Generalize abort tracking to page end state tracking

tools/metrics/histograms/histograms.xml

Index: tools/metrics/histograms/histograms.xml
diff --git a/tools/metrics/histograms/histograms.xml b/tools/metrics/histograms/histograms.xml
index f7fd8937c039f19f0327adf824671dab93cf2849..6735460f1689c8fd9858eaea7fceaa9ffc2b7986 100644
--- a/tools/metrics/histograms/histograms.xml
+++ b/tools/metrics/histograms/histograms.xml
@@ -44074,6 +44074,9 @@ http://cs/file:chrome/histograms.xml - but prefer this file for new entries.
 
 <histogram base="true" name="PageLoad.Experimental.AbortTiming.ClientRedirect"
     units="ms">
+  <obsolete>
+    Deprecated in favor of PageLoad.Internal.ClientRedirect.*.
+  </obsolete>
   <owner>csharrison@chromium.org</owner>
   <summary>
     This metric is still experimental and not yet ready to be relied upon.
@@ -44215,6 +44218,24 @@ http://cs/file:chrome/histograms.xml - but prefer this file for new entries.
   </summary>
 </histogram>
 
+<histogram name="PageLoad.Experimental.PageTiming.FirstPaintToPageEnd"
+    units="ms">
+  <owner>bmcquade@chromium.org</owner>
+  <summary>
+    Measures the total time the page load was active after first paint, for page
+    loads that spend the entire time in the foreground.
+  </summary>
+</histogram>
+
+<histogram name="PageLoad.Experimental.PageTiming.NavigationToPageEnd"
+    units="ms">
+  <owner>bmcquade@chromium.org</owner>
+  <summary>
+    Measures the total time the page load was active, for page loads that spend
+    the entire time in the foreground.
+  </summary>
+</histogram>
+
 <histogram
     name="PageLoad.Experimental.PaintTiming.FirstMeaningfulPaintSignalStatus"
     enum="FirstMeaningfulPaintSignalStatus">
@@ -44452,6 +44473,43 @@ http://cs/file:chrome/histograms.xml - but prefer this file for new entries.
   </summary>
 </histogram>
 
+<histogram name="PageLoad.PageTiming.FirstPaintToFirstBackground" units="ms">
+  <owner>bmcquade@chromium.org</owner>
+  <summary>
+    Measures the total time the page load was active after first paint, up until
+    being backgrounded, for page loads that started in the foreground.

[Ilya Sherman, 2017/02/21 23:08:29]

Is this only recorded for pages that were ever backgrounded, or is some special
value recorded for pages that stayed in the foreground indefinitely?  What's
recorded if I open a page and don't close it for a month, during which time my
computer goes to sleep multiple times (i.e. base::TimeTicks pauses)?  (If the
timing is limited to just the page load, then I guess that question is probably
irrelevant, but it's a bit unclear -- maybe because the context of "what page
loads is this metric recorded for?" is not entirely clear.)

[Bryan McQuade, 2017/02/22 00:57:33]

This is only recorded for pages that were ever backgrounded.

This, along with the two other new histograms in this change
'FirstPaintToPageEnd' and 'FirstPaintToPageEnd.NoEndTime' help us to understand
how often these 3 cases of page termination are encountered for page loads
processed by the page load metrics infrastructure.

This is recorded for the entire duration from when a page paints to when it
first goes into the background. This histogram will have a very large tail,
which is expected / intended. Happy to try to clarify this in the name or
description - just make suggestions for the text & I can try to incorporate it.

RE: base::TimeTicks pauses, these are general issues that affect all histograms
that use TimeTicks (we have many), so I'm disinclined to mention this in this
histogram. In this case, the pause during suspend behavior is desirable - this
metric is a proxy for the time that a user spends on a page, so time spent
suspended shouldn't be counted.

Given that the other 2 histograms are in PageLoad.Experimental, I've moved this
one to PageLoad.Experimental as well. The goal here for the time being is to get
some data to understand how often these 3 cases are encountered (as well as what
the histogram distribution looks like), as a first step towards building a
better non-experimental histogram that can give insight into the time users
spend on web pages.

+  </summary>
+</histogram>
+
+<histogram name="PageLoad.PageTiming.NavigationToFailedProvisionalLoad"
+    units="ms">
+  <owner>bmcquade@chromium.org</owner>
+  <owner>shivanisha@chromium.org</owner>
+  <summary>
+    Measures the time from navigation timing's navigation start to the time the
+    provisional load failed. Only measures provisional loads that failed in the
+    foreground.
+  </summary>
+</histogram>
+
+<histogram name="PageLoad.PageTiming.NavigationToFirstBackground" units="ms">
+  <owner>bmcquade@chromium.org</owner>
+  <summary>
+    Measures the total time the page load was active, up until being
+    backgrounded, for page loads that started in the foreground.
+  </summary>
+</histogram>
+
+<histogram name="PageLoad.PageTiming.NavigationToFirstForeground" units="ms">
+  <owner>bmcquade@chromium.org</owner>
+  <owner>csharrison@chromium.org</owner>
+  <summary>
+    Measures the time from navigation timing's navigation start to the time the
+    user first foregrounds an initially backgrounded tab. Only measures
+    navigations that started in the background.

