existing problem, I
can do it myself in another CL if you prefer. (Currently, under a modern kernel,
calling alarm() would kill the process).

If you wanted to, I think you could use the Watchdog class from base instead of
using alarm().

Otherwise: LGTM.

[jam, 2012-07-30 21:01:57]

On 2012/07/30 20:51:19, Julien Tinnes wrote:
> We need to allow the alarm() system call in the sandbox as well. This is in
> content/common/sandbox_init_linux.cc. Since this would be an existing problem,
I
> can do it myself in another CL if you prefer. (Currently, under a modern
kernel,
> calling alarm() would kill the process).
> 
> If you wanted to, I think you could use the Watchdog class from base instead
of
> using alarm().
> 
> Otherwise: LGTM.

If you can take care of it in another cl, that'd be great, since I'm not
familiar with this code. Currently this works because the only two places that
use this are python scripts (modified in this change) that also use --no-sandbox

[asharif1, 2012-07-30 21:06:51]

On 2012/07/30 21:01:57, John Abd-El-Malek wrote:
> On 2012/07/30 20:51:19, Julien Tinnes wrote:
> > We need to allow the alarm() system call in the sandbox as well. This is in
> > content/common/sandbox_init_linux.cc. Since this would be an existing
problem,
> I
> > can do it myself in another CL if you prefer. (Currently, under a modern
> kernel,
> > calling alarm() would kill the process).
> > 
> > If you wanted to, I think you could use the Watchdog class from base instead
> of
> > using alarm().
> > 
> > Otherwise: LGTM.
> 
> If you can take care of it in another cl, that'd be great, since I'm not
> familiar with this code. Currently this works because the only two places that
> use this are python scripts (modified in this change) that also use
--no-sandbox

lgtm but someone else must approve.