telemetry: Add a page set for blink's memory usage measurement

tools/perf/benchmarks/renderer_memory.py

Index: tools/perf/benchmarks/renderer_memory.py
diff --git a/tools/perf/benchmarks/memory_health_plan.py b/tools/perf/benchmarks/renderer_memory.py
similarity index 52%
copy from tools/perf/benchmarks/memory_health_plan.py
copy to tools/perf/benchmarks/renderer_memory.py
index 14be34cbf3f10ca62aec10f793ad8f95beee3e28..55490c77830892ebb29d250f87f7502ebbc61ce3 100644
--- a/tools/perf/benchmarks/memory_health_plan.py
+++ b/tools/perf/benchmarks/renderer_memory.py
@@ -5,39 +5,35 @@
 import re
 
 from core import perf_benchmark
-
-from telemetry import benchmark
 from telemetry.timeline import tracing_category_filter
 from telemetry.web_perf import timeline_based_measurement
 
 import page_sets
 
 
-RE_BENCHMARK_VALUES = re.compile('(fore|back)ground-memory_')
+_RENDERER_MEMORY_VALUE_REGEXP = re.compile('.+-memory_.+_renderer')
 
 
-@benchmark.Enabled('android')
-class MemoryHealthPlan(perf_benchmark.PerfBenchmark):
-  """Timeline based benchmark for the Memory Health Plan."""
+class RendererMemoryBlinkMemoryMobile(perf_benchmark.PerfBenchmark):
+  """Timeline based benchmark for measuring memory consumption on mobile
+  sites on which blink's memory consumption is relatively high."""
 
-  page_set = page_sets.MemoryHealthStory
+  page_set = page_sets.BlinkMemoryMobilePageSet
 
   def SetExtraBrowserOptions(self, options):
-    # TODO(perezju): Temporary workaround to disable periodic memory dumps.
-    # See: http://crbug.com/513692
+    # TODO(bashi): Remove this workaround, see http://crbug.com/513692
     options.AppendExtraBrowserArgs('--enable-memory-benchmarking')

[perezju, 2015/08/28 12:54:32]

I worry a bit about duplicating these temporary workarounds on all memory
benchmarks we might create in the future. Not sure what Ned thinks about this,
but how about creating a perf_benchmark.MemoryBenchmark class where these tweaks
can live, and then subclass from it on both memory_health_plan.py and here?

[bashi, 2015/08/31 23:47:28]

Sounds a good idea. Done. Ned, what do you think?

[nednguyen, 2015/09/01 17:52:38]

caveat: the naming should be perf_benchmark._MemoryBenchmark so run_benchmark
doesn't discover it.
SGTM. We do this in many places in perf:
https://code.google.com/p/chromium/codesearch#chromium/src/tools/perf/benchma...

[bashi, 2015/09/01 22:37:00]

Done. But I got following lint warning.

"Access to a protected member _MemoryBenchmark of a client class
(protected-access)"

Just ignore the warning?

 
   def CreateTimelineBasedMeasurementOptions(self):
-    # Enable only memory-infra, to get memory dumps, and blink.console, to get
-    # the timeline markers used for mapping threads to tabs.
+    # See the comment in MemoryHealthPlan.CreateTimelineBasedMeasurementOptions
     trace_memory = tracing_category_filter.TracingCategoryFilter(
-      filter_string='-*,blink.console,disabled-by-default-memory-infra')
+        filter_string='-*,blink.console,disabled-by-default-memory-infra')
     return timeline_based_measurement.Options(overhead_level=trace_memory)

[perezju, 2015/08/28 12:54:32]

Same as above, maybe move to a perf_benchmark.MemroyBenchmark parent class.

[bashi, 2015/08/31 23:47:28]

Done.

 
   @classmethod
   def Name(cls):
-    return 'memory.memory_health_plan'
+    return 'renderer_memory.blink_memory_mobile'
 
   @classmethod
   def ValueCanBeAddedPredicate(cls, value, is_first_result):
-    return bool(RE_BENCHMARK_VALUES.match(value.name))
+    return bool(_RENDERER_MEMORY_VALUE_REGEXP.match(value.name))