[Ilya Sherman, 2017/02/21 23:08:29]

What is recorded if the user never foregrounds the tab?  What if the user waits
a month to foreground it?

[Bryan McQuade, 2017/02/22 00:57:33]

This histogram will have a long tail, which is fine. This histogram only logs
metrics for pages that start in the background and are eventually foregrounded.
This is just a rename of the existing
PageLoad.Timing2.NavigationToFirstForeground - we've wanted to get rid of
Timing2 for a while but didn't have a good category name to put this one in -
the 'PageTiming' category is a better name & home for this one.

+  </summary>
+</histogram>
+
 <histogram name="PageLoad.PaintTiming.ForegroundToFirstPaint" units="ms">
   <owner>pkotwicz@chromium.org</owner>
   <summary>
@@ -44667,6 +44725,9 @@ http://cs/file:chrome/histograms.xml - but prefer this file for new entries.
 </histogram>
 
 <histogram name="PageLoad.Timing2.NavigationToFailedProvisionalLoad" units="ms">
+  <obsolete>
+    deprecated in favor of PageLoad.PageTiming.NavigationToFailedProvisionalLoad
+  </obsolete>
   <owner>bmcquade@chromium.org</owner>
   <owner>shivanisha@chromium.org</owner>
   <summary>
@@ -44702,6 +44763,9 @@ http://cs/file:chrome/histograms.xml - but prefer this file for new entries.
 </histogram>
 
 <histogram name="PageLoad.Timing2.NavigationToFirstForeground" units="ms">
+  <obsolete>
+    deprecated in favor of PageLoad.PageTiming.NavigationToFirstForeground
+  </obsolete>
   <owner>bmcquade@chromium.org</owner>
   <owner>csharrison@chromium.org</owner>
   <summary>
@@ -95397,7 +95461,7 @@ http://cs/file:chrome/histograms.xml - but prefer this file for new entries.
   <int value="2" label="IPC received from a frame we navigated away from"/>
   <int value="3" label="IPC received from a bad URL scheme"/>
   <int value="4" label="No IPCs received for this navigation"/>
-  <int value="5" label="Abort reported before navigation start"/>
+  <int value="5" label="Page end reported before navigation start"/>

[Ilya Sherman, 2017/02/21 23:08:29]

Hmm, why did this label change?  Is it a semantic change, or just a
clarification of what the semantics were all along?

[Bryan McQuade, 2017/02/22 00:57:33]

There is no behavior change, but the backing implementation details changed to
use the name 'page end' in some places where we used the name 'abort'
previously, so this keeps the histogram name in sync with the code.

   <int value="6"
       label="Multiple aborted provisional loads at navigation start"/>
   <int value="7"
@@ -95405,6 +95469,7 @@ http://cs/file:chrome/histograms.xml - but prefer this file for new entries.
              (deprecated)"/>
   <int value="8" label="Inter process TimeTicks skew"/>
   <int value="9" label="No commit or failed provisional load received"/>
+  <int value="10" label="No page load end time recorded"/>
 </enum>
 
 <enum name="InterruptReason" type="int">
@@ -117721,6 +117786,14 @@ value.
   <affected-histogram name="PageLoad.ParseTiming.NavigationToParseStart"/>
 </histogram_suffixes>
 
+<histogram_suffixes name="PageLoadMetricsNoEndTime" separator=".">
+  <suffix name="NoEndTime" label="The page load had no recorded end time."/>

[Ilya Sherman, 2017/02/21 23:08:29]

Hmm, this sounds like the metric is recording (\inf - t), where t is the first
paint or navigation time.  Is the value always recorded into the overflow
bucket?  I'm not really understanding what sort of data I would expect to see
for histograms with this suffix, nor what it would mean...

[Bryan McQuade, 2017/02/22 00:57:33]

Yeah, this is a bad description, thanks for catching it! I've updated it to be a
little clearer. We want to understand how frequently this case is happening and
whether its histogram shape differs from cases where an end time was explicitly
recorded. This is an experimental histogram (PageLoad.Experimental) and I expect
to fold it into a simpler non-experimental histogram once we gather data and
learn a bit more about what we see in the field.

+  <affected-histogram
+      name="PageLoad.Experimental.PageTiming.FirstPaintToPageEnd"/>
+  <affected-histogram
+      name="PageLoad.Experimental.PageTiming.NavigationToPageEnd"/>
+</histogram_suffixes>
+
 <histogram_suffixes name="PageLoadMetricsUserGesture" separator=".">
   <suffix name="UserGesture"
       label="Restricted to pages loaded via a user gesture."/>