Add benchmark that imitates news reading.

tools/perf/benchmarks/news.py

Index: tools/perf/benchmarks/news.py
diff --git a/tools/perf/benchmarks/news.py b/tools/perf/benchmarks/news.py
new file mode 100644
index 0000000000000000000000000000000000000000..af06225eee0b4f8269ab32df2b2804887ab3505a
--- /dev/null
+++ b/tools/perf/benchmarks/news.py
@@ -0,0 +1,94 @@
+# Copyright 2016 The Chromium Authors. All rights reserved.
+# Use of this source code is governed by a BSD-style license that can be
+# found in the LICENSE file.
+
+import re
+
+from core import perf_benchmark
+from telemetry.timeline import chrome_trace_config
+from telemetry.timeline import chrome_trace_category_filter
+from telemetry.web_perf import timeline_based_measurement
+import page_sets
+
+
+# See tr.v.Numeric.getSummarizedScalarNumericsWithNames()
+# https://github.com/catapult-project/catapult/blob/master/tracing/tracing/value/numeric.html#L323
+_IGNORED_MEMORY_STATS_RE = re.compile(r'_(std|count|min|sum|pct_\d{4}(_\d+)?)$')
+_MEMORY_STATS_RE = re.compile(r'renderer_processes')

[petrcermak, 2016/07/07 14:55:55]

You can replace this by |'renderer_processes' in value.name|. No need for a
regex.

[ulan, 2016/07/07 17:34:38]

Done.

[petrcermak, 2016/07/08 09:17:53]

There's no point in having the regular expression anymore ;-)

[ulan, 2016/07/08 09:50:57]

Done.

+
+# Track only the high-level GC stats to reduce the data load on dashboard.
+_IGNORED_V8_STATS_RE = re.compile(
+    r'_(idle_deadline_overrun|percentage_idle|outside_idle)')

[petrcermak, 2016/07/07 14:55:56]

Should this only match the end of the name ("$" at the end of the regular
expression)?

[ulan, 2016/07/07 17:34:38]

We can have outside_idle_(max|sum|min...).

[petrcermak, 2016/07/08 09:17:53]

Acknowledged.

+_V8_GC_HIGH_LEVEL_STATS_RE = re.compile(
+    r'v8-gc-full-mark-compactor_|'

[petrcermak, 2016/07/07 14:55:55]

Let's use the power of regular expressions when you're using them ;-)

r'v8-gc-((full|latency|memory)-mark-compactor|incremental-(finalize|step)|scavenger|total)_'

I suppose the inner simplifications might make the regex too unreadable, so the
following might be better (your call):

r'v8-gc-(full-mark-compactor|latency-mark-compactor|memory-mark-compactor|incremental-finalize|incremental-step|scavenger|total)_'

Also, shouldn't this only match the beginning of the name ("^" at the beginning
of the regular expression)?

[ulan, 2016/07/07 17:34:38]

Thanks, I chose the second option as it seems more readable to me.
And added "^".

[petrcermak, 2016/07/08 09:17:53]

Acknowledged.

+    r'v8-gc-incremental-finalize_|'
+    r'v8-gc-incremental-step_|'
+    r'v8-gc-latency-mark-compactor_|'
+    r'v8-gc-memory-mark-compactor_|'
+    r'v8-gc-scavenger_|'
+    r'v8-gc-total_')
+
+
+class _NewsBenchmark(perf_benchmark.PerfBenchmark):
+  """ Base class for news browsing benchmarks.

[petrcermak, 2016/07/07 14:55:55]

nit: Remove space after quotes

[ulan, 2016/07/07 17:34:38]

Done.

+  This benchmark measures memory usage with periodic memory dumps and v8 times.
+  See page_sets.news_stories._NewsStory for workload description.
+  """
+
+  def CreateTimelineBasedMeasurementOptions(self):
+    categories = [
+      # Disable all categories by default.
+      '-*',
+      # Memory categories.
+      'disabled-by-default-memory-infra',
+      # V8 categories.
+      'blink.console',
+      'disabled-by-default-v8.gc',
+      'renderer.scheduler',
+      'v8',
+      'webkit.console',
+    ]
+    options = timeline_based_measurement.Options(
+        chrome_trace_category_filter.ChromeTraceCategoryFilter(
+            ','.join(categories)))
+    options.config.enable_android_graphics_memtrack = True
+    # Trigger periodic light memory dumps every 1000 ms.
+    memory_dump_config = chrome_trace_config.MemoryDumpConfig()
+    memory_dump_config.AddTrigger('light', 1000)
+    options.config.chrome_trace_config.SetMemoryDumpConfig(memory_dump_config)
+    options.SetTimelineBasedMetric('v8AndMemoryMetrics')
+    return options
+
+  @classmethod
+  def Name(cls):
+    return 'news_%s' % cls.page_set.PLATFORM
+
+  @classmethod
+  def ValueCanBeAddedPredicate(cls, value, is_first_result):
+    # TODO(crbug.com/610962): Remove this stopgap when the perf dashboard
+    # is able to cope with the data load generated by TBMv2 metrics.
+    if 'memory:chrome' in value.name:
+      return (_MEMORY_STATS_RE.search(value.name) and
+              not _IGNORED_MEMORY_STATS_RE.search(value.name))

[petrcermak, 2016/07/07 14:55:55]

You have this in both statements. I suggest you factor it out:

if _IGNORED_MEMORY_STATS_RE.search(value.name):
  return False
if 'memory:chrome' in value.name:
  return bool(_MEMORY_STATS_RE.search(value.name))
else:
  return bool(_V8_GC_HIGH_LEVEL_STATS_RE.search(value.name))

[ulan, 2016/07/07 17:34:38]

Note that the branch below has ignored V8 stats, not memory stats.

[petrcermak, 2016/07/08 09:17:53]

Ack. Sorry, I didn't realize that.

+    return (_V8_GC_HIGH_LEVEL_STATS_RE.search(value.name) and
+            not _IGNORED_V8_STATS_RE.search(value.name))
+
+  @classmethod
+  def ShouldTearDownStateAfterEachStoryRun(cls):
+    return True
+
+
+class DesktopNewsBenchmark(_NewsBenchmark):
+  page_set = page_sets.DesktopNewsStorySet
+
+  @classmethod
+  def ShouldDisable(cls, possible_browser):
+    return possible_browser.platform.GetDeviceTypeName() != 'Desktop'
+
+
+class MobileNewsBenchmark(_NewsBenchmark):
+  page_set = page_sets.MobileNewsStorySet
+
+  @classmethod
+  def ShouldDisable(cls, possible_browser):
+    return possible_browser.platform.GetDeviceTypeName() == 'Desktop'