Measure and report time between user-space process crashes.

Measure and report time between user-space process crashes.

BUG=none
TEST=unit tests, gmerged on the device and inspected
logs, about:histograms,etc.

[petkov, 2010-06-11 20:46:46]

Here's a version for an early review.

Some known issues that we may or may not want fixed:

- Crash intervals are reported in seconds. Would milliseconds make more sense?
- Crash intervals of 0 are not reported at all. Therefore the time between two
crashes back-to-back in less than a second, or while the screenlocker is on and
no active use in between will not be reported.

[kmixter1, 2010-06-11 21:19:44]

First, thanks for looking at this!

I don't think sub-1 second resolution is so important, however counting them
still seems important.  So I guess it would be better to still sample "0" when a
crash occurs at less than 1s interval.

http://codereview.chromium.org/2736008/diff/3001/4001
File metrics_daemon.cc (right):

http://codereview.chromium.org/2736008/diff/3001/4001#newcode62
metrics_daemon.cc:62: const int MetricsDaemon::kMetricUserCrashIntervalMax =
kSecondsPerDay;
I think it would be interesting to know if the time between crashes was 48 hours
vs 100 hours vs a month.  I guess above a month we aren't as concerned.  But we
are concerned above 24 hours.

[petkov, 2010-06-16 21:01:03]

Sub-one seconds crashes and crashes while the screen saver is active are
reported with an interval of 0 now.

PTAL.

http://codereview.chromium.org/2736008/diff/3001/4001
File metrics_daemon.cc (right):

http://codereview.chromium.org/2736008/diff/3001/4001#newcode62
metrics_daemon.cc:62: const int MetricsDaemon::kMetricUserCrashIntervalMax =
kSecondsPerDay;
On 2010/06/11 21:19:44, kmixter1 wrote:
> I think it would be interesting to know if the time between crashes was 48
hours
> vs 100 hours vs a month.  I guess above a month we aren't as concerned.  But
we
> are concerned above 24 hours.

Increased this to 4 weeks. Again, the mean would be correct either way. Wonder
if we should switch to minutes rather than seconds?

[kmixter1, 2010-06-16 22:41:05]

LGTM

http://codereview.chromium.org/2736008/diff/6001/7004
File metrics_daemon.cc (right):

http://codereview.chromium.org/2736008/diff/6001/7004#newcode345
metrics_daemon.cc:345: if (user_active_) {
This is subtle.  Do you want to create something like UpdateActiveState which
does this and the above SetUserActiveState for use when called asynchronously
through an RPC?

[petkov, 2010-06-16 22:50:27]

I'll push this as soon as http://codereview.chromium.org/2716010/show LGTY.

http://codereview.chromium.org/2736008/diff/6001/7004
File metrics_daemon.cc (right):

http://codereview.chromium.org/2736008/diff/6001/7004#newcode345
metrics_daemon.cc:345: if (user_active_) {
On 2010/06/16 22:41:05, kmixter1 wrote:
> This is subtle.  Do you want to create something like UpdateActiveState which
> does this and the above SetUserActiveState for use when called asynchronously
> through an RPC?

Actually, I've remove the monitor rescheduling on crashes. There's really no
need to restart the back off on such events.