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 b736f7de677dd451846ee0a2d042034a871a62f9..b1e842a52bec8cd60fc8dbf20acd99dddbc7ee85 100644
--- a/tools/metrics/histograms/histograms.xml
+++ b/tools/metrics/histograms/histograms.xml
@@ -7962,9 +7962,12 @@ Therefore, the affected-histogram name has to have at least one dot in it.
   </summary>
 </histogram>
 
-<histogram name="Extensions.BackgroundPageLoadTime" units="milliseconds">

[Ilya Sherman, 2015/02/06 21:26:44]

Please mark the old histogram as <obsolete>, rather than deleting it.

[not at google - send to devlin, 2015/02/06 22:51:07]

Done.

+<histogram name="Extensions.BackgroundPageLoadTime2" units="milliseconds">
   <owner>kalman@chromium.org</owner>
-  <summary>The time for an extension's background page to load.</summary>
+  <summary>
+    The time taken for an extension's persistent background page to load its
+    initial URL.
+  </summary>
 </histogram>
 
 <histogram name="Extensions.BackgroundPageType"
@@ -8069,11 +8072,6 @@ Therefore, the affected-histogram name has to have at least one dot in it.
   </summary>
 </histogram>
 
-<histogram name="Extensions.DialogLoadTime" units="milliseconds">
-  <owner>Please list the metric's owners. Add more owner tags as needed.</owner>
-  <summary>The time for a dialog-hosted extension to load.</summary>
-</histogram>
-
 <histogram name="Extensions.DidCreateScriptContext_Blessed"
     units="milliseconds">
   <owner>kalman@chromium.org</owner>
@@ -8196,9 +8194,12 @@ Therefore, the affected-histogram name has to have at least one dot in it.
   </summary>
 </histogram>
 
-<histogram name="Extensions.EventPageActiveTime" units="milliseconds">
+<histogram name="Extensions.EventPageActiveTime2" units="milliseconds">
   <owner>kalman@chromium.org</owner>
-  <summary>The time an extension's event page has spent loaded.</summary>
+  <summary>
+    The time between an extension's event page loading its first URL and the
+    event page later shutting down.
+  </summary>
 </histogram>
 
 <histogram name="Extensions.EventPageIdleTime" units="milliseconds">
@@ -8206,9 +8207,11 @@ Therefore, the affected-histogram name has to have at least one dot in it.
   <summary>The time an extension's event page has spent unloaded.</summary>
 </histogram>
 
-<histogram name="Extensions.EventPageLoadTime" units="milliseconds">
+<histogram name="Extensions.EventPageLoadTime2" units="milliseconds">
   <owner>kalman@chromium.org</owner>
-  <summary>The time for an extension's event page to load.</summary>
+  <summary>
+    The time taken for an extension's event page to load its initial URL.
+  </summary>
 </histogram>
 
 <histogram name="Extensions.ExtensionCacheCount">
@@ -9008,6 +9011,15 @@ Therefore, the affected-histogram name has to have at least one dot in it.
   </summary>
 </histogram>
 
+<histogram name="Extensions.PopupLoadTime2" units="milliseconds">

[Ilya Sherman, 2015/02/06 21:26:44]

I don't see a PopupLoadTime1.  Does this just have a "2" to be consistent with
the other histograms that you're renaming?  It's probably not viable for all
extensions time metrics to have a "2" in them going forward, so it might not
really make sense to include a "2" here.  Up to you, though.

[not at google - send to devlin, 2015/02/06 22:51:07]

Yes that's mostly why I did the "2" thing, to be consistent. However - if there
*had* been an Extensions.PopupLoadTime - which there was supposed to be, and we
may already have data uploaded about it (?) - I'd have needed to change it.

[Ilya Sherman, 2015/02/06 23:19:49]

Hmm, if you think that there might already be data being uploaded for a metric
named "Extensions.PopupLoadTime", then please add that metric to histograms.xml
and mark it as <obsolete>.  If you're not sure, it doesn't hurt to add it, and
it's useful to be able to look at real metric names rather than just numeric
hashes when viewing the data server-side, including for pipeline debugging :)

[not at google - send to devlin, 2015/02/06 23:32:13]

Done.

+  <owner>kalman@chromium.org</owner>
+  <summary>
+    The time taken for an extension popup to load its initial URL. This may not
+    include the time taken for an extension process to start, if the extension
+    already has a background page running.

[Ilya Sherman, 2015/02/06 21:26:44]

Would it be worth splitting this metric into two cases, one for extensions that
were already running, and one for extensions that were not?  Bimodal histograms
tend to be hard to work with.

[not at google - send to devlin, 2015/02/06 22:51:07]

I did have a go at that actually but it wasn't clear that it'd be useful.

Ultimately it's time to render that we care about, how many resources are
consumed in opening a popup. The process startup thing is potentially a
side-effect of needing that which is hard to reason about. One useful bit of
data it'd give would be to compare cost of starting up a process vs not I
suppose, and the difference in whether there's a background page or not.
However, given these aren't actually real processes anyway - eh. Pretty subtle.

One way to split would be on whether there is any background page. That's
straightforward. It's not a metric I care about particularly, the only purpose
it would serve is some sort of normalization, potentially.

The other way would be whether the load caused a process to start vs the view
being attached to an existing process. What I'm unsure about here is races, what
the lifetime of a RenderProcessHost is vs a RenderViewHost. E.g. does Chrome
claim that it's made a RPH even though it hasn't quite finish, and happily
attaches a RVH to it anyway?

Alright - I think I'll leave it the way it is. If the data turns out to be
bimodal then I can make this change and call it something different (like
PopupLoadTimeWarm vs PopupLoadTimeCold).

[Ilya Sherman, 2015/02/06 23:19:49]

Fair enough.

+  </summary>
+</histogram>
+
 <histogram name="Extensions.ProcessManagerStartupHostsTime"
     units="milliseconds">
   <owner>yoz@chromium.org</owner>