[ETW Export] Add polling for ETW keyword.

[ETW Export] Add polling for ETW keyword.

Couple of notes:
- TraceEventETWExport is a friend of TraceLog (see comment in code).
- Restored the original implementation of IsETWExportEnabled (needed until command line flag is removed). Also moved body to cc for consistency (and rearranged order of functions)
- The timer for now is also behind the command line flag. But eventually, when it's activated by default, I'm worried about power usage, as this will force the CPU to wake up once a sec.

BUG=491909

[Georges Khalil, 2015-06-26 21:32:25]

Patchset #1 (id:1) has been deleted

[Georges Khalil, 2015-06-26 21:43:23]

Patchset #1 (id:20001) has been deleted

[Georges Khalil, 2015-06-26 21:47:27]

georgesak@chromium.org changed reviewers:
+ brucedawson@chromium.org

[Georges Khalil, 2015-06-26 21:47:38]

Patchset #1 (id:40001) has been deleted

[Georges Khalil, 2015-06-26 21:47:48]

Bruce, PTAL.

[brucedawson, 2015-06-26 22:24:24]

Two comments...

https://codereview.chromium.org/1213523003/diff/60001/base/trace_event/trace_...
File base/trace_event/trace_event_etw_export_win.cc (right):

https://codereview.chromium.org/1213523003/diff/60001/base/trace_event/trace_...
base/trace_event/trace_event_etw_export_win.cc:178: return (GetInstance() &&
GetInstance()->etw_export_enabled_);
If you move this and change it in the same CL then the change is very hard to
see. Consider leaving it in its original location.

A separate CL to move it (before or after) would also work.

https://codereview.chromium.org/1213523003/diff/60001/content/browser/browser...
File content/browser/browser_main_loop.cc (right):

https://codereview.chromium.org/1213523003/diff/60001/content/browser/browser...
content/browser/browser_main_loop.cc:704:
base::trace_event::TraceEventETWExport::StartKeywordUpdateTimer();
Does this get called in just the browser or in all processes?

If in just the browser then the other processes will never emit events.
If in all processes then we can end up with arbitrarily many wakeups per second.

Ideal implementation would do the timer-based checking in just the browser
process and then send an IPC message to the other processes when there is a
change. I'm not sure if that is overkill or ideal.

[brucedawson, 2015-08-07 20:28:56]

I'd like to request/recommend that you land this change behind the
--trace-export-events-to-etw flag. That is, if this flag is specified then check
the CHROME_Context.MatchAnyKeyword value once a second in every Chrome process.

This isn't ideal, since it still requires that users add a command-line flag in
order to get these events, but it would get rid of a significant source of error
and weirdness. Right now users have to add the command-line flag *and* they need
to ensure that tracing is running when Chrome launches *and* they can't change
the categories without relaunching Chrome. It's very confusing.

Doing away with the flag completely can be done at some later point.

I realized how annoying the lack of any polling ability is while I was telling a
user how to record an ETW trace with Chrome events in it, and *I* was getting
confused by the steps involved.

[Georges Khalil, 2015-08-10 21:27:30]

On 2015/08/07 20:28:56, brucedawson wrote:
> I'd like to request/recommend that you land this change behind the
> --trace-export-events-to-etw flag. That is, if this flag is specified then
check
> the CHROME_Context.MatchAnyKeyword value once a second in every Chrome
process.
> 
> This isn't ideal, since it still requires that users add a command-line flag
in
> order to get these events, but it would get rid of a significant source of
error
> and weirdness. Right now users have to add the command-line flag *and* they
need
> to ensure that tracing is running when Chrome launches *and* they can't change
> the categories without relaunching Chrome. It's very confusing.
> 
> Doing away with the flag completely can be done at some later point.
> 
> I realized how annoying the lack of any polling ability is while I was telling
a
> user how to record an ETW trace with Chrome events in it, and *I* was getting
> confused by the steps involved.

Closing this. New approach in https://codereview.chromium.org/1279353002/