Log UMA for navigation start timestamp skew for browser-initiated loads

content/renderer/render_frame_impl.cc

Index: content/renderer/render_frame_impl.cc
diff --git a/content/renderer/render_frame_impl.cc b/content/renderer/render_frame_impl.cc
index 6f08693d2fbfb13d5200a140879e5bc08bda2b4b..8e60f8f04a849469a6c885cb03f945715b36569a 100644
--- a/content/renderer/render_frame_impl.cc
+++ b/content/renderer/render_frame_impl.cc
@@ -475,9 +475,19 @@ base::TimeTicks SanitizeNavigationTiming(
   DCHECK(!browser_navigation_start.is_null());
   base::TimeTicks navigation_start =
       std::min(browser_navigation_start, renderer_navigation_start);
-  // TODO(csharrison) Investigate how big a problem the cross process
-  // monotonicity really is and on what platforms. Log UMA for:
-  // |renderer_navigation_start - browser_navigation_start|
+  base::TimeDelta difference =
+      renderer_navigation_start - browser_navigation_start;
+  if (difference > base::TimeDelta()) {
+    UMA_HISTOGRAM_CUSTOM_TIMES(
+        "Navigation.Start.RendererBrowserDifference.Positive", difference,
+        base::TimeDelta::FromMilliseconds(1), base::TimeDelta::FromSeconds(5),

[Alexei Svitkine (slow), 2015/11/06 16:53:18]

Suggest using UMA_HISTOGRAM_TIMES().

It gives you a range from 1ms to 10s, which is very similar to your params here.
Given that the distribution is exponential, you'll still get lots of resolution
at the lower buckets.

+        50);
+  } else {
+    UMA_HISTOGRAM_CUSTOM_TIMES(
+        "Navigation.Start.RendererBrowserDifference.Negative", -difference,
+        base::TimeDelta::FromMilliseconds(1), base::TimeDelta::FromSeconds(5),
+        50);
+  }
   return navigation_start;
 }