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 a08a797caadefaa69ebfc2739e4bfd0fda291e82..66c3eb5b2fb38f45b0d2def4a8f917ec5d1f957d 100644
--- a/tools/metrics/histograms/histograms.xml
+++ b/tools/metrics/histograms/histograms.xml
@@ -75250,6 +75250,26 @@ http://cs/file:chrome/histograms.xml - but prefer this file for new entries.
   </summary>
 </histogram>
 
+<histogram name="V8.ArrayBufferBigAllocations" units="log2(Size)">

[Ilya Sherman, 2017/03/31 22:00:36]

Why do you record log2 of the size?  Histogram buckets are already exponentially
spaced.  Also, the size in what units?

[kschimpf, 2017/03/31 22:26:32]

I switched to log2 size (in bytes) because the representation used by V8 to
collect histograms uses type "int" while the size of the request is size_t. This
disallows numbers greater than 2**31 which can possibly happen with WASM
modules.

Since the code in V8 for counters is used in many places with an "int"
assumption, I did not want to change things so close to the 59 release. On the
other hand, we would like some feedback on these concepts as soon as possible.

Updating the units to log2(bytes).

+  <owner>bradnelson@chromium.org</owner>
+  <owner>titzer@chromium.org</owner>
+  <owner>kschimpf@chromium.org</owner>
+  <summary>
+    Number of bytes requested in array buffer allocation. Only tracks memory
+    allocations over 1Mb.

[Ilya Sherman, 2017/03/31 22:00:36]

Why does the histogram only track allocations over 1MB?  It seems simple enough
to toss the rest into an underflow bucket.

[kschimpf, 2017/03/31 22:26:32]

This is definitely a legitimate answer. I probably worried too much about the
overhead   of collecting this data. I did not want to slow down buffer array
allocations. I also did not want to flood the histogram with a bucket that is so
large that visually, it would not be easy to see problem sized allocations
(which are large ones).

That said, if you still feel that all allocations should be tracked, I will
change the code.

+  </summary>
+</histogram>
+
+<histogram name="V8.ArrayBufferNewSizeFailures" units="log2(Size)">
+  <owner>bradnelson@chromium.org</owner>
+  <owner>titzer@chromium.org</owner>
+  <owner>kschimpf@chromium.org</owner>
+  <summary>
+    Number of bytes requested in array buffer allocation that V8 was not able to
+    meet the corresponding request.

[Ilya Sherman, 2017/03/31 22:00:36]

This sentence is a bit confusing.  Could you please try to reword it a bit to be
clearer?

+  </summary>
+</histogram>
+
 <histogram name="V8.AsmWasmTranslationMicroSeconds" units="microseconds">
   <owner>bradnelson@chromium.org</owner>
   <owner>titzer@chromium.org</owner>