Port part of media.tough_video_cases to TBMv2

tools/perf/benchmarks/media.py

Index: tools/perf/benchmarks/media.py
diff --git a/tools/perf/benchmarks/media.py b/tools/perf/benchmarks/media.py
index c5c6ba613934b033b5f76770000c5d16809565bf..63538151bf06e81a9709cef63fd07696eb0867b9 100644
--- a/tools/perf/benchmarks/media.py
+++ b/tools/perf/benchmarks/media.py
@@ -6,8 +6,10 @@ from core import perf_benchmark
 
 from telemetry import benchmark
 from telemetry.page import legacy_page_test
+from telemetry.timeline import chrome_trace_category_filter
 from telemetry.value import list_of_scalar_values
 from telemetry.value import scalar
+from telemetry.web_perf import timeline_based_measurement
 
 from measurements import media
 import page_sets
@@ -40,7 +42,7 @@ class _MSEMeasurement(legacy_page_test.LegacyPageTest):
 # android: See media.android.tough_video_cases below
 # crbug.com/565180: Only include cases that report time_to_play
 @benchmark.Disabled('android')
-class Media(perf_benchmark.PerfBenchmark):
+class MediaToughVideoCases(perf_benchmark.PerfBenchmark):
   """Obtains media metrics for key user scenarios."""
   test = media.Media
   page_set = page_sets.ToughVideoCasesPageSet
@@ -50,6 +52,27 @@ class Media(perf_benchmark.PerfBenchmark):
     return 'media.tough_video_cases'
 
 
+@benchmark.Disabled('android')
+class MediaToughVideoCasesTBMv2(perf_benchmark.PerfBenchmark):

[watk, 2017/03/14 00:08:45]

Not a comment for this CL, just a naming bikeshed :) for when we eventually
rename this: I'd prefer if we called this either just Media, ToughMediaCases
because it isn't only for video. There are audio stories too.

[CalebRouleau, 2017/03/14 00:12:46]

John and I already discussed this. Our current idea is that after we remove
legacy media.tough_video_cases, we will rename TBMv2 tough_media_cases to
media.src_cases. This is because it will be parallel with media.mse_cases and
(not yet written) media.eme_cases. Then that breaks down media use cases into
src, eme, and mse, which I think is most close to how the actual code works.
WDYT?

[charliea (OOO until 10-5), 2017/03/14 19:04:20]

nednguyen@google.com: if I understand your long-term naming suggestions, you
want one benchmark per team with the user stories separated by tags, correct? so
there's just two benchmarks, media.desktop and media.mobile, and there's
multiple tag sets within those benchmarks ("mse" might be one tag set and "src"
might be another). Is this right?

I'm not sure how much we want to publicize this long-term plan now given that so
little of it is implemented. Regardless, I don't think it affects this CL.

+  """Obtains media metrics using TBMv2. Will eventually replace Media class."""
+  page_set = page_sets.ToughVideoCasesPageSet
+
+  def CreateTimelineBasedMeasurementOptions(self):
+    category_filter = chrome_trace_category_filter.ChromeTraceCategoryFilter(
+        filter_string='toplevel')
+    options = timeline_based_measurement.Options(category_filter)
+    options.config.chrome_trace_config.category_filter.AddFilterString('rail')

[watk, 2017/03/14 00:08:45]

nit: it's weird that you add "toplevel" above and then "rail" here with a gap.
Can you group all the category filters into one block so it's easy to see? Take
a look at the v8 link below for an example.

Also I think it's really useful to have a comment indicating why the category is
enabled. It can be quite difficult to find out why a category is enabled from
looking at the metrics. e.g., I just had a look at powerMetric.html and it
doesn't say which categories it it needs. 

So if we later change the metrics it's difficult to know which categories we
should change.

I like the way it's done for some v8 ones:
https://cs.chromium.org/chromium/src/tools/perf/benchmarks/v8.py?l=262&rcl=95...

+    options.config.enable_atrace_trace = True
+    options.config.atrace_config.categories = ['sched']
+    options.config.enable_battor_trace = True
+    options.SetTimelineBasedMetrics(['powerMetric', 'cpuTimeMetric'])
+    return options
+
+  @classmethod
+  def Name(cls):
+    return 'media.tough_video_cases_tbmv2'
+
+
 # crbug.com/565180: Only include cases that don't report time_to_play
 @benchmark.Disabled('android')
 class MediaExtra(perf_benchmark.PerfBenchmark):