Add Media.AudioRendererEvents histogram to measure how often OnRenderError() is called.

media/filters/audio_renderer_impl.cc

Index: media/filters/audio_renderer_impl.cc
diff --git a/media/filters/audio_renderer_impl.cc b/media/filters/audio_renderer_impl.cc
index fd0a4104842ef8c97b247ad1515ad9dea0c33ae4..598269cd6ec1bc0836ac856f4bdebd754050f0b3 100644
--- a/media/filters/audio_renderer_impl.cc
+++ b/media/filters/audio_renderer_impl.cc
@@ -14,6 +14,7 @@
 #include "base/command_line.h"
 #include "base/logging.h"
 #include "base/message_loop_proxy.h"
+#include "base/metrics/histogram.h"
 #include "media/audio/audio_util.h"
 #include "media/base/audio_splicer.h"
 #include "media/base/bind_to_loop.h"
@@ -284,6 +285,8 @@ void AudioRendererImpl::OnDecoderSelected(
 
   state_ = kPaused;
 
+  HISTOGRAM_BOOLEAN("Media.AudioRendererSinkErrors", false);

[DaleCurtis, 2013/04/24 20:56:39]

I believe you want UMA_*, otherwise this won't get uploaded?  Or is w/o UMA_ the
new public pattern?

I'm also not sure histograms work this way.  Won't this end up recording a false
and true value if an error occurs?  Is it smart enough to only record one value?

[jar (doing other things), 2013/04/24 21:51:07]

+1 for needing the prefix UMA_HISTOGRAM_BOOLEAN

The code, as written, will work, producing counts in two distinct buckets.  I
always hate the chance of (for example) misspelling the histogram in one place
or the other, so I'd suggest factoring out:

void histogram_renderer_status(bool error) {
  UMA_HISTOGRAM_BOOLEAN("Media.AudioRendererSinkErrors", error);
}

You can then call this from line 288 and 635.


On 2013/04/24 20:56:39, DaleCurtis wrote:

[scherkus (not reviewing), 2013/04/25 00:04:46]

Done.

+
   sink_->Initialize(audio_parameters_, weak_this_);
   sink_->Start();
 
@@ -629,6 +632,7 @@ void AudioRendererImpl::UpdateEarliestEndTime_Locked(
 }
 
 void AudioRendererImpl::OnRenderError() {
+  HISTOGRAM_BOOLEAN("Media.AudioRendererSinkErrors", true);

[scherkus (not reviewing), 2013/04/24 20:27:41]

we can also record audio_parameters_ when hitting an error, but I'm inclined to
wait to see how often this gets hit (I long for way to record histograms for
parameters without having a histogram for each individual parameter!)

[jar (doing other things), 2013/04/24 21:51:07]

Some of this stuff *might* be pulled from dremel logs, looking for correlation
between failure reports, and parameters seen in other histograms.  I agree that
first you should see how significant the issue is. ;-)

On 2013/04/24 20:27:41, scherkus wrote:

   disabled_cb_.Run();
 }