Force a crash of the browser if IO thread hasn't responded

Force a crash of the browser if IO thread hasn't responded
3 times for the ping message and if only one other thread
is responding.

Collect stats when it is the first time we have detected
that the watched thread has not responded atleast 3 times
(that is when we haven't done any thread hang processing).


BUG=82336
TEST=thread watcher unit tests
R=jar
Committed: http://src.chromium.org/viewvc/chrome?view=rev&revision=85631

[ramant (doing other things), 2011-05-16 18:28:44]

Hi Jim,
  In Canary build 13.0.766.0 that was released yesterday, we would have crashed
4 times (bucket #1). For 2 threads not responding, we would have crashed 61
times.

  This change crashes the browser if only one thread is responding and all other
threads are not responding.

  Would appreciate your opinion/comments.

thanks,
raman

http://www.corp.google.com/~chrome-eng/dashboards/histograms/

Count: 1,280; Sum: 4,768; Mean: 4

[jar (doing other things), 2011-05-16 21:24:25]

LGTM

[Peter Kasting, 2011-05-17 00:41:05]

http://codereview.chromium.org/7030020/diff/3002/chrome/browser/metrics/threa...
File chrome/browser/metrics/thread_watcher.cc (right):

http://codereview.chromium.org/7030020/diff/3002/chrome/browser/metrics/threa...
chrome/browser/metrics/thread_watcher.cc:271: CHECK(crash++);
Why is this better than simply NOTREACHED();?

[jar (doing other things), 2011-05-17 01:11:26]

We were anticipating having distinct crashes based on distinct watchdog
threads going south.

We expect the optimizer to make it difficult to tell which thread was
reporting a crash.

The plan is to access slightly different illegal addresses (offsets from
NULL pointer), depending on which thread crashes.  This illegal access will
appear in the crash dump.

This first version is not very exciting, as there is only one thread
monitored to induce a crash (so it is clear what caused the crash).

NOTREACHED would give us no way to distinguish the broken threads.

Jim

On Mon, May 16, 2011 at 5:41 PM, <pkasting@chromium.org> wrote:

>
>
>
http://codereview.chromium.org/7030020/diff/3002/chrome/browser/metrics/threa...
> File chrome/browser/metrics/thread_watcher.cc (right):
>
>
>
http://codereview.chromium.org/7030020/diff/3002/chrome/browser/metrics/threa...
> chrome/browser/metrics/thread_watcher.cc:271: CHECK(crash++);
> Why is this better than simply NOTREACHED();?
>
>
> http://codereview.chromium.org/7030020/
>