Add responsiveness_metric for timeline_based_measurement

tools/telemetry/telemetry/web_perf/metrics/mainthread_jank.py

Index: tools/telemetry/telemetry/web_perf/metrics/mainthread_jank.py
diff --git a/tools/telemetry/telemetry/web_perf/metrics/mainthread_jank.py b/tools/telemetry/telemetry/web_perf/metrics/mainthread_jank.py
new file mode 100644
index 0000000000000000000000000000000000000000..4fc905dd6cc2cedd5958abb157053abeaef9618a
--- /dev/null
+++ b/tools/telemetry/telemetry/web_perf/metrics/mainthread_jank.py
@@ -0,0 +1,61 @@
+# Copyright 2014 The Chromium Authors. All rights reserved.

[chrishenry, 2014/05/13 03:22:38]

nit: Filename should probably be main_thread_jank to be consistent.

+# Use of this source code is governed by a BSD-style license that can be
+# found in the LICENSE file.
+
+from telemetry.web_perf.metrics import timeline_based_metric
+
+
+PERCEIVABLE_DELAY_MS = 50

[chrishenry, 2014/05/13 03:22:38]

Please doc this. Maybe name it as USER_PERCEIVABLE_DELAY_THRESHOLD_MS

[nednguyen, 2014/05/14 17:32:27]

Done.

+
+
+class _MainThreadJankStat(object):
+  def __init__(self):
+    self.sum_top_slices_thread_time = 0

[chrishenry, 2014/05/13 03:22:38]

Hm, slightly inconsistent. We seem to use top slices and top message loop for
the same thing in this file. Should this be top_message_loop?

[nednguyen, 2014/05/14 17:32:27]

Done.

+    # slices whose overlapped duration that is more than PERCEIVABLE_DELAY_MS
+    self.num_big_top_slices = 0
+    self.sum_big_top_slices_thread_time = 0
+    self.biggest_top_slice_thread_time = 0

[chrishenry, 2014/05/13 03:22:38]

Document each of this stats (if not here, somewhere else where it makes sense to
document).

+
+def _ComputeMainThreadJankStats(renderer_thread, thread_start, thread_end):
+  stat = _MainThreadJankStat()
+  for s in renderer_thread.IterAllSlicesOverlappedWithThreadTimeRange(

[chrishenry, 2014/05/13 03:22:38]

Can you explain again as to the difference between start/end and
thread_start/thread_end? I had thought that thread_start/thread_end is when the
event is actually being scheduled by the kernel (and start, end is the actual
wall time). If that's the case, shouldn't this be using the start/end instead of
thread_start/thread_end?

[nednguyen, 2014/05/14 17:32:27]

Since what happens outside of the browser is uncontrollable by us, so we want to
denoise the metrics by using thread_start, thread_end.

[chrishenry, 2014/05/19 23:16:50]

Ok, I agree.

+    thread_start, thread_end):

[chrishenry, 2014/05/13 03:22:38]

4-col indent.

[nednguyen, 2014/05/14 17:32:27]

Done. Just realize that my vim indent for python was messed up. :-)

+    if not s.IsTopMessageLoop():

[chrishenry, 2014/05/13 03:22:38]

Btw, maybe this should be a static function instead of a member function of
TimelineEvent.

[nednguyen, 2014/05/14 17:32:27]

Done.

+      continue
+    jank_thread_duration = s.GetOverlappedThreadTimeDurationWithRange(

[chrishenry, 2014/05/13 03:22:38]

Question on thread_time/thread_end applies here too.

[nednguyen, 2014/05/14 17:32:27]

Done.

+      thread_start, thread_end)
+    stat.sum_top_slices_thread_time += jank_thread_duration
+    stat.biggest_top_slice_thread_time = max(
+        stat.biggest_top_slice_thread_time, jank_thread_duration)
+    if jank_thread_duration > PERCEIVABLE_DELAY_MS:
+      stat.num_big_top_slices += 1
+      stat.sum_big_top_slices_thread_time += jank_thread_duration
+  return stat
+
+
+class MainthreadJankMetric(timeline_based_metric.TimelineBasedMetric):

[chrishenry, 2014/05/13 03:22:38]

This is inconsistent, above you use MainThread, and here Mainthread (hence also
my comment on file name). We should pick one and stick with it. (:

[nednguyen, 2014/05/14 17:32:27]

Picked Mainthread.

+  def __init__(self):
+    super(MainthreadJankMetric, self).__init__()
+
+  def AddResults(self, _, renderer_thread, interaction_records, results):
+    self.VerifyNonOverlappedRecords(interaction_records)

[chrishenry, 2014/05/13 03:22:38]

Why must the records be non-overlapping? Shouldn't we be outputting results on
per-timeline record basis instead of a combined results?

[nednguyen, 2014/05/14 17:32:27]

This is supposed to return a summarized results for multiple records. To get one
results for each record, one needs to call this multiple time (and
timeline_based_measurement does so). 

+    num_big_janks = 0
+    total_jank_thread_time = 0
+    total_big_jank_thread_time = 0
+    biggest_jank_thread_time = 0
+
+    for record in interaction_records:
+      record_jank_stat = _ComputeMainThreadJankStats(
+        renderer_thread, record.thread_start, record.thread_end)

[chrishenry, 2014/05/13 03:22:38]

4-col hanging indent, same everywhere below too.

[chrishenry, 2014/05/13 03:22:38]

renderer_thread isn't documented even in the base class, we should document what
it means.

[nednguyen, 2014/05/14 17:32:27]

Ok. Will escalate this to Nat, who knows about this better than me.

+      num_big_janks += record_jank_stat.num_big_top_slices
+      total_jank_thread_time += record_jank_stat.sum_top_slices_thread_time
+      total_big_jank_thread_time += (
+        record_jank_stat.sum_big_top_slices_thread_time)
+      biggest_jank_thread_time = (
+        max(biggest_jank_thread_time,
+            record_jank_stat.biggest_top_slice_thread_time))
+
+    results.Add('num_big_janks', 'score', num_big_janks)
+    results.Add('total_jank_thread_time', 'ms', total_jank_thread_time)
+    results.Add('total_big_jank_thread_time', 'ms', total_big_jank_thread_time)
+    results.Add('biggest_jank_thread_time', 'ms', biggest_jank_thread_time)